Percona toolkit 2_1_operations_manual

Percona Toolkit Documentation
Release 2.1.1

Percona Inc

April 04, 2012

CONTENTS

1 Getting Percona Toolkit 3
1.1 Installation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2 Tools 5
2.1 pt-align . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 pt-archiver . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3 pt-config-diff . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.4 pt-deadlock-logger . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.5 pt-diskstats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.6 pt-duplicate-key-checker . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
2.7 pt-fifo-split . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
2.8 pt-find . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.9 pt-fingerprint . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
2.10 pt-fk-error-logger . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
2.11 pt-heartbeat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
2.12 pt-index-usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
2.13 pt-ioprofile . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
2.14 pt-kill . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
2.15 pt-log-player . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
2.16 pt-mext . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
2.17 pt-mysql-summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
2.18 pt-online-schema-change . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
2.19 pt-pmp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
2.20 pt-query-advisor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
2.21 pt-query-digest . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
2.22 pt-show-grants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
2.23 pt-sift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
2.24 pt-slave-delay . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
2.25 pt-slave-find . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
2.26 pt-slave-restart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
2.27 pt-stalk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
2.28 pt-summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
2.29 pt-table-checksum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
2.30 pt-table-sync . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
2.31 pt-table-usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
2.32 pt-tcp-model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
2.33 pt-trend . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258
2.34 pt-upgrade . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
2.35 pt-variable-advisor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270

i

2.36 pt-visual-explain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281

3 Conﬁguration 293
3.1 CONFIGURATION FILES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293
3.2 DSN (DATA SOURCE NAME) SPECIFICATIONS . . . . . . . . . . . . . . . . . . . . . . . . . . 294
3.3 ENVIRONMENT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296
3.4 SYSTEM REQUIREMENTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296

4 Miscellaneous 299
4.1 BUGS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
4.2 AUTHORS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
4.3 COPYRIGHT, LICENSE, AND WARRANTY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
4.4 VERSION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
4.5 Release Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300

Index 305

ii

Percona Toolkit Documentation, Release 2.1.1

Percona Toolkit is a collection of advanced command-line tools used by Percona (http://www.percona.com/) support
staff to perform a variety of MySQL and system tasks that are too difﬁcult or complex to perform manually.
These tools are ideal alternatives to private or “one-off” scripts because they are professionally developed, formally
tested, and fully documented. They are also fully self-contained, so installation is quick and easy and no libraries are
installed.
Percona Toolkit is derived from Maatkit and Aspersa, two of the best-known toolkits for MySQL server administration.
It is developed and supported by Percona Inc. For more information and other free, open-source software developed
by Percona, visit http://www.percona.com/software/.

CONTENTS 1


2 CONTENTS

CHAPTER

ONE

GETTING PERCONA TOOLKIT

1.1 Installation

Visit http://www.percona.com/software/percona-toolkit/ to download the latest release of Percona Toolkit. Or, get the
latest release from the command line:
wget percona.com/get/percona-toolkit.tar.gz

wget percona.com/get/percona-toolkit.rpm

wget percona.com/get/percona-toolkit.deb

You can also get individual tools from the latest release:
wget percona.com/get/TOOL

Replace TOOL with the name of any tool.

3


4 Chapter 1. Getting Percona Toolkit

CHAPTER

TWO

TOOLS

2.1 pt-align

2.1.1 NAME

pt-align - Align output from other tools to columns.

2.1.2 SYNOPSIS

Usage

pt-align [FILES]

pt-align aligns output from other tools to columns. If no FILES are speciﬁed, STDIN is read.
If a tool prints the following output,
DATABASE TABLE ROWS
foo bar 100
long_db_name table 1
another long_name 500

then pt-align reprints the output as,
DATABASE TABLE ROWS
foo bar 100
long_db_name table 1
another long_name 500

2.1.3 RISKS

The following section is included to inform users about the potential risks, whether known or unknown, of using this
tool. The two main categories of risks are those created by the nature of the tool (e.g. read-only tools vs. read-write
tools) and those created by bugs.
pt-align is a read-only tool. It should be very low-risk.
At the time of this release, we know of no bugs that could cause serious harm to users.
The authoritative source for updated information is always the online issue tracking system. Issues that affect this tool
will be marked as such. You can see a list of such issues at the following URL: http://www.percona.com/bugs/pt-align.

5


See also “BUGS” for more information on filing bugs and getting help.

2.1.4 DESCRIPTION

pt-align reads lines and splits them into words. It counts how many words each line has, and if there is one number that
predominates, it assumes this is the number of words in each line. Then it discards all lines that don’t have that many
words, and looks at the 2nd line that does. It assumes this is the first non-header line. Based on whether each word
looks numeric or not, it decides on column alignment. Finally, it goes through and decides how wide each column
should be, and then prints them out.
This is useful for things like aligning the output of vmstat or iostat so it is easier to read.

2.1.5 OPTIONS

This tool does not have any command-line options.

2.1.6 ENVIRONMENT

This tool does not use any environment variables.

2.1.7 SYSTEM REQUIREMENTS

You need Perl, and some core packages that ought to be installed in any reasonably new version of Perl.

2.1.8 BUGS

For a list of known bugs, see http://www.percona.com/bugs/pt-align.
Please report bugs at https://bugs.launchpad.net/percona-toolkit. Include the following information in your bug report:
• Complete command-line used to run the tool
• Tool --version
• MySQL version of all servers involved
• Output from the tool including STDERR
• Input files (log/dump/config files, etc.)
If possible, include debugging output by running the tool with PTDEBUG; see “ENVIRONMENT”.

2.1.9 DOWNLOADING




6 Chapter 2. Tools




2.1.10 AUTHORS

Baron Schwartz, Brian Fraser, and Daniel Nichter

2.1.11 ABOUT PERCONA TOOLKIT

This tool is part of Percona Toolkit, a collection of advanced command-line tools developed by Percona for MySQL
support and consulting. Percona Toolkit was forked from two projects in June, 2011: Maatkit and Aspersa. Those
projects were created by Baron Schwartz and developed primarily by him and Daniel Nichter, both of whom are
employed by Percona. Visit http://www.percona.com/software/ for more software developed by Percona.

2.1.12 COPYRIGHT, LICENSE, AND WARRANTY

This program is copyright 2010-2011 Baron Schwartz, 2011-2012 Percona Inc. Feedback and improvements are
welcome.
THIS PROGRAM IS PROVIDED “AS IS” AND WITHOUT ANY EXPRESS OR IMPLIED WARRANTIES, IN-
CLUDING, WITHOUT LIMITATION, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
FOR A PARTICULAR PURPOSE.
This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public
License as published by the Free Software Foundation, version 2; OR the Perl Artistic License. On UNIX and similar
systems, you can issue ‘man perlgpl’ or ‘man perlartistic’ to read these licenses.
You should have received a copy of the GNU General Public License along with this program; if not, write to the Free
Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA.

2.1.13 VERSION

pt-align 2.1.1

2.2 pt-archiver

2.2.1 NAME

pt-archiver - Archive rows from a MySQL table into another table or a ﬁle.

2.2.2 SYNOPSIS

Usage

pt-archiver [OPTION...] --source DSN --where WHERE

2.2. pt-archiver 7


pt-archiver nibbles records from a MySQL table. The –source and –dest arguments use DSN syntax; if COPY is yes,
–dest defaults to the key’s value from –source.

Examples

Archive all rows from oltp_server to olap_server and to a file:
pt-archiver --source h=oltp_server,D=test,t=tbl --dest h=olap_server
--file ’/var/log/archive/%Y-%m-%d-%D.%t’
--where "1=1" --limit 1000 --commit-each

Purge (delete) orphan rows from child table:
pt-archiver --source h=host,D=db,t=child --purge
--where ’NOT EXISTS(SELECT * FROM parent WHERE col=child.col)’

2.2.3 RISKS

pt-archiver is a read-write tool. It deletes data from the source by default, so you should test your archiving jobs with
the --dry-run option if you’re not sure about them. It is designed to have as little impact on production systems as
possible, but tuning with --limit, --txn-size and similar options might be a good idea too.
If you write or use --plugin modules, you should ensure they are good quality and well-tested.
At the time of this release there is an unverified bug with --bulk-insert that may cause data loss.
The authoritative source for updated information is always the online issue tracking system. Issues that affect this
tool will be marked as such. You can see a list of such issues at the following URL: http://www.percona.com/bugs/pt-
archiver.

2.2.4 DESCRIPTION

pt-archiver is the tool I use to archive tables as described in http://tinyurl.com/mysql-archiving. The goal is a low-
impact, forward-only job to nibble old data out of the table without impacting OLTP queries much. You can insert the
data into another table, which need not be on the same server. You can also write it to a file in a format suitable for
LOAD DATA INFILE. Or you can do neither, in which case it’s just an incremental DELETE.
pt-archiver is extensible via a plugin mechanism. You can inject your own code to add advanced archiving logic that
could be useful for archiving dependent data, applying complex business rules, or building a data warehouse during
the archiving process.
You need to choose values carefully for some options. The most important are --limit, --retries, and
--txn-size.
The strategy is to find the first row(s), then scan some index forward-only to find more rows efficiently. Each sub-
sequent query should not scan the entire table; it should seek into the index, then scan until it finds more archivable
rows. Specifying the index with the ‘i’ part of the --source argument can be crucial for this; use --dry-run to
examine the generated queries and be sure to EXPLAIN them to see if they are efficient (most of the time you prob-
ably want to scan the PRIMARY key, which is the default). Even better, profile pt-archiver with mk-query-profiler
(http://maatkit.org/get/mk-query-profiler) and make sure it is not scanning the whole table every query.

8 Chapter 2. Tools


You can disable the seek-then-scan optimizations partially or wholly with --no-ascend and --ascend-first.
Sometimes this may be more efficient for multi-column keys. Be aware that pt-archiver is built to start at the beginning
of the index it chooses and scan it forward-only. This might result in long table scans if you’re trying to nibble from
the end of the table by an index other than the one it prefers. See --source and read the documentation on the i
part if this applies to you.

2.2.5 OUTPUT

If you specify --progress, the output is a header row, plus status output at intervals. Each row in the status output
lists the current date and time, how many seconds pt-archiver has been running, and how many rows it has archived.
If you specify --statistics, pt-archiver outputs timing and other information to help you identify which part of
your archiving process takes the most time.

2.2.6 ERROR-HANDLING

pt-archiver tries to catch signals and exit gracefully; for example, if you send it SIGTERM (Ctrl-C on UNIX-ish
systems), it will catch the signal, print a message about the signal, and exit fairly normally. It will not execute
--analyze or --optimize, because these may take a long time to finish. It will run all other code normally,
including calling after_finish() on any plugins (see “EXTENDING”).
In other words, a signal, if caught, will break out of the main archiving loop and skip optimize/analyze.

2.2.7 OPTIONS

Specify at least one of --dest, --file, or --purge.
--ignore and --replace are mutually exclusive.
--txn-size and --commit-each are mutually exclusive.
--low-priority-insert and --delayed-insert are mutually exclusive.
--share-lock and --for-update are mutually exclusive.
--analyze and --optimize are mutually exclusive.
--no-ascend and --no-delete are mutually exclusive.
DSN values in --dest default to values from --source if COPY is yes.
-analyze
type: string
Run ANALYZE TABLE afterwards on --source and/or --dest.
Runs ANALYZE TABLE after finishing. The argument is an arbitrary string. If it contains the letter ‘s’, the
source will be analyzed. If it contains ‘d’, the destination will be analyzed. You can specify either or both. For
example, the following will analyze both:
--analyze=ds

See http://dev.mysql.com/doc/en/analyze-table.html for details on ANALYZE TABLE.
-ascend-first
Ascend only first column of index.
If you do want to use the ascending index optimization (see --no-ascend), but do not want to incur the
overhead of ascending a large multi-column index, you can use this option to tell pt-archiver to ascend only the

2.2. pt-archiver 9


leftmost column of the index. This can provide a significant performance boost over not ascending the index at
all, while avoiding the cost of ascending the whole index.
See “EXTENDING” for a discussion of how this interacts with plugins.
-ask-pass
Prompt for a password when connecting to MySQL.
-buffer
Buffer output to --file and flush at commit.
Disables autoflushing to --file and flushes --file to disk only when a transaction commits. This typically
means the file is block-flushed by the operating system, so there may be some implicit flushes to disk between
commits as well. The default is to flush --file to disk after every row.
The danger is that a crash might cause lost data.
The performance increase I have seen from using --buffer is around 5 to 15 percent. Your mileage may vary.
-bulk-delete
Delete each chunk with a single statement (implies --commit-each).
Delete each chunk of rows in bulk with a single DELETE statement. The statement deletes every row between
the first and last row of the chunk, inclusive. It implies --commit-each, since it would be a bad idea to
INSERT rows one at a time and commit them before the bulk DELETE.
The normal method is to delete every row by its primary key. Bulk deletes might be a lot faster. They also
might not be faster if you have a complex WHERE clause.
This option completely defers all DELETE processing until the chunk of rows is finished. If you have a plugin
on the source, its before_delete method will not be called. Instead, its before_bulk_delete method
is called later.
WARNING: if you have a plugin on the source that sometimes doesn’t return true from is_archivable(),
you should use this option only if you understand what it does. If the plugin instructs pt-archiver not to archive
a row, it will still be deleted by the bulk delete!
-[no]bulk-delete-limit
default: yes
Add --limit to --bulk-delete statement.
This is an advanced option and you should not disable it unless you know what you are doing and why! By
default, --bulk-delete appends a --limit clause to the bulk delete SQL statement. In certain cases, this
clause can be omitted by specifying --no-bulk-delete-limit. --limit must still be specified.
-bulk-insert
Insert each chunk with LOAD DATA INFILE (implies --bulk-delete --commit-each).
Insert each chunk of rows with LOAD DATA LOCAL INFILE. This may be much faster than inserting a row
at a time with INSERT statements. It is implemented by creating a temporary file for each chunk of rows, and
writing the rows to this file instead of inserting them. When the chunk is finished, it uploads the rows.
To protect the safety of your data, this option forces bulk deletes to be used. It would be unsafe to delete each
row as it is found, before inserting the rows into the destination first. Forcing bulk deletes guarantees that the
deletion waits until the insertion is successful.
The --low-priority-insert, --replace, and --ignore options work with this option, but
--delayed-insert does not.
-charset
short form: -A; type: string

10 Chapter 2. Tools


Default character set. If the value is utf8, sets Perl’s binmode on STDOUT to utf8, passes the mysql_enable_utf8
option to DBD::mysql, and runs SET NAMES UTF8 after connecting to MySQL. Any other value sets binmode
on STDOUT without the utf8 layer, and runs SET NAMES after connecting to MySQL.
See also --[no]check-charset.
-[no]check-charset
default: yes
Ensure connection and table character sets are the same. Disabling this check may cause text to be erroneously
converted from one character set to another (usually from utf8 to latin1) which may cause data loss or mojibake.
Disabling this check may be useful or necessary when character set conversions are intended.
-[no]check-columns
default: yes
Ensure --source and --dest have same columns.
Enabled by default; causes pt-archiver to check that the source and destination tables have the same columns. It
does not check column order, data type, etc. It just checks that all columns in the source exist in the destination
and vice versa. If there are any differences, pt-archiver will exit with an error.
To disable this check, specify –no-check-columns.
-check-interval
type: time; default: 1s
How often to check for slave lag if --check-slave-lag is given.
-check-slave-lag
type: string
Pause archiving until the specified DSN’s slave lag is less than --max-lag.
-columns
short form: -c; type: array
Comma-separated list of columns to archive.
Specify a comma-separated list of columns to fetch, write to the file, and insert into the destination table. If
specified, pt-archiver ignores other columns unless it needs to add them to the SELECT statement for ascending
an index or deleting rows. It fetches and uses these extra columns internally, but does not write them to the file
or to the destination table. It does pass them to plugins.
See also --primary-key-only.
-commit-each
Commit each set of fetched and archived rows (disables --txn-size).
Commits transactions and flushes --file after each set of rows has been archived, before fetching the next set
of rows, and before sleeping if --sleep is specified. Disables --txn-size; use --limit to control the
transaction size with --commit-each.
This option is useful as a shortcut to make --limit and --txn-size the same value, but more importantly
it avoids transactions being held open while searching for more rows. For example, imagine you are archiving
old rows from the beginning of a very large table, with --limit 1000 and --txn-size 1000. After some
period of finding and archiving 1000 rows at a time, pt-archiver finds the last 999 rows and archives them,
then executes the next SELECT to find more rows. This scans the rest of the table, but never finds any more
rows. It has held open a transaction for a very long time, only to determine it is finished anyway. You can use
--commit-each to avoid this.
-config
type: Array

2.2. pt-archiver 11


Read this comma-separated list of config files; if specified, this must be the first option on the command line.
-delayed-insert
Add the DELAYED modifier to INSERT statements.
Adds the DELAYED modifier to INSERT or REPLACE statements. See
http://dev.mysql.com/doc/en/insert.html for details.
-dest
type: DSN
DSN specifying the table to archive to.
This item specifies a table into which pt-archiver will insert rows archived from --source. It uses the same
key=val argument format as --source. Most missing values default to the same values as --source, so
you don’t have to repeat options that are the same in --source and --dest. Use the --help option to see
which values are copied from --source.
WARNING: Using a default options file (F) DSN option that defines a socket for --source causes pt-
archiver to connect to --dest using that socket unless another socket for --dest is specified. This means
that pt-archiver may incorrectly connect to --source when it connects to --dest. For example:
--source F=host1.cnf,D=db,t=tbl --dest h=host2

When pt-archiver connects to --dest, host2, it will connect via the --source, host1, socket defined in
host1.cnf.
-dry-run
Print queries and exit without doing anything.
Causes pt-archiver to exit after printing the filename and SQL statements it will use.
-file
type: string
File to archive to, with DATE_FORMAT()-like formatting.
Filename to write archived rows to. A subset of MySQL’s DATE_FORMAT() formatting codes are allowed in
the filename, as follows:
%d Day of the month, numeric (01..31)
%H Hour (00..23)
%i Minutes, numeric (00..59)
%m Month, numeric (01..12)
%s Seconds (00..59)
%Y Year, numeric, four digits

You can use the following extra format codes too:
%D Database name
%t Table name

Example:
--file ’/var/log/archive/%Y-%m-%d-%D.%t’

The file’s contents are in the same format used by SELECT INTO OUTFILE, as documented in the MySQL
manual: rows terminated by newlines, columns terminated by tabs, NULL characters are represented by N, and
special characters are escaped by . This lets you reload a file with LOAD DATA INFILE’s default settings.
If you want a column header at the top of the file, see --header. The file is auto-flushed by default; see
--buffer.

12 Chapter 2. Tools


-for-update
Adds the FOR UPDATE modifier to SELECT statements.
For details, see http://dev.mysql.com/doc/en/innodb-locking-reads.html.
-header
Print column header at top of --file.
Writes column names as the first line in the file given by --file. If the file exists, does not write headers; this
keeps the file loadable with LOAD DATA INFILE in case you append more output to it.
-help
Show help and exit.
-high-priority-select
Adds the HIGH_PRIORITY modifier to SELECT statements.
See http://dev.mysql.com/doc/en/select.html for details.
-host
short form: -h; type: string
Connect to host.
-ignore
Use IGNORE for INSERT statements.
Causes INSERTs into --dest to be INSERT IGNORE.
-limit
type: int; default: 1
Number of rows to fetch and archive per statement.
Limits the number of rows returned by the SELECT statements that retrieve rows to archive. Default is one
row. It may be more efficient to increase the limit, but be careful if you are archiving sparsely, skipping over
many rows; this can potentially cause more contention with other queries, depending on the storage engine,
transaction isolation level, and options such as --for-update.
-local
Do not write OPTIMIZE or ANALYZE queries to binlog.
Adds the NO_WRITE_TO_BINLOG modifier to ANALYZE and OPTIMIZE queries. See --analyze for
details.
-low-priority-delete
Adds the LOW_PRIORITY modifier to DELETE statements.
See http://dev.mysql.com/doc/en/delete.html for details.
-low-priority-insert
Adds the LOW_PRIORITY modifier to INSERT or REPLACE statements.
See http://dev.mysql.com/doc/en/insert.html for details.
-max-lag
Pause archiving if the slave given by --check-slave-lag lags.
This option causes pt-archiver to look at the slave every time it’s about to fetch another row. If the slave’s lag
is greater than the option’s value, or if the slave isn’t running (so its lag is NULL), pt-table-checksum sleeps
for --check-interval seconds and then looks at the lag again. It repeats until the slave is caught up, then
proceeds to fetch and archive the row.

2.2. pt-archiver 13


This option may eliminate the need for --sleep or --sleep-coef.
-no-ascend
Do not use ascending index optimization.
The default ascending-index optimization causes pt-archiver to optimize repeated SELECT queries so they
seek into the index where the previous query ended, then scan along it, rather than scanning from the beginning
of the table every time. This is enabled by default because it is generally a good strategy for repeated accesses.
Large, multiple-column indexes may cause the WHERE clause to be complex enough that this could actually
be less efficient. Consider for example a four-column PRIMARY KEY on (a, b, c, d). The WHERE clause to
start where the last query ended is as follows:
WHERE (a > ?)
OR (a = ? AND b > ?)
OR (a = ? AND b = ? AND c > ?)
OR (a = ? AND b = ? AND c = ? AND d >= ?)

Populating the placeholders with values uses memory and CPU, adds network traffic and parsing overhead, and
may make the query harder for MySQL to optimize. A four-column key isn’t a big deal, but a ten-column key
in which every column allows NULL might be.
Ascending the index might not be necessary if you know you are simply removing rows from the beginning
of the table in chunks, but not leaving any holes, so starting at the beginning of the table is actually the most
efficient thing to do.
See also --ascend-first. See “EXTENDING” for a discussion of how this interacts with plugins.
-no-delete
Do not delete archived rows.
Causes pt-archiver not to delete rows after processing them. This disallows --no-ascend, because enabling
them both would cause an infinite loop.
If there is a plugin on the source DSN, its before_delete method is called anyway, even though pt-archiver
will not execute the delete. See “EXTENDING” for more on plugins.
-optimize
type: string
Run OPTIMIZE TABLE afterwards on --source and/or --dest.
Runs OPTIMIZE TABLE after finishing. See --analyze for the option syntax and
http://dev.mysql.com/doc/en/optimize-table.html for details on OPTIMIZE TABLE.
-password
short form: -p; type: string
Password to use when connecting.
-pid
type: string
Create the given PID file when daemonized. The file contains the process ID of the daemonized instance. The
PID file is removed when the daemonized instance exits. The program checks for the existence of the PID file
when starting; if it exists and the process with the matching PID exists, the program exits.
-plugin
type: string
Perl module name to use as a generic plugin.
Specify the Perl module name of a general-purpose plugin. It is currently used only for statistics (see
--statistics) and must have new() and a statistics() method.

14 Chapter 2. Tools


The new( src = $src, dst => $dst, opts => $o )> method gets the source and destination DSNs, and their
database connections, just like the connection-specific plugins do. It also gets an OptionParser object ($o) for
accessing command-line options (example: ‘‘$o-‘‘get(‘purge’);>).
The statistics(%stats, $time) method gets a hashref of the statistics collected by the archiving
job, and the time the whole job started.
-port
short form: -P; type: int
Port number to use for connection.
-primary-key-only
Primary key columns only.
A shortcut for specifying --columns with the primary key columns. This is an efficiency if you just want
to purge rows; it avoids fetching the entire row, when only the primary key columns are needed for DELETE
statements. See also --purge.
-progress
type: int
Print progress information every X rows.
Prints current time, elapsed time, and rows archived every X rows.
-purge
Purge instead of archiving; allows omitting --file and --dest.
Allows archiving without a --file or --dest argument, which is effectively a purge since the rows are just
deleted.
If you just want to purge rows, consider specifying the table’s primary key columns with
--primary-key-only. This will prevent fetching all columns from the server for no reason.
-quick-delete
Adds the QUICK modifier to DELETE statements.
See http://dev.mysql.com/doc/en/delete.html for details. As stated in the documentation, in some cases it may
be faster to use DELETE QUICK followed by OPTIMIZE TABLE. You can use --optimize for this.
-quiet
short form: -q
Do not print any output, such as for --statistics.
Suppresses normal output, including the output of --statistics, but doesn’t suppress the output from
--why-quit.
-replace
Causes INSERTs into --dest to be written as REPLACE.
-retries
Number of retries per timeout or deadlock.
Specifies the number of times pt-archiver should retry when there is an InnoDB lock wait timeout or deadlock.
When retries are exhausted, pt-archiver will exit with an error.
Consider carefully what you want to happen when you are archiving between a mixture of transactional and
non-transactional storage engines. The INSERT to --dest and DELETE from --source are on separate
connections, so they do not actually participate in the same transaction even if they’re on the same server.

2.2. pt-archiver 15


However, pt-archiver implements simple distributed transactions in code, so commits and rollbacks should
happen as desired across the two connections.
At this time I have not written any code to handle errors with transactional storage engines other than InnoDB.
Request that feature if you need it.
-run-time
type: time
Time to run before exiting.
Optional suffix s=seconds, m=minutes, h=hours, d=days; if no suffix, s is used.
-[no]safe-auto-increment
default: yes
Do not archive row with max AUTO_INCREMENT.
Adds an extra WHERE clause to prevent pt-archiver from removing the newest row when ascending a single-
column AUTO_INCREMENT key. This guards against re-using AUTO_INCREMENT values if the server
restarts, and is enabled by default.
The extra WHERE clause contains the maximum value of the auto-increment column as of the beginning of the
archive or purge job. If new rows are inserted while pt-archiver is running, it will not see them.
-sentinel
type: string; default: /tmp/pt-archiver-sentinel
Exit if this file exists.
The presence of the file specified by --sentinel will cause pt-archiver to stop archiving and exit. The
default is /tmp/pt-archiver-sentinel. You might find this handy to stop cron jobs gracefully if necessary. See also
--stop.
-set-vars
type: string; default: wait_timeout=10000
Set these MySQL variables.
Specify any variables you want to be set immediately after connecting to MySQL. These will be included in a
SET command.
-share-lock
Adds the LOCK IN SHARE MODE modifier to SELECT statements.
See http://dev.mysql.com/doc/en/innodb-locking-reads.html.
-skip-foreign-key-checks
Disables foreign key checks with SET FOREIGN_KEY_CHECKS=0.
-sleep
type: int
Sleep time between fetches.
Specifies how long to sleep between SELECT statements. Default is not to sleep at all. Transactions are NOT
committed, and the --file file is NOT flushed, before sleeping. See --txn-size to control that.
If --commit-each is specified, committing and flushing happens before sleeping.
-sleep-coef
type: float
Calculate --sleep as a multiple of the last SELECT time.

16 Chapter 2. Tools


If this option is specified, pt-archiver will sleep for the query time of the last SELECT multiplied by the
specified coefficient.
This is a slightly more sophisticated way to throttle the SELECTs: sleep a varying amount of time between each
SELECT, depending on how long the SELECTs are taking.
-socket
short form: -S; type: string
Socket file to use for connection.
-source
type: DSN
DSN specifying the table to archive from (required). This argument is a DSN. See DSN OPTIONS for the
syntax. Most options control how pt-archiver connects to MySQL, but there are some extended DSN options
in this tool’s syntax. The D, t, and i options select a table to archive:
--source h=my_server,D=my_database,t=my_tbl

The a option specifies the database to set as the connection’s default with USE. If the b option is true, it disables
binary logging with SQL_LOG_BIN. The m option specifies pluggable actions, which an external Perl module
can provide. The only required part is the table; other parts may be read from various places in the environment
(such as options files).
The ‘i’ part deserves special mention. This tells pt-archiver which index it should scan to archive. This appears
in a FORCE INDEX or USE INDEX hint in the SELECT statements used to fetch archivable rows. If you don’t
specify anything, pt-archiver will auto-discover a good index, preferring a PRIMARY KEY if one exists. In my
experience this usually works well, so most of the time you can probably just omit the ‘i’ part.
The index is used to optimize repeated accesses to the table; pt-archiver remembers the last row it retrieves from
each SELECT statement, and uses it to construct a WHERE clause, using the columns in the specified index,
that should allow MySQL to start the next SELECT where the last one ended, rather than potentially scanning
from the beginning of the table with each successive SELECT. If you are using external plugins, please see
“EXTENDING” for a discussion of how they interact with ascending indexes.
The ‘a’ and ‘b’ options allow you to control how statements flow through the binary log. If you specify the ‘b’
option, binary logging will be disabled on the specified connection. If you specify the ‘a’ option, the connection
will USE the specified database, which you can use to prevent slaves from executing the binary log events with
--replicate-ignore-db options. These two options can be used as different methods to achieve the same
goal: archive data off the master, but leave it on the slave. For example, you can run a purge job on the master
and prevent it from happening on the slave using your method of choice.
WARNING: Using a default options file (F) DSN option that defines a socket for --source causes pt-
archiver to connect to --dest using that socket unless another socket for --dest is specified. This means
that pt-archiver may incorrectly connect to --source when it is meant to connect to --dest. For example:
--source F=host1.cnf,D=db,t=tbl --dest h=host2

When pt-archiver connects to --dest, host2, it will connect via the --source, host1, socket defined in
host1.cnf.
-statistics
Collect and print timing statistics.
Causes pt-archiver to collect timing statistics about what it does. These statistics are available to the plugin
specified by --plugin
Unless you specify --quiet, pt-archiver prints the statistics when it exits. The statistics look like this:

2.2. pt-archiver 17


Started at 2008-07-18T07:18:53, ended at 2008-07-18T07:18:53
Source: D=db,t=table
SELECT 4
INSERT 4
DELETE 4
Action Count Time Pct
commit 10 0.1079 88.27
select 5 0.0047 3.87
deleting 4 0.0028 2.29
inserting 4 0.0028 2.28
other 0 0.0040 3.29

The first two (or three) lines show times and the source and destination tables. The next three lines show how
many rows were fetched, inserted, and deleted.
The remaining lines show counts and timing. The columns are the action, the total number of times that action
was timed, the total time it took, and the percent of the program’s total runtime. The rows are sorted in order of
descending total time. The last row is the rest of the time not explicitly attributed to anything. Actions will vary
depending on command-line options.
If --why-quit is given, its behavior is changed slightly. This option causes it to print the reason for exiting
even when it’s just because there are no more rows.
This option requires the standard Time::HiRes module, which is part of core Perl on reasonably new Perl re-
leases.
-stop
Stop running instances by creating the sentinel file.
Causes pt-archiver to create the sentinel file specified by --sentinel and exit. This should have the effect
of stopping all running instances which are watching the same sentinel file.
-txn-size
Number of rows per transaction.
Specifies the size, in number of rows, of each transaction. Zero disables transactions altogether. After pt-
archiver processes this many rows, it commits both the --source and the --dest if given, and flushes the
file given by --file.
This parameter is critical to performance. If you are archiving from a live server, which for example is doing
heavy OLTP work, you need to choose a good balance between transaction size and commit overhead. Larger
transactions create the possibility of more lock contention and deadlocks, but smaller transactions cause more
frequent commit overhead, which can be significant. To give an idea, on a small test set I worked with while
writing pt-archiver, a value of 500 caused archiving to take about 2 seconds per 1000 rows on an otherwise
quiet MySQL instance on my desktop machine, archiving to disk and to another table. Disabling transactions
with a value of zero, which turns on autocommit, dropped performance to 38 seconds per thousand rows.
If you are not archiving from or to a transactional storage engine, you may want to disable transactions so
pt-archiver doesn’t try to commit.
-user
short form: -u; type: string
User for login if not current user.
-version
Show version and exit.
-where
type: string

18 Chapter 2. Tools


WHERE clause to limit which rows to archive (required).
Specifies a WHERE clause to limit which rows are archived. Do not include the word WHERE. You may need
to quote the argument to prevent your shell from interpreting it. For example:
--where ’ts < current_date - interval 90 day’

For safety, --where is required. If you do not require a WHERE clause, use --where 1=1.
-why-quit
Print reason for exiting unless rows exhausted.
Causes pt-archiver to print a message if it exits for any reason other than running out of rows to archive.
This can be useful if you have a cron job with --run-time specified, for example, and you want to be sure
pt-archiver is finishing before running out of time.
If --statistics is given, the behavior is changed slightly. It will print the reason for exiting even when it’s
just because there are no more rows.
This output prints even if --quiet is given. That’s so you can put pt-archiver in a cron job and get an email
if there’s an abnormal exit.

2.2.8 DSN OPTIONS

These DSN options are used to create a DSN. Each option is given like option=value. The options are case-
sensitive, so P and p are not the same option. There cannot be whitespace before or after the = and if the value
contains whitespace it must be quoted. DSN options are comma-separated. See the percona-toolkit manpage for full
details.
• a
copy: no
Database to USE when executing queries.
• A
dsn: charset; copy: yes
Default character set.
• b
copy: no
If true, disable binlog with SQL_LOG_BIN.
• D
dsn: database; copy: yes
Database that contains the table.
• F
dsn: mysql_read_default_file; copy: yes
Only read default options from the given file
• h
dsn: host; copy: yes
Connect to host.
• i

2.2. pt-archiver 19


copy: yes
Index to use.
• m
copy: no
Plugin module name.
• p
dsn: password; copy: yes
• P
dsn: port; copy: yes
• S
dsn: mysql_socket; copy: yes
• t
copy: yes
Table to archive from/to.
• u
dsn: user; copy: yes

2.2.9 EXTENDING

pt-archiver is extensible by plugging in external Perl modules to handle some logic and/or actions. You can specify a
module for both the --source and the --dest, with the ‘m’ part of the speciﬁcation. For example:
--source D=test,t=test1,m=My::Module1 --dest m=My::Module2,t=test2

This will cause pt-archiver to load the My::Module1 and My::Module2 packages, create instances of them, and then
make calls to them during the archiving process.
You can also specify a plugin with --plugin.
The module must provide this interface:
new(dbh => $dbh, db => $db_name, tbl => $tbl_name)
The plugin’s constructor is passed a reference to the database handle, the database name, and table name.
The plugin is created just after pt-archiver opens the connection, and before it examines the table given
in the arguments. This gives the plugin a chance to create and populate temporary tables, or do other setup
work.
before_begin(cols => @cols, allcols => @allcols)
This method is called just before pt-archiver begins iterating through rows and archiving them, but after
it does all other setup work (examining table structures, designing SQL queries, and so on). This is the
only time pt-archiver tells the plugin column names for the rows it will pass the plugin while archiving.

20 Chapter 2. Tools


The cols argument is the column names the user requested to be archived, either by default or by the
--columns option. The allcols argument is the list of column names for every row pt-archiver will
fetch from the source table. It may fetch more columns than the user requested, because it needs some
columns for its own use. When subsequent plugin functions receive a row, it is the full row containing all
the extra columns, if any, added to the end.
is_archivable(row => @row)
This method is called for each row to determine whether it is archivable. This applies only to --source.
The argument is the row itself, as an arrayref. If the method returns true, the row will be archived;
otherwise it will be skipped.
Skipping a row adds complications for non-unique indexes. Normally pt-archiver uses a WHERE clause
designed to target the last processed row as the place to start the scan for the next SELECT statement. If
you have skipped the row by returning false from is_archivable(), pt-archiver could get into an infinite
loop because the row still exists. Therefore, when you specify a plugin for the --source argument, pt-
archiver will change its WHERE clause slightly. Instead of starting at “greater than or equal to” the last
processed row, it will start “strictly greater than.” This will work fine on unique indexes such as primary
keys, but it may skip rows (leave holes) on non-unique indexes or when ascending only the first column
of an index.
pt-archiver will change the clause in the same way if you specify --no-delete, because again an
infinite loop is possible.
If you specify the --bulk-delete option and return false from this method, pt-archiver may not do
what you want. The row won’t be archived, but it will be deleted, since bulk deletes operate on ranges of
rows and don’t know which rows the plugin selected to keep.
If you specify the --bulk-insert option, this method’s return value will influence whether the row
is written to the temporary file for the bulk insert, so bulk inserts will work as expected. However, bulk
inserts require bulk deletes.
before_delete(row => @row)
This method is called for each row just before it is deleted. This applies only to --source. This is a
good place for you to handle dependencies, such as deleting things that are foreign-keyed to the row you
are about to delete. You could also use this to recursively archive all dependent tables.
This plugin method is called even if --no-delete is given, but not if --bulk-delete is given.
before_bulk_delete(first_row => @row, last_row => @row)
This method is called just before a bulk delete is executed. It is similar to the before_delete method,
except its arguments are the first and last row of the range to be deleted. It is called even if --no-delete
is given.
before_insert(row => @row)
This method is called for each row just before it is inserted. This applies only to --dest. You could
use this to insert the row into multiple tables, perhaps with an ON DUPLICATE KEY UPDATE clause to
build summary tables in a data warehouse.
This method is not called if --bulk-insert is given.
before_bulk_insert(first_row => @row, last_row => @row, filename => bulk_insert_filename)
This method is called just before a bulk insert is executed. It is similar to the before_insert method,
except its arguments are the first and last row of the range to be deleted.
custom_sth(row => @row, sql => $sql)
This method is called just before inserting the row, but after “before_insert()”. It allows the plugin to
specify different INSERT statement if desired. The return value (if any) should be a DBI statement

2.2. pt-archiver 21


handle. The sql parameter is the SQL text used to prepare the default INSERT statement. This method
is not called if you specify --bulk-insert.
If no value is returned, the default INSERT statement handle is used.
This method applies only to the plugin specified for --dest, so if your plugin isn’t doing what you
expect, check that you’ve specified it for the destination and not the source.
custom_sth_bulk(first_row => @row, last_row => @row, sql => $sql, filename => $bulk_insert_filename)
If you’ve specified --bulk-insert, this method is called just before the bulk insert, but after “be-
fore_bulk_insert()”, and the arguments are different.
This method’s return value etc is similar to the “custom_sth()” method.
after_finish()
This method is called after pt-archiver exits the archiving loop, commits all database handles, closes
--file, and prints the final statistics, but before pt-archiver runs ANALYZE or OPTIMIZE (see
--analyze and --optimize).
If you specify a plugin for both --source and --dest, pt-archiver constructs, calls before_begin(), and calls
after_finish() on the two plugins in the order --source, --dest.
pt-archiver assumes it controls transactions, and that the plugin will NOT commit or roll back the database handle.
The database handle passed to the plugin’s constructor is the same handle pt-archiver uses itself. Remember that
--source and --dest are separate handles.
A sample module might look like this:
package My::Module;

sub new {
my ( $class, %args ) = @_;
return bless(%args, $class);
}

sub before_begin {
my ( $self, %args ) = @_;
# Save column names for later
$self->{cols} = $args{cols};
}

sub is_archivable {
my ( $self, %args ) = @_;
# Do some advanced logic with $args{row}
return 1;
}

sub before_delete {} # Take no action
sub before_insert {} # Take no action
sub custom_sth {} # Take no action
sub after_finish {} # Take no action

1;

2.2.10 ENVIRONMENT

The environment variable PTDEBUG enables verbose debugging output to STDERR. To enable debugging and capture
all output to a file, run the tool like:

22 Chapter 2. Tools


PTDEBUG=1 pt-archiver ... > FILE 2>&1

Be careful: debugging output is voluminous and can generate several megabytes of output.


You need Perl, DBI, DBD::mysql, and some core packages that ought to be installed in any reasonably new version of
Perl.

2.2.12 BUGS

For a list of known bugs, see http://www.percona.com/bugs/pt-archiver.
• Tool --version

2.2.13 DOWNLOADING






2.2.14 AUTHORS

Baron Schwartz

2.2.15 ACKNOWLEDGMENTS

Andrew O’Brien

2.2. pt-archiver 23





welcome.

2.2.18 VERSION

pt-archiver 2.1.1

2.3 pt-config-diff

2.3.1 NAME

pt-config-diff - Diff MySQL configuration files and server variables.

2.3.2 SYNOPSIS

Usage

pt-config-diff [OPTION...] CONFIG CONFIG [CONFIG...]

pt-config-diff diffs MySQL configuration files and server variables. CONFIG can be a filename or a DSN. At least
two CONFIG sources must be given. Like standard Unix diff, there is no output if there are no differences.
Diff host1 config from SHOW VARIABLES against host2:
pt-config-diff h=host1 h=host2

Diff config from [mysqld] section in my.cnf against host1 config:
pt-config-diff /etc/my.cnf h=host1

Diff the [mysqld] section of two option files:

24 Chapter 2. Tools


pt-config-diff /etc/my-small.cnf /etc/my-large.cnf

2.3.3 RISKS

pt-config-diff reads MySQL’s configuration and examines it and is thus very low risk.
At the time of this release there are no known bugs that pose a serious risk.
config-diff.

2.3.4 DESCRIPTION

pt-config-diff diffs MySQL configurations by examining the values of server system variables from two or more
CONFIG sources specified on the command line. A CONFIG source can be a DSN or a filename containing the
output of mysqld --help --verbose, my_print_defaults, SHOW VARIABLES, or an option file (e.g.
my.cnf).
For each DSN CONFIG, pt-config-diff connects to MySQL and gets variables and values by executing SHOW
/*!40103 GLOBAL*/ VARIABLES. This is an “active config” because it shows what server values MySQL is
actively (currently) running with.
Only variables that all CONFIG sources have are compared because if a variable is not present then we cannot know
or safely guess its value. For example, if you compare an option file (e.g. my.cnf) to an active config (i.e. SHOW
VARIABLES from a DSN CONFIG), the option file will probably only have a few variables, whereas the active config
has every variable. Only values of the variables present in both configs are compared.
Option file and DSN configs provide the best results.

2.3.5 OUTPUT

There is no output when there are no differences. When there are differences, pt-config-diff prints a report to STDOUT
that looks similar to the following:
2 config differences
Variable my.master.cnf my.slave.cnf
========================= =============== ===============
datadir /tmp/12345/data /tmp/12346/data
port 12345 12346

Comparing MySQL variables is difficult because there are many variations and subtleties across the many versions
and distributions of MySQL. When a comparison fails, the tool prints a warning to STDERR, such as the following:
Comparing log_error values (mysqld.log, /tmp/12345/data/mysqld.log)
caused an error: Argument "/tmp/12345/data/mysqld.log" isn’t numeric
in numeric eq (==) at ./pt-config-diff line 2311.

Please report these warnings so the comparison functions can be improved.

2.3. pt-config-diff 25


2.3.6 EXIT STATUS

pt-conﬁg-diff exits with a zero exit status when there are no differences, and 1 if there are.

2.3.7 OPTIONS

This tool accepts additional command-line arguments. Refer to the “SYNOPSIS” and usage information for details.
-ask-pass
-charset
-config
type: Array
(This option does not specify a CONFIG; it’s equivalent to --defaults-file.)
-daemonize
Fork to the background and detach from the shell. POSIX operating systems only.
-defaults-file
short form: -F; type: string
Only read mysql options from the given ﬁle. You must give an absolute pathname.
-help
Show help and exit.
-host
Connect to host.
-ignore-variables
type: array
Ignore, do not compare, these variables.
-password
Password to use for connection.
-pid
type: string
-port

26 Chapter 2. Tools


-[no]report
default: yes
Print the MySQL config diff report to STDOUT. If you just want to check if the given configs are different or
not by examining the tool’s exit status, then specify --no-report to suppress the report.
-report-width
Truncate report lines to this many characters. Since some variable values can be long, or when comparing
multiple configs, it may help to increase the report width so values are not truncated beyond readability.
-set-vars
Set these MySQL variables. Immediately after connecting to MySQL, this string will be appended to SET and
executed.
-socket
-user
MySQL user if not current user.
-version

2.3.8 DSN OPTIONS

details.
• A
• D
Default database.
• F
• h
Connect to host.
• p

2.3. pt-config-diff 27


• P
• S
• u

2.3.9 ENVIRONMENT

PTDEBUG=1 pt-config-diff ... > FILE 2>&1



Perl.

2.3.11 BUGS

For a list of known bugs, see http://www.percona.com/bugs/pt-conﬁg-diff.
• Tool --version

2.3.12 DOWNLOADING


28 Chapter 2. Tools







2.3.13 AUTHORS

Baron Schwartz and Daniel Nichter




This program is copyright 2011-2012 Percona Inc. Feedback and improvements are welcome.

2.3.16 VERSION

pt-conﬁg-diff 2.1.1

2.4 pt-deadlock-logger

2.4.1 NAME

pt-deadlock-logger - Extract and log MySQL deadlock information.

2.4. pt-deadlock-logger 29


2.4.2 SYNOPSIS

Usage

pt-deadlock-logger [OPTION...] SOURCE_DSN

pt-deadlock-logger extracts and saves information about the most recent deadlock in a MySQL server.
Print deadlocks on SOURCE_DSN:
pt-deadlock-logger SOURCE_DSN

Store deadlock information from SOURCE_DSN in test.deadlocks table on SOURCE_DSN (source and destination
are the same host):
pt-deadlock-logger SOURCE_DSN --dest D=test,t=deadlocks

Store deadlock information from SOURCE_DSN in test.deadlocks table on DEST_DSN (source and destination are
different hosts):
pt-deadlock-logger SOURCE_DSN --dest DEST_DSN,D=test,t=deadlocks

Daemonize and check for deadlocks on SOURCE_DSN every 30 seconds for 4 hours:
pt-deadlock-logger SOURCE_DSN --dest D=test,t=deadlocks --daemonize --run-time 4h --interval 30s

2.4.3 RISKS

pt-deadlock-logger is a read-only tool unless you specify a --dest table. In some cases polling SHOW INNODB
STATUS too rapidly can cause extra load on the server. If you’re using it on a production server under very heavy
load, you might want to set --interval to 30 seconds or more.
deadlock-logger.

2.4.4 DESCRIPTION

pt-deadlock-logger extracts deadlock data from a MySQL server. Currently only InnoDB deadlock information is
available. You can print the information to standard output, store it in a database table, or both. If neither --print
nor --dest are given, then the deadlock information is printed by default. If only --dest is given, then the deadlock
information is only stored. If both options are given, then the deadlock information is printed and stored.
The source host can be speciﬁed using one of two methods. The ﬁrst method is to use at least one of the standard
connection-related command line options: --defaults-file, --password, --host, --port, --socket
or --user. These options only apply to the source host; they cannot be used to specify the destination host.
The second method to specify the source host, or the optional destination host using --dest, is a DSN. A
DSN is a special syntax that can be either just a hostname (like server.domain.com or 1.2.3.4), or a
key=value,key=value string. Keys are a single letter:

30 Chapter 2. Tools


KEY MEANING
=== =======
h Connect to host
P Port number to use for connection
S Socket file to use for connection
u User for login if not current user
p Password to use when connecting
F Only read default options from the given file

If you omit any values from the destination host DSN, they are ﬁlled in with values from the source host, so you don’t
need to specify them in both places. pt-deadlock-logger reads all normal MySQL option ﬁles, such as ~/.my.cnf, so
you may not need to specify username, password and other common options at all.

2.4.5 OUTPUT

You can choose which columns are output and/or saved to --dest with the --columns argument. The default
columns are as follows:
server
The (source) server on which the deadlock occurred. This might be useful if you’re tracking deadlocks on
many servers.
ts
The date and time of the last detected deadlock.
thread
The MySQL thread number, which is the same as the connection ID in SHOW FULL PROCESSLIST.
txn_id
The InnoDB transaction ID, which InnoDB expresses as two unsigned integers. I have multiplied them
out to be one number.
txn_time
How long the transaction was active when the deadlock happened.
user
The connection’s database username.
hostname
The connection’s host.
ip
The connection’s IP address. If you specify --numeric-ip, this is converted to an unsigned integer.
db
The database in which the deadlock occurred.
tbl
The table on which the deadlock occurred.
idx
The index on which the deadlock occurred.
lock_type



The lock type the transaction held on the lock that caused the deadlock.
lock_mode
The lock mode of the lock that caused the deadlock.
wait_hold
Whether the transaction was waiting for the lock or holding the lock. Usually you will see the two waited-
for locks.
victim
Whether the transaction was selected as the deadlock victim and rolled back.
query
The query that caused the deadlock.

2.4.6 INNODB CAVEATS AND DETAILS

InnoDB’s output is hard to parse and sometimes there’s no way to do it right.
Sometimes not all information (for example, username or IP address) is included in the deadlock information. In this
case there’s nothing for the script to put in those columns. It may also be the case that the deadlock output is so long
(because there were a lot of locks) that the whole thing is truncated.
Though there are usually two transactions involved in a deadlock, there are more locks than that; at a minimum,
one more lock than transactions is necessary to create a cycle in the waits-for graph. pt-deadlock-logger prints the
transactions (always two in the InnoDB output, even when there are more transactions in the waits-for graph than that)
and ﬁlls in locks. It prefers waited-for over held when choosing lock information to output, but you can ﬁgure out the
rest with a moment’s thought. If you see one wait-for and one held lock, you’re looking at the same lock, so of course
you’d prefer to see both wait-for locks and get more information. If the two waited-for locks are not on the same table,
more than two transactions were involved in the deadlock.

2.4.7 OPTIONS

-ask-pass
-charset
-clear-deadlocks
type: string
Use this table to create a small deadlock. This usually has the effect of clearing out a huge deadlock, which
otherwise consumes the entire output of SHOW INNODB STATUS. The table must not exist. pt-deadlock-
logger will create it with the following MAGIC_clear_deadlocks structure:
CREATE TABLE test.deadlock_maker(a INT PRIMARY KEY) ENGINE=InnoDB;

After creating the table and causing a small deadlock, the tool will drop the table again.

32 Chapter 2. Tools


-[no]collapse
Collapse whitespace in queries to a single space. This might make it easier to inspect on the command line or
in a query. By default, whitespace is collapsed when printing with --print, but not modified when storing to
--dest. (That is, the default is different for each action).
-columns
type: hash
Output only this comma-separated list of columns. See “OUTPUT” for more details on columns.
-config
type: Array
-create-dest-table
Create the table specified by --dest.
Normally the --dest table is expected to exist already. This option causes pt-deadlock-logger to create the
table automatically using the suggested table structure.
-daemonize
-defaults-file
-dest
type: DSN
DSN for where to store deadlocks; specify at least a database (D) and table (t).
Missing values are filled in with the same values from the source host, so you can usually omit most parts of
this argument if you’re storing deadlocks on the same server on which they happen.
By default, whitespace in the query column is left intact; use --[no]collapse if you want whitespace
collapsed.
The following MAGIC_dest_table is suggested if you want to store all the information pt-deadlock-logger can
extract about deadlocks:
CREATE TABLE deadlocks (
server char(20) NOT NULL,
ts datetime NOT NULL,
thread int unsigned NOT NULL,
txn_id bigint unsigned NOT NULL,
txn_time smallint unsigned NOT NULL,
user char(16) NOT NULL,
hostname char(20) NOT NULL,
ip char(15) NOT NULL, -- alternatively, ip int unsigned NOT NULL
db char(64) NOT NULL,
tbl char(64) NOT NULL,
idx char(64) NOT NULL,
lock_type char(16) NOT NULL,
lock_mode char(1) NOT NULL,
wait_hold char(1) NOT NULL,
victim tinyint unsigned NOT NULL,
query text NOT NULL,
PRIMARY KEY (server,ts,thread)
) ENGINE=InnoDB



If you use --columns, you can omit whichever columns you don’t want to store.
-help
Show help and exit.
-host
Connect to host.
-interval
type: time
How often to check for deadlocks. If no --run-time is specified, pt-deadlock-logger runs forever, checking
for deadlocks at every interval. See also --run-time.
-log
type: string
Print all output to this file when daemonized.
-numeric-ip
Express IP addresses as integers.
-password
-pid
type: string
-port
-print
Print results on standard output. See “OUTPUT” for more. By default, enables --[no]collapse unless you
explicitly disable it.
If --interval or --run-time is specified, only new deadlocks are printed at each interval. A fingerprint
for each deadlock is created using --columns server, ts and thread (even if those columns were not specified
by --columns) and if the current deadlock’s fingerprint is different from the last deadlock’s fingerprint, then
it is printed.
-run-time
type: time
How long to run before exiting. By default pt-deadlock-logger runs once, checks for deadlocks, and exits. If
--run-time is specified but no --interval is specified, a default 1 second interval will be used.
-set-vars
executed.
-socket

34 Chapter 2. Tools


-tab
Print tab-separated columns, instead of aligned.
-user
-version

2.4.8 DSN OPTIONS

details.
• A
• D
Default database.
• F
• h
Connect to host.
• p
• P
• S
• t
Table in which to store deadlock information.
• u




2.4.9 ENVIRONMENT

PTDEBUG=1 pt-deadlock-logger ... > FILE 2>&1



Perl.

2.4.11 BUGS

For a list of known bugs, see http://www.percona.com/bugs/pt-deadlock-logger.
• Tool --version

2.4.12 DOWNLOADING






2.4.13 AUTHORS

Baron Schwartz

36 Chapter 2. Tools





welcome.

2.4.16 VERSION

pt-deadlock-logger 2.1.1

2.5 pt-diskstats

2.5.1 NAME

pt-diskstats - An interactive I/O monitoring tool for GNU/Linux.

2.5.2 SYNOPSIS

Usage

pt-diskstats [OPTION...] [FILES]

pt-diskstats prints disk I/O statistics for GNU/Linux. It is somewhat similar to iostat, but it is interactive and more
detailed. It can analyze samples gathered from another machine.

2.5.3 RISKS

pt-diskstats simply reads /proc/diskstats. It should be very low-risk.

2.5. pt-diskstats 37


diskstats.

2.5.4 DESCRIPTION

The pt-diskstats tool is similar to iostat, but has some advantages. It prints read and write statistics separately, and
has more columns. It is menu-driven and interactive, with several different ways to aggregate the data. It integrates
well with the pt-stalk tool. It also does the “right thing” by default, such as hiding disks that are idle. These properties
make it very convenient for quickly drilling down into I/O performance and inspecting disk behavior.
This program works in two modes. The default is to collect samples of /proc/diskstats and print out the formatted
statistics at intervals. The other mode is to process a file that contains saved samples of /proc/diskstats; there is a shell
script later in this documentation that shows how to collect such a file.
In both cases, the tool is interactively controlled by keystrokes, so you can redisplay and slice the data flexibly and
easily. It loops forever, until you exit with the ‘q’ key. If you press the ‘?’ key, you will bring up the interactive help
menu that shows which keys control the program.
When the program is gathering samples of /proc/diskstats and refreshing its display, it prints information about the
newest sample each time it refreshes. When it is operating on a file of saved samples, it redraws the entire file’s
contents every time you change an option.
The program doesn’t print information about every block device on the system. It hides devices that it has never
observed to have any activity. You can enable and disable this by pressing the ‘i’ key.

2.5.5 OUTPUT

In the rest of this documentation, we will try to clarify the distinction between block devices (/dev/sda1, for example),
which the kernel presents to the application via a filesystem, versus the (usually) physical device underneath the block
device, which could be a disk, a RAID controller, and so on. We will sometimes refer to logical I/O operations, which
occur at the block device, versus physical I/Os which are performed on the underlying device. When we refer to the
queue, we are speaking of the queue associated with the block device, which holds requests until they’re issued to the
physical device.
The program’s output looks like the following sample, which is too wide for this manual page, so we have formatted
it as several samples with line breaks:
#ts device rd_s rd_avkb rd_mb_s rd_mrg rd_cnc rd_rt
{6} sda 0.9 4.2 0.0 0% 0.0 17.9
{6} sdb 0.4 4.0 0.0 0% 0.0 26.1
{6} dm-0 0.0 4.0 0.0 0% 0.0 13.5
{6} dm-1 0.8 4.0 0.0 0% 0.0 16.0

... wr_s wr_avkb wr_mb_s wr_mrg wr_cnc wr_rt
... 99.7 6.2 0.6 35% 3.7 23.7
... 14.5 15.8 0.2 75% 0.5 9.2
... 1.0 4.0 0.0 0% 0.0 2.3
... 117.7 4.0 0.5 0% 4.1 35.1

... busy in_prg io_s qtime stime
... 6% 0 100.6 23.3 0.4
... 4% 0 14.9 8.6 0.6
... 0% 0 1.1 1.5 1.2
... 5% 0 118.5 34.5 0.4

38 Chapter 2. Tools


The columns are as follows:
#ts
This column’s contents vary depending on the tool’s aggregation mode. In the default mode, when each
line contains information about a single disk but possibly aggregates across several samples from that disk,
this column shows the number of samples that were included into the line of output, in {curly braces}. In
the example shown, each line of output aggregates {10} samples of /proc/diskstats.
In the “all” group-by mode, this column shows timestamp offsets, relative to the time the tool began
aggregating or the timestamp of the previous lines printed, depending on the mode. The output can be
confusing to explain, but it’s rather intuitive when you see the lines appearing on your screen periodically.
Similarly, in “sample” group-by mode, the number indicates the total time span that is grouped into each
sample.
If you specify --show-timestamps, this field instead shows the timestamp at which the sample was
taken; if multiple timestamps are present in a single line of output, then the first timestamp is used.
device
The device name. If there is more than one device, then instead the number of devices aggregated into the
line is shown, in {curly braces}.
rd_s
The average number of reads per second. This is the number of I/O requests that were sent to the underly-
ing device. This usually is a smaller number than the number of logical IO requests made by applications.
More requests might have been queued to the block device, but some of them usually are merged before
being sent to the disk.
This field is computed from the contents of /proc/diskstats as follows. See “KERNEL DOCUMENTA-
TION” below for the meaning of the field numbers:
delta[field1] / delta[time]

rd_avkb
The average size of the reads, in kilobytes. This field is computed as follows:
2 * delta[field3] / delta[field1]

rd_mb_s
The average number of megabytes read per second. Computed as follows:
2 * delta[field3] / delta[time]

rd_mrg
The percentage of read requests that were merged together in the queue scheduler before being sent to the
physical device. The field is computed as follows:
100 * delta[field2] / (delta[field2] + delta[field1])

rd_cnc
The average concurrency of the read operations, as computed by Little’s Law. This is the end-to-end
concurrency on the block device, not the underlying disk’s concurrency. It includes time spent in the
queue. The field is computed as follows:
delta[field4] / delta[time] / 1000 / devices-in-group

rd_rt



The average response time of the read operations, in milliseconds. This is the end-to-end response time,
including time spent in the queue. It is the response time that the application making I/O requests sees,
not the response time of the physical disk underlying the block device. It is computed as follows:
delta[field4] / (delta[field1] + delta[field2])

wr_s, wr_avkb, wr_mb_s, wr_mrg, wr_cnc, wr_rt
These columns show write activity, and they match the corresponding columns for read activity.
busy
The fraction of wall-clock time that the device had at least one request in progress; this is what iostat
calls %util, and indeed it is utilization, depending on how you define utilization, but that is sometimes
ambiguous in common parlance. It may also be called the residence time; the time during which at least
one request was resident in the system. It is computed as follows:
100 * delta[field10] / (1000 * delta[time])

This field cannot exceed 100% unless there is a rounding error, but it is a common mistake to think that a
device that’s busy all the time is saturated. A device such as a RAID volume should support concurrency
higher than 1, and solid-state drives can support very high concurrency. Concurrency can grow without
bound, and is a more reliable indicator of how loaded the device really is.
in_prg
The number of requests that were in progress. Unlike the read and write concurrencies, which are averages
that are generated from reliable numbers, this number is an instantaneous sample, and you can see that
it might represent a spike of requests, rather than the true long-term average. If this number is large, it
essentially means that the device is heavily loaded. It is computed as follows:
field9

ios_s
The average throughput of the physical device, in I/O operations per second (IOPS). This column shows
the total IOPS the underlying device is handling. It is the sum of rd_s and wr_s.
qtime
The average queue time; that is, time a request spends in the device scheduler queue before being sent to
the physical device. This is an average over reads and writes.
It is computed in a slightly complex way: the average response time seen by the application, minus the
average service time (see the description of the next column). This is derived from the queueing theory
formula for response time, R = W + S: response time = queue time + service time. This is solved for W,
of course, to give W = R - S. The computation follows:
delta[field11] / (delta[field1, 2, 5, 6] + delta[field9])
- delta[field10] / delta[field1, 2, 5, 6]

See the description for stime for more details and cautions.
stime
The average service time; that is, the time elapsed while the physical device processes the request, after
the request finishes waiting in the queue. This is an average over reads and writes. It is computed from
the queueing theory utilization formula, U = SX, solved for S. This means that utilization divided by
throughput gives service time:
delta[field10] / (delta[field1, 2, 5, 6])

40 Chapter 2. Tools


Note, however, that there can be some kernel bugs that cause field 9 in /proc/diskstats to become negative,
and this can cause field 10 to be wrong, thus making the service time computation not wholly trustworthy.
Note that in the above formula we use utilization very specifically. It is a duration, not a percentage.
You can compare the stime and qtime columns to see whether the response time for reads and writes is
spent in the queue or on the physical device. However, you cannot see the difference between reads and
writes. Changing the block device scheduler algorithm might improve queue time greatly. The default
algorithm, cfq, is very bad for servers, and should only be used on laptops and workstations that perform
tasks such as working with spreadsheets and surfing the Internet.
If you are used to using iostat, you might wonder where you can find the same information in pt-diskstats. Here are
two samples of output from both tools on the same machine at the same time, for /dev/sda, wrapped to fit:
#ts dev rd_s rd_avkb rd_mb_s rd_mrg rd_cnc rd_rt
08:50:10 sda 0.0 0.0 0.0 0% 0.0 0.0
08:50:20 sda 0.4 4.0 0.0 0% 0.0 15.5
08:50:30 sda 2.1 4.4 0.0 0% 0.0 21.1
08:50:40 sda 2.4 4.0 0.0 0% 0.0 15.4
08:50:50 sda 0.1 4.0 0.0 0% 0.0 33.0

wr_s wr_avkb wr_mb_s wr_mrg wr_cnc wr_rt
7.7 25.5 0.2 84% 0.0 0.3
49.6 6.8 0.3 41% 2.4 28.8
210.1 5.6 1.1 28% 7.4 25.2
297.1 5.4 1.6 26% 11.4 28.3
11.9 11.7 0.1 66% 0.2 4.9

busy in_prg io_s qtime stime
1% 0 7.7 0.1 0.2
6% 0 50.0 28.1 0.7
12% 0 212.2 24.8 0.4
16% 0 299.5 27.8 0.4
1% 0 12.0 4.7 0.3

Dev rrqm/s wrqm/s r/s w/s rMB/s wMB/s
08:50:10 sda 0.00 41.40 0.00 7.70 0.00 0.19
08:50:20 sda 0.00 34.70 0.40 49.60 0.00 0.33
08:50:30 sda 0.00 83.30 2.10 210.10 0.01 1.15
08:50:40 sda 0.00 105.10 2.40 297.90 0.01 1.58
08:50:50 sda 0.00 22.50 0.10 11.10 0.00 0.13

avgrq-sz avgqu-sz await svctm %util
51.01 0.02 2.04 1.25 0.96
13.55 2.44 48.76 1.16 5.79
11.15 7.45 35.10 0.55 11.76
10.81 11.40 37.96 0.53 15.97
24.07 0.17 15.60 0.87 0.97

The correspondence between the columns is not one-to-one. In particular:
rrqm/s, wrqm/s
These columns in iostat are replaced by rd_mrg and wr_mrg in pt-diskstats.
avgrq-sz
This column is in sectors in iostat, and is a combination of reads and writes. The pt-diskstats output
breaks these out separately and shows them in kB. You can derive it via a weighted average of rd_avkb
and wr_avkb in pt-diskstats, and then multiply by 2 to get sectors (each sector is 512 bytes).



avgqu-sz
This column really represents concurrency at the block device scheduler. The pt-diskstats output shows
concurrency for reads and writes separately: rd_cnc and wr_cnc.
await
This column is the average response time from the beginning to the end of a request to the block device,
including queue time and service time, and is not shown in pt-diskstats. Instead, pt-diskstats shows
individual response times at the disk level for reads and writes (rd_rt and wr_rt), as well as queue time
versus service time for reads and writes in aggregate.
svctm
This column is the average service time at the disk, and is shown as stime in pt-diskstats.
%util
This column is called busy in pt-diskstats. Utilization is usually defined as the portion of time during
which there was at least one active request, not as a percentage, which is why we chose to avoid this
confusing term.

2.5.6 COLLECTING DATA

It is straightforward to gather a sample of data for this tool. Files should have this format, with a timestamp line
preceding each sample of statistics:
TS <timestamp>
<contents of /proc/diskstats>
TS <timestamp>
<contents of /proc/diskstats>
... et cetera

You can simply use pt-diskstats with --save-samples to collect this data for you. If you wish to capture samples
as part of some other tool, and use pt-diskstats to analyze them, you can include a snippet of shell script such as the
following:
INTERVAL=1
while true; do
sleep=$(date +%s.%N | awk "{print $INTERVAL - ($1 % $INTERVAL)}")
sleep $sleep
date +"TS %s.%N %F %T" >> diskstats-samples.txt
cat /proc/diskstats >> diskstats-samples.txt
done

2.5.7 KERNEL DOCUMENTATION

This documentation supplements the official documentation|http://www.kernel.org/doc/Documentation/iostats.txt on
the contents of /proc/diskstats. That documentation can sometimes be difficult to understand for those who are not
familiar with Linux kernel internals. The contents of /proc/diskstats are generated by the diskstats_show()
function in the kernel source file block/genhd.c.
Here is a sample of /proc/diskstats on a recent kernel.
8 1 sda1 426 243 3386 2056 3 0 18 87 0 2135 2142

The fields in this sample are as follows. The first three fields are the major and minor device numbers (8, 1), and the
device name (sda1). They are followed by 11 fields of statistics:

42 Chapter 2. Tools


1. The number of reads completed. This is the number of physical reads done by the underlying disk, not the
number of reads that applications made from the block device. This means that 426 actual reads have completed
successfully to the disk on which /dev/sda1 resides. Reads are not counted until they complete.
2. The number of reads merged because they were adjacent. In the sample, 243 reads were merged. This means
that /dev/sda1 actually received 869 logical reads, but sent only 426 physical reads to the underlying physical
device.
3. The number of sectors read successfully. The 426 physical reads to the disk read 3386 sectors. Sectors are 512
bytes, so a total of about 1.65MB have been read from /dev/sda1.
4. The number of milliseconds spent reading. This counts only reads that have completed, not reads that are in
progress. It counts the time spent from when requests are placed on the queue until they complete, not the
time that the underlying disk spends servicing the requests. That is, it measures the total response time seen by
applications, not disk response times.
5. Ditto for field 1, but for writes.
9. The number of I/Os currently in progress, that is, they’ve been scheduled by the queue scheduler and issued to
the disk (submitted to the underlying disk’s queue), but not yet completed. There are bugs in some kernels that
cause this number, and thus fields 10 and 11, to be wrong sometimes.
10. The total number of milliseconds spent doing I/Os. This is not the total response time seen by the applications;
it is the total amount of time during which at least one I/O was in progress. If one I/O is issued at time 100,
another comes in at 101, and both of them complete at 102, then this field increments by 2, not 3.
11. This field counts the total response time of all I/Os. In contrast to field 10, it counts double when two I/Os
overlap. In our previous example, this field would increment by 3, not 2.

2.5.8 OPTIONS

-columns-regex
type: string; default: .
Print columns that match this Perl regex.
-config
type: Array
-devices-regex
type: string
Print devices that match this Perl regex.
-group-by
type: string; default: disk
Group-by mode: disk, sample, or all. In disk mode, each line of output shows one disk device. In sample mode,
each line of output shows one sample of statistics. In all mode, each line of output shows one sample and one
disk device.



-headers
type: Hash; default: group,scroll
If group is present, each sample will be separated by a blank line, unless the sample is only one line. If
scroll is present, the tool will print the headers as often as needed to prevent them from scrolling out of view.
Note that you can press the space bar, or the enter key, to reprint headers at will.
-help
Show help and exit.
-interval
When in interactive mode, wait N seconds before printing to the screen. Also, how often the tool should sample
/proc/diskstats.
The tool attempts to gather statistics exactly on even intervals of clock time. That is, if you specify a 5-second
interval, it will try to capture samples at 12:00:00, 12:00:05, and so on; it will not gather at 12:00:01, 12:00:06
and so forth.
This can lead to slightly odd delays in some circumstances, because the tool waits one full cycle before printing
out the first set of lines. (Unlike iostat and vmstat, pt-diskstats does not start with a line representing the
averages since the computer was booted.) Therefore, the rule has an exception to avoid very long delays.
Suppose you specify a 10-second interval, but you start the tool at 12:00:00.01. The tool might wait until
12:00:20 to print its first lines of output, and in the intervening 19.99 seconds, it would appear to do nothing.
To alleviate this, the tool waits until the next even interval of time to gather, unless more than 20% of that interval
remains. This means the tool will never wait more than 120% of the sampling interval to produce output, e.g if
you start the tool at 12:00:53 with a 10-second sampling interval, then the first sample will be only 7 seconds
long, not 10 seconds.
-iterations
type: int
When in interactive mode, stop after N samples. Run forever by default.
-sample-time
In –group-by sample mode, include N seconds of samples per group.
-save-samples
type: string
File to save diskstats samples in; these can be used for later analysis.
-show-inactive
Show inactive devices.
-show-timestamps
Show a ‘HH:MM:SS’ timestamp in the #ts column. If multiple timestamps are aggregated into one line, the
first timestamp is shown.
-version

2.5.9 ENVIRONMENT


44 Chapter 2. Tools


PTDEBUG=1 pt-diskstats ... > FILE 2>&1



This tool requires Perl v5.8.0 or newer and the /proc ﬁlesystem, unless reading from ﬁles.

2.5.11 BUGS

For a list of known bugs, see http://www.percona.com/bugs/pt-diskstats.
• Tool --version

2.5.12 DOWNLOADING






2.5.13 AUTHORS

Baron Schwartz, Brian Fraser, and Daniel Nichter






welcome.

2.5.16 VERSION

pt-diskstats 2.1.1

2.6 pt-duplicate-key-checker

2.6.1 NAME

pt-duplicate-key-checker - Find duplicate indexes and foreign keys on MySQL tables.

2.6.2 SYNOPSIS

Usage

pt-duplicate-key-checker [OPTION...] [DSN]

pt-duplicate-key-checker examines MySQL tables for duplicate or redundant indexes and foreign keys. Connection
options are read from MySQL option ﬁles.
pt-duplicate-key-checker --host host1

2.6.3 RISKS

pt-duplicate-key-checker is a read-only tool that executes SHOW CREATE TABLE and related queries to inspect
table structures, and thus is very low-risk.
At the time of this release, there is an unconﬁrmed bug that causes the tool to crash.
duplicate-key-checker.

46 Chapter 2. Tools


2.6.4 DESCRIPTION

This program examines the output of SHOW CREATE TABLE on MySQL tables, and if it finds indexes that cover
the same columns as another index in the same order, or cover an exact leftmost prefix of another index, it prints
out the suspicious indexes. By default, indexes must be of the same type, so a BTREE index is not a duplicate of a
FULLTEXT index, even if they have the same columns. You can override this.
It also looks for duplicate foreign keys. A duplicate foreign key covers the same columns as another in the same table,
and references the same parent table.

2.6.5 OPTIONS

-all-structs
Compare indexes with different structs (BTREE, HASH, etc).
By default this is disabled, because a BTREE index that covers the same columns as a FULLTEXT index is not
really a duplicate, for example.
-ask-pass
-charset
-[no]clustered
default: yes
PK columns appended to secondary key is duplicate.
Detects when a suffix of a secondary key is a leftmost prefix of the primary key, and treats it as a duplicate
key. Only detects this condition on storage engines whose primary keys are clustered (currently InnoDB and
solidDB).
Clustered storage engines append the primary key columns to the leaf nodes of all secondary keys anyway, so
you might consider it redundant to have them appear in the internal nodes as well. Of course, you may also want
them in the internal nodes, because just having them at the leaf nodes won’t help for some queries. It does help
for covering index queries, however.
Here’s an example of a key that is considered redundant with this option:
PRIMARY KEY (‘a‘)
KEY ‘b‘ (‘b‘,‘a‘)

The use of such indexes is rather subtle. For example, suppose you have the following query:
SELECT ... WHERE b=1 ORDER BY a;

This query will do a filesort if we remove the index on b,a. But if we shorten the index on b,a to just b and
also remove the ORDER BY, the query should return the same results.
The tool suggests shortening duplicate clustered keys by dropping the key and re-adding it without the primary
key prefix. The shortened clustered key may still duplicate another key, but the tool cannot currently detect
when this happens without being ran a second time to re-check the newly shortened clustered keys. Therefore,
if you shorten any duplicate clustered keys, you should run the tool again.

2.6. pt-duplicate-key-checker 47


-config
type: Array
-databases
short form: -d; type: hash
Check only this comma-separated list of databases.
-defaults-file
-engines
short form: -e; type: hash
Check only tables whose storage engine is in this comma-separated list.
-help
Show help and exit.
-host
Connect to host.
-ignore-databases
type: Hash
Ignore this comma-separated list of databases.
-ignore-engines
type: Hash
Ignore this comma-separated list of storage engines.
-ignore-order
Ignore index order so KEY(a,b) duplicates KEY(b,a).
-ignore-tables
type: Hash
Ignore this comma-separated list of tables. Table names may be qualified with the database name.
-key-types
type: string; default: fk
Check for duplicate f=foreign keys, k=keys or fk=both.
-password
-pid
type: string
Create the given PID file. The file contains the process ID of the script. The PID file is removed when the script
exits. Before starting, the script checks if the PID file already exists. If it does not, then the script creates and
writes its own PID to it. If it does, then the script checks the following: if the file contains a PID and a process
is running with that PID, then the script dies; or, if there is no process running with that PID, then the script
overwrites the file with its own PID and starts; else, if the file contains no PID, then the script dies.

48 Chapter 2. Tools


-port
-set-vars
executed.
-socket
-[no]sql
default: yes
Print DROP KEY statement for each duplicate key. By default an ALTER TABLE DROP KEY statement is
printed below each duplicate key so that, if you want to remove the duplicate key, you can copy-paste the
statement into MySQL.
To disable printing these statements, specify --no-sql.
-[no]summary
default: yes
Print summary of indexes at end of output.
-tables
short form: -t; type: hash
Check only this comma-separated list of tables.
Table names may be qualiﬁed with the database name.
-user
-verbose
short form: -v
Output all keys and/or foreign keys found, not just redundant ones.
-version

2.6.6 DSN OPTIONS

details.
• A
• D



Default database.
• F
• h
Connect to host.
• p
• P
• S
• u

2.6.7 ENVIRONMENT

PTDEBUG=1 pt-duplicate-key-checker ... > FILE 2>&1



Perl.

2.6.9 BUGS

For a list of known bugs, see http://www.percona.com/bugs/pt-duplicate-key-checker.
• Tool --version

50 Chapter 2. Tools



2.6.10 DOWNLOADING






2.6.11 AUTHORS





welcome.

2.6.14 VERSION

pt-duplicate-key-checker 2.1.1



2.7 pt-fifo-split

2.7.1 NAME

pt-fifo-split - Split files and pipe lines to a fifo without really splitting.

2.7.2 SYNOPSIS

Usage

pt-fifo-split [options] [FILE ...]

pt-fifo-split splits FILE and pipes lines to a fifo. With no FILE, or when FILE is -, read standard input.
Read hugefile.txt in chunks of a million lines without physically splitting it:
pt-fifo-split --lines 1000000 hugefile.txt
while [ -e /tmp/pt-fifo-split ]; do cat /tmp/pt-fifo-split; done

2.7.3 RISKS

pt-fifo-split creates and/or deletes the --fifo file. Otherwise, no other files are modified, and it merely reads lines
from the file given on the command-line. It should be very low-risk.
will be marked as such. You can see a list of such issues at the following URL: http://www.percona.com/bugs/pt-fifo-
split.

2.7.4 DESCRIPTION

pt-fifo-split lets you read from a file as though it contains only some of the lines in the file. When you read from it
again, it contains the next set of lines; when you have gone all the way through it, the file disappears. This works only
on Unix-like operating systems.
You can specify multiple files on the command line. If you don’t specify any, or if you use the special filename -,
lines are read from standard input.

2.7.5 OPTIONS

-config
type: Array

52 Chapter 2. Tools


-fifo
type: string; default: /tmp/pt-fifo-split
The name of the fifo from which the lines can be read.
-force
Remove the fifo if it exists already, then create it again.
-help
Show help and exit.
-lines
The number of lines to read in each chunk.
-offset
Begin at the Nth line. If the argument is 0, all lines are printed to the fifo. If 1, then beginning at the first line,
lines are printed (exactly the same as 0). If 2, the first line is skipped, and the 2nd and subsequent lines are
printed to the fifo.
-pid
type: string
-statistics
Print out statistics between chunks. The statistics are the number of chunks, the number of lines, elapsed time,
and lines per second overall and during the last chunk.
-version

2.7.6 ENVIRONMENT

PTDEBUG=1 pt-fifo-split ... > FILE 2>&1



Perl.

2.7.8 BUGS

For a list of known bugs, see http://www.percona.com/bugs/pt-fifo-split.

2.7. pt-fifo-split 53


• Tool --version

2.7.9 DOWNLOADING






2.7.10 AUTHORS

Baron Schwartz




welcome.

54 Chapter 2. Tools


2.7.13 VERSION

pt-fifo-split 2.1.1

2.8 pt-find

2.8.1 NAME

pt-find - Find MySQL tables and execute actions, like GNU find.

2.8.2 SYNOPSIS

Usage

pt-find [OPTION...] [DATABASE...]

pt-find searches for MySQL tables and executes actions, like GNU find. The default action is to print the database
and table name.
Find all tables created more than a day ago, which use the MyISAM engine, and print their names:
pt-find --ctime +1 --engine MyISAM

Find InnoDB tables that haven’t been updated in a month, and convert them to MyISAM storage engine (data ware-
housing, anyone?):
pt-find --mtime +30 --engine InnoDB --exec "ALTER TABLE %D.%N ENGINE=MyISAM"

Find tables created by a process that no longer exists, following the name_sid_pid naming convention, and remove
them.
pt-find --connection-id ’D_d+_(d+)$’ --server-id ’D_(d+)_d+$’ --exec-plus "DROP TABLE %s"

Find empty tables in the test and junk databases, and delete them:
pt-find --empty junk test --exec-plus "DROP TABLE %s"

Find tables more than five gigabytes in total size:
pt-find --tablesize +5G

Find all tables and print their total data and index size, and sort largest tables first (sort is a different program, by the
way).
pt-find --printf "%Tt%D.%Nn" | sort -rn

As above, but this time, insert the data back into the database for posterity:
pt-find --noquote --exec "INSERT INTO sysdata.tblsize(db, tbl, size) VALUES(’%D’, ’%N’, %T)"

2.8.3 RISKS


2.8. pt-find 55


pt-find only reads and prints information by default, but --exec and --exec-plus can execute user-defined SQL.
You should be as careful with it as you are with any command-line tool that can execute queries against your database.
will be marked as such. You can see a list of such issues at the following URL: http://www.percona.com/bugs/pt-find.

2.8.4 DESCRIPTION

pt-find looks for MySQL tables that pass the tests you specify, and executes the actions you specify. The default
action is to print the database and table name to STDOUT.
pt-find is simpler than GNU find. It doesn’t allow you to specify complicated expressions on the command line.
pt-find uses SHOW TABLES when possible, and SHOW TABLE STATUS when needed.

2.8.5 OPTION TYPES

There are three types of options: normal options, which determine some behavior or setting; tests, which determine
whether a table should be included in the list of tables found; and actions, which do something to the tables pt-find
finds.
pt-find uses standard Getopt::Long option parsing, so you should use double dashes in front of long option names,
unlike GNU find.

2.8.6 OPTIONS

-ask-pass
-case-insensitive
Specifies that all regular expression searches are case-insensitive.
-charset
-config
type: Array
-day-start
Measure times (for --mmin, etc) from the beginning of today rather than from the current time.
-defaults-file

56 Chapter 2. Tools


-help
Show help and exit.
-host
Connect to host.
-or
Combine tests with OR, not AND.
By default, tests are evaluated as though there were an AND between them. This option switches it to OR.
Option parsing is not implemented by pt-find itself, so you cannot specify complicated expressions with paren-
theses and mixtures of OR and AND.
-password
-pid
type: string
-port
-[no]quote
default: yes
Quotes MySQL identifier names with MySQL’s standard backtick character.
Quoting happens after tests are run, and before actions are run.
-set-vars
executed.
-socket
-user
-version

2.8. pt-find 57


2.8.7 TESTS

Most tests check some criterion against a column of SHOW TABLE STATUS output. Numeric arguments can be
specified as +n for greater than n, -n for less than n, and n for exactly n. All numeric options can take an optional
suffix multiplier of k, M or G (1_024, 1_048_576, and 1_073_741_824 respectively). All patterns are Perl regular
expressions (see ‘man perlre’) unless specified as SQL LIKE patterns.
Dates and times are all measured relative to the same instant, when pt-find first asks the database server what time it
is. All date and time manipulation is done in SQL, so if you say to find tables modified 5 days ago, that translates to
SELECT DATE_SUB(CURRENT_TIMESTAMP, INTERVAL 5 DAY). If you specify --day-start, if course it’s
relative to CURRENT_DATE instead.
However, table sizes and other metrics are not consistent at an instant in time. It can take some time for MySQL to
process all the SHOW queries, and pt-find can’t do anything about that. These measurements are as of the time they’re
taken.
If you need some test that’s not in this list, file a bug report and I’ll enhance pt-find for you. It’s really easy.
-autoinc
type: string; group: Tests
Table’s next AUTO_INCREMENT is n. This tests the Auto_increment column.
-avgrowlen
type: size; group: Tests
Table avg row len is n bytes. This tests the Avg_row_length column. The specified size can be “NULL” to test
where Avg_row_length IS NULL.
-checksum
Table checksum is n. This tests the Checksum column.
-cmin
Table was created n minutes ago. This tests the Create_time column.
-collation
Table collation matches pattern. This tests the Collation column.
-column-name
A column name in the table matches pattern.
-column-type
A column in the table matches this type (case-insensitive).
Examples of types are: varchar, char, int, smallint, bigint, decimal, year, timestamp, text, enum.
-comment
Table comment matches pattern. This tests the Comment column.
-connection-id

58 Chapter 2. Tools


Table name has nonexistent MySQL connection ID. This tests the table name for a pattern. The argument to this
test must be a Perl regular expression that captures digits like this: (d+). If the table name matches the pattern,
these captured digits are taken to be the MySQL connection ID of some process. If the connection doesn’t exist
according to SHOW FULL PROCESSLIST, the test returns true. If the connection ID is greater than pt-find‘s
own connection ID, the test returns false for safety.
Why would you want to do this? If you use MySQL statement-based replication, you probably know the trouble
temporary tables can cause. You might choose to work around this by creating real tables with unique names,
instead of temporary tables. One way to do this is to append your connection ID to the end of the table, thusly:
scratch_table_12345. This assures the table name is unique and lets you have a way to find which connection
it was associated with. And perhaps most importantly, if the connection no longer exists, you can assume the
connection died without cleaning up its tables, and this table is a candidate for removal.
This is how I manage scratch tables, and that’s why I included this test in pt-find.
The argument I use to --connection-id is “D_(d+)$”. That finds tables with a series of numbers at the
end, preceded by an underscore and some non-number character (the latter criterion prevents me from examining
tables with a date at the end, which people tend to do: baron_scratch_2007_05_07 for example). It’s better to
keep the scratch tables separate of course.
If you do this, make sure the user pt-find runs as has the PROCESS privilege! Otherwise it will only see
connections from the same user, and might think some tables are ready to remove when they’re still in use. For
safety, pt-find checks this for you.
See also --server-id.
-createopts
Table create option matches pattern. This tests the Create_options column.
-ctime
Table was created n days ago. This tests the Create_time column.
-datafree
Table has n bytes of free space. This tests the Data_free column. The specified size can be “NULL” to test
where Data_free IS NULL.
-datasize
Table data uses n bytes of space. This tests the Data_length column. The specified size can be “NULL” to test
where Data_length IS NULL.
-dblike
Database name matches SQL LIKE pattern.
-dbregex
Database name matches this pattern.
-empty
group: Tests
Table has no rows. This tests the Rows column.

2.8. pt-find 59


-engine
Table storage engine matches this pattern. This tests the Engine column, or in earlier versions of MySQL, the
Type column.
-function
Function definition matches pattern.
-indexsize
Table indexes use n bytes of space. This tests the Index_length column. The specified size can be “NULL” to
test where Index_length IS NULL.
-kmin
Table was checked n minutes ago. This tests the Check_time column.
-ktime
Table was checked n days ago. This tests the Check_time column.
-mmin
Table was last modified n minutes ago. This tests the Update_time column.
-mtime
Table was last modified n days ago. This tests the Update_time column.
-procedure
Procedure definition matches pattern.
-rowformat
Table row format matches pattern. This tests the Row_format column.
-rows
Table has n rows. This tests the Rows column. The specified size can be “NULL” to test where Rows IS NULL.
-server-id
Table name contains the server ID. If you create temporary tables with the naming convention explained in
--connection-id, but also add the server ID of the server on which the tables are created, then you can use
this pattern match to ensure tables are dropped only on the server they’re created on. This prevents a table from
being accidentally dropped on a slave while it’s in use (provided that your server IDs are all unique, which they
should be for replication to work).
For example, on the master (server ID 22) you create a table called scratch_table_22_12345. If you see this
table on the slave (server ID 23), you might think it can be dropped safely if there’s no such connection 12345.
But if you also force the name to match the server ID with --server-id ’D_(d+)_d+$’, the table
won’t be dropped on the slave.

60 Chapter 2. Tools


-tablesize
Table uses n bytes of space. This tests the sum of the Data_length and Index_length columns.
-tbllike
Table name matches SQL LIKE pattern.
-tblregex
Table name matches this pattern.
-tblversion
Table version is n. This tests the Version column.
-trigger
Trigger action statement matches pattern.
-trigger-table
--trigger is defined on table matching pattern.
-view
CREATE VIEW matches this pattern.

2.8.8 ACTIONS

The --exec-plus action happens after everything else, but otherwise actions happen in an indeterminate order. If
you need determinism, file a bug report and I’ll add this feature.
-exec
type: string; group: Actions
Execute this SQL with each item found. The SQL can contain escapes and formatting directives (see
--printf).
-exec-dsn
Specify a DSN in key-value format to use when executing SQL with --exec and --exec-plus. Any values
not specified are inherited from command-line arguments.
-exec-plus
Execute this SQL with all items at once. This option is unlike --exec. There are no escaping or formatting
directives; there is only one special placeholder for the list of database and table names, %s. The list of tables
found will be joined together with commas and substituted wherever you place %s.
You might use this, for example, to drop all the tables you found:
DROP TABLE %s

2.8. pt-find 61


This is sort of like GNU find’s “-exec command {} +” syntax. Only it’s not totally cryptic. And it doesn’t
require me to write a command-line parser.
-print
group: Actions
Print the database and table name, followed by a newline. This is the default action if no other action is specified.
-printf
Print format on the standard output, interpreting ‘’ escapes and ‘%’ directives. Escapes are backslashed char-
acters, like n and t. Perl interprets these, so you can use any escapes Perl knows about. Directives are replaced
by %s, and as of this writing, you can’t add any special formatting instructions, like field widths or alignment
(though I’m musing over ways to do that).
Here is a list of the directives. Note that most of them simply come from columns of SHOW TABLE STATUS.
If the column is NULL or doesn’t exist, you get an empty string in the output. A % character followed by any
character not in the following list is discarded (but the other character is printed).
CHAR DATA SOURCE NOTES
---- ------------------ ------------------------------------------
a Auto_increment
A Avg_row_length
c Checksum
C Create_time
D Database The database name in which the table lives
d Data_length
E Engine In older versions of MySQL, this is Type
F Data_free
f Innodb_free Parsed from the Comment field
I Index_length
K Check_time
L Collation
M Max_data_length
N Name
O Comment
P Create_options
R Row_format
S Rows
T Table_length Data_length+Index_length
U Update_time
V Version

2.8.9 DSN OPTIONS

details.
• A
• D

62 Chapter 2. Tools


Default database.
• F
• h
Connect to host.
• p
• P
• S
• u

2.8.10 ENVIRONMENT

PTDEBUG=1 pt-find ... > FILE 2>&1



Perl.

2.8.12 BUGS

For a list of known bugs, see http://www.percona.com/bugs/pt-ﬁnd.
• Tool --version

2.8. pt-ﬁnd 63



2.8.13 DOWNLOADING






2.8.14 AUTHORS

Baron Schwartz




welcome.

2.8.17 VERSION

pt-ﬁnd 2.1.1

64 Chapter 2. Tools


2.9 pt-fingerprint

2.9.1 NAME

pt-fingerprint - Convert queries into fingerprints.

2.9.2 SYNOPSIS

Usage

pt-fingerprint [OPTIONS] [FILES]

pt-fingerprint converts queries into fingerprints. With the –query option, converts the option’s value into a fingerprint.
With no options, treats command-line arguments as FILEs and reads and converts semicolon-separated queries from
the FILEs. When FILE is -, it read standard input.
Convert a single query:
pt-fingerprint --query "select a, b, c from users where id = 500"

Convert a file full of queries:
pt-fingerprint /path/to/file.txt

2.9.3 RISKS

The pt-fingerprint tool simply reads data and transforms it, so risks are minimal.

2.9.4 DESCRIPTION

A query fingerprint is the abstracted form of a query, which makes it possible to group similar queries together.
Abstracting a query removes literal values, normalizes whitespace, and so on. For example, consider these two queries:
SELECT name, password FROM user WHERE id=’12823’;
select name, password from user
where id=5;

Both of those queries will fingerprint to
select name, password from user where id=?

Once the query’s fingerprint is known, we can then talk about a query as though it represents all similar queries.
Query fingerprinting accommodates a great many special cases, which have proven necessary in the real world. For
example, an IN list with 5 literals is really equivalent to one with 4 literals, so lists of literals are collapsed to a single
one. If you want to understand more about how and why all of these cases are handled, please review the test cases in
the Subversion repository. If you find something that is not fingerprinted properly, please submit a bug report with a
reproducible test case. Here is a list of transformations during fingerprinting, which might not be exhaustive:

2.9. pt-fingerprint 65


• Group all SELECT queries from mysqldump together, even if they are against different tables. Ditto for all of
pt-table-checksum’s checksum queries.
• Shorten multi-value INSERT statements to a single VALUES() list.
• Strip comments.
• Abstract the databases in USE statements, so all USE statements are grouped together.
• Replace all literals, such as quoted strings. For efficiency, the code that replaces literal numbers is somewhat
non-selective, and might replace some things as numbers when they really are not. Hexadecimal literals are
also replaced. NULL is treated as a literal. Numbers embedded in identifiers are also replaced, so tables named
similarly will be fingerprinted to the same values (e.g. users_2009 and users_2010 will fingerprint identically).
• Collapse all whitespace into a single space.
• Lowercase the entire query.
• Replace all literals inside of IN() and VALUES() lists with a single placeholder, regardless of cardinality.
• Collapse multiple identical UNION queries into a single one.

2.9.5 OPTIONS

-config
type: Array
-help
Show help and exit.
-match-embedded-numbers
Match numbers embedded in words and replace as single values. This option causes the tool to be more careful
about matching numbers so that words with numbers, like catch22 are matched and replaced as a single ?
placeholder. Otherwise the default number matching pattern will replace catch22 as catch?.
This is helpful if database or table names contain numbers.
-match-md5-checksums
Match MD5 checksums and replace as single values. This option causes the tool to be more careful about
matching numbers so that MD5 checksums like fbc5e685a5d3d45aa1d0347fdb7c4d35 are matched
and replaced as a single ? placeholder. Otherwise, the default number matching pattern will replace
fbc5e685a5d3d45aa1d0347fdb7c4d35 as fbc?.
-query
type: string
The query to convert into a fingerprint.
-version

2.9.6 ENVIRONMENT


66 Chapter 2. Tools


PTDEBUG=1 pt-fingerprint ... > FILE 2>&1



Perl.

2.9.8 BUGS

For a list of known bugs, see http://www.percona.com/bugs/pt-ﬁngerprint.
• Tool --version

2.9.9 DOWNLOADING






2.9.10 AUTHORS




2.9. pt-ﬁngerprint 67




2.9.13 VERSION

pt-ﬁngerprint 2.1.1

2.10 pt-fk-error-logger

2.10.1 NAME

pt-fk-error-logger - Extract and log MySQL foreign key errors.

2.10.2 SYNOPSIS

Usage

pt-fk-error-logger [OPTION...] SOURCE_DSN

pt-fk-error-logger extracts and saves information about the most recent foreign key errors in a MySQL server.
Print foreign key errors on host1:
pt-fk-error-logger h=host1

Save foreign key errors on host1 to db.foreign_key_errors table on host2:
pt-fk-error-logger h=host1 --dest h=host1,D=db,t=foreign_key_errors

2.10.3 RISKS

pt-fk-error-logger is read-only unless you specify --dest. It should be very low-risk.
will be marked as such. You can see a list of such issues at the following URL: http://www.percona.com/bugs/pt-fk-
error-logger.

68 Chapter 2. Tools



2.10.4 DESCRIPTION

pt-fk-error-logger prints or saves the foreign key errors text from SHOW INNODB STATUS. The errors are not
parsed or interpreted in any way. Foreign key errors are uniquely identiﬁed by their timestamp. Only new (more
recent) errors are printed or saved.

2.10.5 OUTPUT

If --print is given or no --dest is given, then pt-fk-error-logger prints the foreign key error text to STDOUT
exactly as it appeared in SHOW INNODB STATUS.

2.10.6 OPTIONS

-ask-pass
-charset
-config
type: Array
-daemonize
-defaults-file
-dest
type: DSN
DSN for where to store foreign key errors; specify at least a database (D) and table (t).
Missing values are ﬁlled in with the same values from the source host, so you can usually omit most parts of
this argument if you’re storing foreign key errors on the same server on which they happen.
The following table is suggested:
CREATE TABLE foreign_key_errors (
ts datetime NOT NULL,
error text NOT NULL,
PRIMARY KEY (ts),
)

The only information saved is the timestamp and the foreign key error text.

2.10. pt-fk-error-logger 69


-help
Show help and exit.
-host
Connect to host.
-interval
type: time; default: 0
How often to check for foreign key errors.
-log
type: string
-password
-pid
type: string
-port
-print
Print results on standard output. See “OUTPUT” for more.
-run-time
type: time
How long to run before exiting.
-set-vars
executed.
-socket
-user
-version

70 Chapter 2. Tools


2.10.7 DSN OPTIONS

details.
• A
• D
Default database.
• F
• h
Connect to host.
• p
• P
• S
• t
Table in which to store foreign key errors.
• u

2.10.8 ENVIRONMENT

PTDEBUG=1 pt-fk-error-logger ... > FILE 2>&1


2.10. pt-fk-error-logger 71



Perl.

2.10.10 BUGS

For a list of known bugs, see http://www.percona.com/bugs/pt-fk-error-logger.
• Tool --version

2.10.11 DOWNLOADING






2.10.12 AUTHORS

Daniel Nichter



72 Chapter 2. Tools




2.10.15 VERSION

pt-fk-error-logger 2.1.1

2.11 pt-heartbeat

2.11.1 NAME

pt-heartbeat - Monitor MySQL replication delay.

2.11.2 SYNOPSIS

Usage

pt-heartbeat [OPTION...] [DSN] --update|--monitor|--check|--stop

pt-heartbeat measures replication lag on a MySQL or PostgreSQL server. You can use it to update a master or monitor
a replica. If possible, MySQL connection options are read from your .my.cnf ﬁle.
Start daemonized process to update test.heartbeat table on master:
pt-heartbeat -D test --update -h master-server --daemonize

Monitor replication lag on slave:
pt-heartbeat -D test --monitor -h slave-server

pt-heartbeat -D test --monitor -h slave-server --dbi-driver Pg

Check slave lag once and exit (using optional DSN to specify slave host):
pt-heartbeat -D test --check h=slave-server

2.11.3 RISKS


2.11. pt-heartbeat 73


pt-heartbeat merely reads and writes a single record in a table. It should be very low-risk.
heartbeat.

2.11.4 DESCRIPTION

pt-heartbeat is a two-part MySQL and PostgreSQL replication delay monitoring system that measures delay by
looking at actual replicated data. This avoids reliance on the replication mechanism itself, which is unreliable. (For
example, SHOW SLAVE STATUS on MySQL).
The first part is an --update instance of pt-heartbeat that connects to a master and updates a timestamp (“heartbeat
record”) every --interval seconds. Since the heartbeat table may contain records from multiple masters (see
“MULTI-SLAVE HIERARCHY”), the server’s ID (@@server_id) is used to identify records.
The second part is a --monitor or --check instance of pt-heartbeat that connects to a slave, examines the
replicated heartbeat record from its immediate master or the specified --master-server-id, and computes the
difference from the current system time. If replication between the slave and the master is delayed or broken, the
computed difference will be greater than zero and potentially increase if --monitor is specified.
You must either manually create the heartbeat table on the master or use --create-table. See
--create-table for the proper heartbeat table structure. The MEMORY storage engine is suggested, but not re-
quired of course, for MySQL.
The heartbeat table must contain a heartbeat row. By default, a heartbeat row is inserted if it doesn’t exist. This
feature can be disabled with the --[no]insert-heartbeat-row option in case the database user does not have
INSERT privileges.
pt-heartbeat depends only on the heartbeat record being replicated to the slave, so it works regardless of the replication
mechanism (built-in replication, a system such as Continuent Tungsten, etc). It works at any depth in the replication
hierarchy; for example, it will reliably report how far a slave lags its master’s master’s master. And if replication is
stopped, it will continue to work and report (accurately!) that the slave is falling further and further behind the master.
pt-heartbeat has a maximum resolution of 0.01 second. The clocks on the master and slave servers must be closely
synchronized via NTP. By default, --update checks happen on the edge of the second (e.g. 00:01) and --monitor
checks happen halfway between seconds (e.g. 00:01.5). As long as the servers’ clocks are closely synchronized and
replication events are propagating in less than half a second, pt-heartbeat will report zero seconds of delay.
pt-heartbeat will try to reconnect if the connection has an error, but will not retry if it can’t get a connection when it
first starts.
The --dbi-driver option lets you use pt-heartbeat to monitor PostgreSQL as well. It is reported to work well
with Slony-1 replication.

2.11.5 MULTI-SLAVE HIERARCHY

If the replication hierarchy has multiple slaves which are masters of other slaves, like “master -> slave1 -
> slave2”, --update instances can be ran on the slaves as well as the master. The default heartbeat ta-
ble (see --create-table) is keyed on the server_id column, so each server will update the row where
server_id=@@server_id.

74 Chapter 2. Tools


For --monitor and --check, if --master-server-id is not speciﬁed, the tool tries to discover and use the
slave’s immediate master. If this fails, or if you want monitor lag from another master, then you can specify the
--master-server-id to use.
For example, if the replication hierarchy is “master -> slave1 -> slave2” with corresponding server IDs 1, 2 and 3, you
can:
pt-heartbeat --daemonize -D test --update -h master
pt-heartbeat --daemonize -D test --update -h slave1

Then check (or monitor) the replication delay from master to slave2:
pt-heartbeat -D test --master-server-id 1 --check slave2

Or check the replication delay from slave1 to slave2:
pt-heartbeat -D test --master-server-id 2 --check slave2

Stopping the --update instance one slave1 will not affect the instance on master.

2.11.6 MASTER AND SLAVE STATUS

The default heartbeat table (see --create-table) has columns for saving information from SHOW MASTER
STATUS and SHOW SLAVE STATUS. These columns are optional. If any are present, their corresponding infor-
mation will be saved.

2.11.7 OPTIONS

Specify at least one of --stop, --update, --monitor, or --check.
--update, --monitor, and --check are mutually exclusive.
--daemonize and --check are mutually exclusive.
-ask-pass
-charset
-check
Check slave delay once and exit. If you also specify --recurse, the tool will try to discover slave’s of the
given slave and check and print their lag, too. The hostname or IP and port for each slave is printed before its
delay. --recurse only works with MySQL.
-config
type: Array
-create-table
Create the heartbeat --table if it does not exist.



This option causes the table specified by --database and --table to be created with the following
MAGIC_create_heartbeat table definition:
CREATE TABLE heartbeat (
ts varchar(26) NOT NULL,
server_id int unsigned NOT NULL PRIMARY KEY,
file varchar(255) DEFAULT NULL, -- SHOW MASTER STATUS
position bigint unsigned DEFAULT NULL, -- SHOW MASTER STATUS
relay_master_log_file varchar(255) DEFAULT NULL, -- SHOW SLAVE STATUS
exec_master_log_pos bigint unsigned DEFAULT NULL -- SHOW SLAVE STATUS
);

The heartbeat table requires at least one row. If you manually create the heartbeat table, then you must insert a
row by doing:
INSERT INTO heartbeat (ts, server_id) VALUES (NOW(), N);

where N is the server’s ID; do not use @@server_id because it will replicate and slaves will insert their own
server ID instead of the master’s server ID.
This is done automatically by --create-table.
A legacy version of the heartbeat table is still supported:
CREATE TABLE heartbeat (
id int NOT NULL PRIMARY KEY,
ts datetime NOT NULL
);

Legacy tables do not support --update instances on each slave of a multi-slave hierarchy like “master ->
slave1 -> slave2”. To manually insert the one required row into a legacy table:
INSERT INTO heartbeat (id, ts) VALUES (1, NOW());

The tool automatically detects if the heartbeat table is legacy.
See also “MULTI-SLAVE HIERARCHY”.
-daemonize
-database
short form: -D; type: string
The database to use for the connection.
-dbi-driver
default: mysql; type: string
Specify a driver for the connection; mysql and Pg are supported.
-defaults-file
-file
type: string
Print latest --monitor output to this file.
When --monitor is given, prints output to the specified file instead of to STDOUT. The file is opened, trun-
cated, and closed every interval, so it will only contain the most recent statistics. Useful when --daemonize
is given.

76 Chapter 2. Tools


-frames
type: string; default: 1m,5m,15m
Timeframes for averages.
Specifies the timeframes over which to calculate moving averages when --monitor is given. Specify as a
comma-separated list of numbers with suffixes. The suffix can be s for seconds, m for minutes, h for hours, or d
for days. The size of the largest frame determines the maximum memory usage, as up to the specified number
of per-second samples are kept in memory to calculate the averages. You can specify as many timeframes as
you like.
-help
Show help and exit.
-host
Connect to host.
-[no]insert-heartbeat-row
default: yes
Insert a heartbeat row in the --table if one doesn’t exist.
The heartbeat --table requires a heartbeat row, else there’s nothing to --update, --monitor, or
--check! By default, the tool will insert a heartbeat row if one is not already present. You can disable this
feature by specifying --no-insert-heartbeat-row in case the database user does not have INSERT
privileges.
-interval
type: float; default: 1.0
How often to update or check the heartbeat --table. Updates and checks begin on the first whole second
then repeat every --interval seconds for --update and every --interval plus --skew seconds for
--monitor.
For example, if at 00:00.4 an --update instance is started at 0.5 second intervals, the first update happens at
00:01.0, the next at 00:01.5, etc. If at 00:10.7 a --monitor instance is started at 0.05 second intervals with
the default 0.5 second --skew, then the first check happens at 00:11.5 (00:11.0 + 0.5) which will be --skew
seconds after the last update which, because the instances are checking at synchronized intervals, happened at
00:11.0.
The tool waits for and begins on the first whole second just to make the interval calculations simpler. Therefore,
the tool could wait up to 1 second before updating or checking.
The minimum (fastest) interval is 0.01, and the maximum precision is two decimal places, so 0.015 will be
rounded to 0.02.
If a legacy heartbeat table (see --create-table) is used, then the maximum precision is 1s because the ts
column is type datetime.
-log
type: string
-master-server-id
type: string
Calculate delay from this master server ID for --monitor or --check. If not given, pt-heartbeat attempts
to connect to the server’s master and determine its server id.



-monitor
Monitor slave delay continuously.
Specifies that pt-heartbeat should check the slave’s delay every second and report to STDOUT (or if --file
is given, to the file instead). The output is the current delay followed by moving averages over the timeframe
given in --frames. For example,
5s [ 0.25s, 0.05s, 0.02s ]

-password
-pid
type: string
-port
-print-master-server-id
Print the auto-detected or given --master-server-id. If --check or --monitor is specified, specify-
ing this option will print the auto-detected or given --master-server-id at the end of each line.
-recurse
type: int
Check slaves recursively to this depth in --check mode.
Try to discover slave servers recursively, to the specified depth. After discovering servers, run the check on each
one of them and print the hostname (if possible), followed by the slave delay.
This currently works only with MySQL. See --recursion-method.
-recursion-method
type: string
Preferred recursion method used to find slaves.
Possible methods are:
METHOD USES
=========== ================
processlist SHOW PROCESSLIST
hosts SHOW SLAVE HOSTS

The processlist method is preferred because SHOW SLAVE HOSTS is not reliable. However, the hosts method
is required if the server uses a non-standard port (not 3306). Usually pt-heartbeat does the right thing and finds
the slaves, but you may give a preferred method and it will be used first. If it doesn’t find any slaves, the other
methods will be tried.
-replace
Use REPLACE instead of UPDATE for –update.
When running in --update mode, use REPLACE instead of UPDATE to set the heartbeat table’s timestamp.
The REPLACE statement is a MySQL extension to SQL. This option is useful when you don’t know whether
the table contains any rows or not. It must be used in conjunction with –update.

78 Chapter 2. Tools


-run-time
type: time
Time to run before exiting.
-sentinel
type: string; default: /tmp/pt-heartbeat-sentinel
-set-vars
executed.
-skew
How long to delay checks.
The default is to delay checks one half second. Since the update happens as soon as possible after the beginning
of the second on the master, this allows one half second of replication delay before reporting that the slave lags
the master by one second. If your clocks are not completely accurate or there is some other reason you’d like to
delay the slave more or less, you can tweak this value. Try setting the PTDEBUG environment variable to see
the effect this has.
-socket
-stop
This should have the effect of stopping all running instances which are watching the same sentinel file. If none
of --update, --monitor or --check is specified, pt-heartbeat will exit after creating the file. If one of
these is specified, pt-heartbeat will wait the interval given by --interval, then remove the file and continue
working.
You might find this handy to stop cron jobs gracefully if necessary, or to replace one running instance with
another. For example, if you want to stop and restart pt-heartbeat every hour (just to make sure that it is
restarted every hour, in case of a server crash or some other problem), you could use a crontab line like this:
0 * * * * :program:‘pt-heartbeat‘ --update -D test --stop
--sentinel /tmp/pt-heartbeat-hourly

The non-default --sentinel will make sure the hourly cron job stops only instances previously started with
the same options (that is, from the same cron job).
See also --sentinel.
-table
type: string; default: heartbeat
The table to use for the heartbeat.
Don’t specify database.table; use --database to specify the database.
See --create-table.
-update
Update a master’s heartbeat.



-user
-version

2.11.8 DSN OPTIONS

details.
• A
• D
Default database.
• F
• h
Connect to host.
• p
• P
• S
• u

80 Chapter 2. Tools


2.11.9 ENVIRONMENT

PTDEBUG=1 pt-heartbeat ... > FILE 2>&1



Perl.

2.11.11 BUGS

For a list of known bugs, see http://www.percona.com/bugs/pt-heartbeat.
• Tool --version

2.11.12 DOWNLOADING






2.11.13 AUTHORS

Proven Scaling LLC, SixApart Ltd, Baron Schwartz, and Daniel Nichter






This program is copyright 2006 Proven Scaling LLC and Six Apart Ltd, 2007-2012 Percona Inc. Feedback and
improvements are welcome.
Feedback and improvements are welcome.

2.11.16 VERSION

pt-heartbeat 2.1.1

2.12 pt-index-usage

2.12.1 NAME

pt-index-usage - Read queries from a log and analyze how they use indexes.

2.12.2 SYNOPSIS

Usage

pt-index-usage [OPTION...] [FILE...]

pt-index-usage reads queries from logs and analyzes how they use indexes.
Analyze queries in slow.log and print reports:
pt-index-usage /path/to/slow.log --host localhost

Disable reports and save results to mk database for later analysis:
pt-index-usage slow.log --no-report --save-results-database mk

82 Chapter 2. Tools


2.12.3 RISKS

This tool is read-only unless you use --save-results-database. It reads a log of queries and EXPLAIN them.
It also gathers information about all tables in all databases. It should be very low-risk.
index-usage.

2.12.4 DESCRIPTION

This tool connects to a MySQL database server, reads through a query log, and uses EXPLAIN to ask MySQL how it
will use each query. When it is finished, it prints out a report on indexes that the queries didn’t use.
The query log needs to be in MySQL’s slow query log format. If you need to input a different format, you can use
pt-query-digest to translate the formats. If you don’t specify a filename, the tool reads from STDIN.
The tool runs two stages. In the first stage, the tool takes inventory of all the tables and indexes in your database, so
it can compare the existing indexes to those that were actually used by the queries in the log. In the second stage, it
runs EXPLAIN on each query in the query log. It uses separate database connections to inventory the tables and run
EXPLAIN, so it opens two connections to the database.
If a query is not a SELECT, it tries to transform it to a roughly equivalent SELECT query so it can be EXPLAINed.
This is not a perfect process, but it is good enough to be useful.
The tool skips the EXPLAIN step for queries that are exact duplicates of those seen before. It assumes that the same
query will generate the same EXPLAIN plan as it did previously (usually a safe assumption, and generally good for
performance), and simply increments the count of times that the indexes were used. However, queries that have the
same fingerprint but different checksums will be re-EXPLAINed. Queries that have different literal constants can have
different execution plans, and this is important to measure.
After EXPLAIN-ing the query, it is necessary to try to map aliases in the query back to the original table names. For
example, consider the EXPLAIN plan for the following query:
SELECT * FROM tbl1 AS foo;

The EXPLAIN output will show access to table foo, and that must be translated back to tbl1. This process involves
complex parsing. It is generally very accurate, but there is some chance that it might not work right. If you find cases
where it fails, submit a bug report and a reproducible test case.
Queries that cannot be EXPLAINed will cause all subsequent queries with the same fingerprint to be blacklisted. This
is to reduce the work they cause, and prevent them from continuing to print error messages. However, at least in this
stage of the tool’s development, it is my opinion that it’s not a good idea to preemptively silence these, or prevent them
from being EXPLAINed at all. I am looking for lots of feedback on how to improve things like the query parsing. So
please submit your test cases based on the errors the tool prints!

2.12.5 OUTPUT

After it reads all the events in the log, the tool prints out DROP statements for every index that was not used. It skips
indexes for tables that were never accessed by any queries in the log, to avoid false-positive results.

2.12. pt-index-usage 83


If you don’t specify --quiet, the tool also outputs warnings about statements that cannot be EXPLAINed and
similar. These go to standard error.
Progress reports are enabled by default (see --progress). These also go to standard error.

2.12.6 OPTIONS

-ask-pass
-charset
-config
type: Array
-create-save-results-database
Create the --save-results-database if it does not exist.
If the --save-results-database already exists and this option is speciﬁed, the database is used and the
necessary tables are created if they do not already exist.
-[no]create-views
Create views for --save-results-database example queries.
Several example queries are given for querying the tables in the --save-results-database. These
example queries are, by default, created as views. Specifying --no-create-views prevents these views
from being created.
-database
-databases
Only get tables and indexes from this comma-separated list of databases.
-databases-regex
type: string
Only get tables and indexes from database whose names match this Perl regex.
-defaults-file
-drop
type: Hash; default: non-unique
Suggest dropping only these types of unused indexes.
By default pt-index-usage will only suggest to drop unused secondary indexes, not primary or unique indexes.
You can specify which types of unused indexes the tool suggests to drop: primary, unique, non-unique, all.

84 Chapter 2. Tools


A separate ALTER TABLE statement for each type is printed. So if you specify --drop all and there is
a primary key and a non-unique index, the ALTER TABLE ... DROP for each will be printed on separate
lines.
-empty-save-results-tables
Drop and re-create all pre-existing tables in the --save-results-database. This allows information
from previous runs to be removed before the current run.
-help
Show help and exit.
-host
Connect to host.
-ignore-databases
type: Hash
-ignore-databases-regex
type: string
Ignore databases whose names match this Perl regex.
-ignore-tables
type: Hash
Ignore this comma-separated list of table names.
-ignore-tables-regex
type: string
Ignore tables whose names match the Perl regex.
-password
-port
-progress
type: array; default: time,30
Print progress reports to STDERR. The value is a comma-separated list with two parts. The ﬁrst part can be
percentage, time, or iterations; the second part speciﬁes how often an update should be printed, in percentage,
seconds, or number of iterations.
-quiet
short form: -q
Do not print any warnings. Also disables --progress.
-[no]report
default: yes
Print the reports for --report-format.



You may want to disable the reports by specifying --no-report if, for example, you also specify
--save-results-database and you only want to query the results tables later.
-report-format
type: Array; default: drop_unused_indexes
Right now there is only one report: drop_unused_indexes. This report prints SQL statements for dropping any
unused indexes. See also --drop.
See also --[no]report.
-save-results-database
type: DSN
Save results to tables in this database. Information about indexes, queries, tables and their usage is stored in
several tables in the speciﬁed database. The tables are auto-created if they do not exist. If the database doesn’t
exist, it can be auto-created with --create-save-results-database. In this case the connection is
initially created with no default database, then after the database is created, it is USE’ed.
pt-index-usage executes INSERT statements to save the results. Therefore, you should be careful if you use
this feature on a production server. It might increase load, or cause trouble if you don’t want the server to be
written to, or so on.
This is a new feature. It may change in future releases.
After a run, you can query the usage tables to answer various questions about index usage. The tables have the
following CREATE TABLE deﬁnitions:
MAGIC_create_indexes:
CREATE TABLE IF NOT EXISTS indexes (
db VARCHAR(64) NOT NULL,
tbl VARCHAR(64) NOT NULL,
idx VARCHAR(64) NOT NULL,
cnt BIGINT UNSIGNED NOT NULL DEFAULT 0,
PRIMARY KEY (db, tbl, idx)
)

MAGIC_create_queries:
CREATE TABLE IF NOT EXISTS queries (
query_id BIGINT UNSIGNED NOT NULL,
fingerprint TEXT NOT NULL,
sample TEXT NOT NULL,
PRIMARY KEY (query_id)
)

MAGIC_create_tables:
CREATE TABLE IF NOT EXISTS tables (
PRIMARY KEY (db, tbl)
)

MAGIC_create_index_usage:
CREATE TABLE IF NOT EXISTS index_usage (
query_id BIGINT UNSIGNED NOT NULL,

86 Chapter 2. Tools


idx VARCHAR(64) NOT NULL,
UNIQUE INDEX (query_id, db, tbl, idx)
)

MAGIC_create_index_alternatives:
CREATE TABLE IF NOT EXISTS index_alternatives (
query_id BIGINT UNSIGNED NOT NULL, -- This query used
db VARCHAR(64) NOT NULL, -- this index, but...
tbl VARCHAR(64) NOT NULL, --
idx VARCHAR(64) NOT NULL, --
alt_idx VARCHAR(64) NOT NULL, -- was an alternative
UNIQUE INDEX (query_id, db, tbl, idx, alt_idx),
INDEX (db, tbl, idx),
INDEX (db, tbl, alt_idx)
)

The following are some queries you can run against these tables to answer common questions you might have.
Each query is also created as a view (with MySQL v5.0 and newer) if :option:‘--[no]create-views‘
is true (it is by default). The view names are the strings after the MAGIC_view_ preﬁx.
Question: which queries sometimes use different indexes, and what fraction of the time is each index chosen?
MAGIC_view_query_uses_several_indexes:
SELECT iu.query_id, CONCAT_WS(’.’, iu.db, iu.tbl, iu.idx) AS idx,
variations, iu.cnt, iu.cnt / total_cnt * 100 AS pct
FROM index_usage AS iu
INNER JOIN (
SELECT query_id, db, tbl, SUM(cnt) AS total_cnt,
COUNT(*) AS variations
FROM index_usage
GROUP BY query_id, db, tbl
HAVING COUNT(*) > 1
) AS qv USING(query_id, db, tbl);

Question: which indexes have lots of alternatives, i.e. are chosen instead of other indexes, and for what queries?
MAGIC_view_index_has_alternates:
SELECT CONCAT_WS(’.’, db, tbl, idx) AS idx_chosen,
GROUP_CONCAT(DISTINCT alt_idx) AS alternatives,
GROUP_CONCAT(DISTINCT query_id) AS queries, SUM(cnt) AS cnt
FROM index_alternatives
GROUP BY db, tbl, idx
HAVING COUNT(*) > 1;

Question: which indexes are considered as alternates for other indexes, and for what queries?
MAGIC_view_index_alternates:
SELECT CONCAT_WS(’.’, db, tbl, alt_idx) AS idx_considered,
GROUP_CONCAT(DISTINCT idx) AS alternative_to,
GROUP BY db, tbl, alt_idx
HAVING COUNT(*) > 1;

Question: which of those are never chosen by any queries, and are therefore superﬂuous?
MAGIC_view_unused_index_alternates:



SELECT CONCAT_WS(’.’, i.db, i.tbl, i.idx) AS idx,
alt.alternative_to, alt.queries, alt.cnt
FROM indexes AS i
INNER JOIN (
SELECT db, tbl, alt_idx, GROUP_CONCAT(DISTINCT idx) AS alternative_to,
GROUP BY db, tbl, alt_idx
HAVING COUNT(*) > 1
) AS alt ON i.db = alt.db AND i.tbl = alt.tbl
AND i.idx = alt.alt_idx
WHERE i.cnt = 0;

Question: given a table, which indexes were used, by how many queries, with how many distinct ﬁngerprints?
Were there alternatives? Which indexes were not used? You can edit the following query’s SELECT list to also
see the query IDs in question. MAGIC_view_index_usage:
SELECT i.idx, iu.usage_cnt, iu.usage_total,
ia.alt_cnt, ia.alt_total
FROM indexes AS i
LEFT OUTER JOIN (
SELECT db, tbl, idx, COUNT(*) AS usage_cnt,
SUM(cnt) AS usage_total, GROUP_CONCAT(query_id) AS used_by
FROM index_usage
) AS iu ON i.db=iu.db AND i.tbl=iu.tbl AND i.idx = iu.idx
LEFT OUTER JOIN (
SELECT db, tbl, idx, COUNT(*) AS alt_cnt,
SUM(cnt) AS alt_total,
GROUP_CONCAT(query_id) AS alt_queries
) AS ia ON i.db=ia.db AND i.tbl=ia.tbl AND i.idx = ia.idx;

Question: which indexes on a given table are vital for at least one query (there is no alternative)?
MAGIC_view_required_indexes:
SELECT i.db, i.tbl, i.idx, no_alt.queries
FROM indexes AS i
INNER JOIN (
SELECT iu.db, iu.tbl, iu.idx,
GROUP_CONCAT(iu.query_id) AS queries
FROM index_usage AS iu
LEFT OUTER JOIN index_alternatives AS ia
USING(db, tbl, idx)
WHERE ia.db IS NULL
GROUP BY iu.db, iu.tbl, iu.idx
) AS no_alt ON no_alt.db = i.db AND no_alt.tbl = i.tbl
AND no_alt.idx = i.idx
ORDER BY i.db, i.tbl, i.idx, no_alt.queries;

-set-vars
executed.
-socket

88 Chapter 2. Tools


-tables
Only get indexes from this comma-separated list of tables.
-tables-regex
type: string
Only get indexes from tables whose names match this Perl regex.
-user
-version

2.12.7 DSN OPTIONS

details.
• A
• D
Database to connect to.
• F
• h
Connect to host.
• p
• P
• S



• u

2.12.8 ENVIRONMENT

PTDEBUG=1 pt-index-usage ... > FILE 2>&1



Perl.

2.12.10 BUGS

For a list of known bugs, see http://www.percona.com/bugs/pt-index-usage.
• Tool --version

2.12.11 DOWNLOADING






90 Chapter 2. Tools


2.12.12 AUTHORS





welcome.

2.12.15 VERSION

pt-index-usage 2.1.1

2.13 pt-ioprofile

2.13.1 NAME

pt-ioprofile - Watch process IO and print a table of file and I/O activity.

2.13.2 SYNOPSIS

Usage

pt-ioprofile [OPTIONS] [FILE]

pt-ioprofile does two things: 1) get lsof+strace for -s seconds, 2) aggregate the result. If you specify a FILE, then step
1) is not performed.

2.13. pt-ioprofile 91


2.13.3 RISKS

pt-ioprofile is a read-only tool, so your data is not at risk. However, it works by attaching strace to the process
using ptrace(), which will make it run very slowly until strace detaches. In addition to freezing the server, there
is also some risk of the process crashing or performing badly after strace detaches from it, or indeed of strace
not detaching cleanly and leaving the process in a sleeping state. As a result, this should be considered an intrusive
tool, and should not be used on production servers unless you are comfortable with that.
ioprofile.

2.13.4 DESCRIPTION

pt-ioprofile uses strace and lsof to watch a process’s IO and print out a table of files and I/O activity. By default,
it watches the mysqld process for 30 seconds. The output is like:
Tue Dec 27 15:33:57 PST 2011
Tracing process ID 1833
total read write lseek ftruncate filename
0.000150 0.000029 0.000068 0.000038 0.000015 /tmp/ibBE5opS

You probably need to run this tool as root.

2.13.5 OPTIONS

-aggregate
short form: -a; type: string; default: sum
The aggregate function, either sum or avg.
If sum, then each cell will contain the sum of the values in it. If avg, then each cell will contain the average of
the values in it.
-cell
short form: -c; type: string; default: times
The cell contents.
Valid values are:
VALUE CELLS CONTAIN
===== =======================
count Count of I/O operations
sizes Sizes of I/O operations
times I/O operation timing

-group-by
short form: -g; type: string; default: filename
The group-by item.

92 Chapter 2. Tools


Valid values are:
VALUE GROUPING
===== ======================================
all Summarize into a single line of output
filename One line of output per filename
pid One line of output per process ID

-help
Print help and exit.
-profile-pid
short form: -p; type: int
The PID to profile, overrides --profile-process.
-profile-process
short form: -b; type: string; default: mysqld
The process name to profile.
-run-time
How long to profile.
-save-samples
type: string
Filename to save samples in; these can be used for later analysis.
-version
Print the tool’s version and exit.

2.13.6 ENVIRONMENT



This tool requires the Bourne shell (/bin/sh).

2.13.8 BUGS

For a list of known bugs, see http://www.percona.com/bugs/pt-ioprofile.
• Tool --version

2.13. pt-ioprofile 93


2.13.9 DOWNLOADING






2.13.10 AUTHORS

Baron Schwartz




welcome.

2.13.13 VERSION

pt-ioproﬁle 2.1.1

94 Chapter 2. Tools


2.14 pt-kill

2.14.1 NAME

pt-kill - Kill MySQL queries that match certain criteria.

2.14.2 SYNOPSIS

Usage

pt-kill [OPTIONS]

pt-kill kills MySQL connections. pt-kill connects to MySQL and gets queries from SHOW PROCESSLIST if no
FILE is given. Else, it reads queries from one or more FILE which contains the output of SHOW PROCESSLIST. If
FILE is -, pt-kill reads from STDIN.
Kill queries running longer than 60s:
pt-kill --busy-time 60 --kill

Print, do not kill, queries running longer than 60s:
pt-kill --busy-time 60 --print

Check for sleeping processes and kill them all every 10s:
pt-kill --match-command Sleep --kill --victims all --interval 10

Print all login processes:
pt-kill --match-state login --print --victims all

See which queries in the processlist right now would match:
mysql -e "SHOW PROCESSLIST" > proclist.txt
pt-kill --test-matching proclist.txt --busy-time 60 --print

2.14.3 RISKS

pt-kill kills queries if you use the --kill option, so it can disrupt your database’s users, of course. You should test
with the <--print> option, which is safe, if you’re unsure what the tool will do.
will be marked as such. You can see a list of such issues at the following URL: http://www.percona.com/bugs/pt-kill.

2.14. pt-kill 95


2.14.4 DESCRIPTION

pt-kill captures queries from SHOW PROCESSLIST, filters them, and then either kills or prints them. This is also
known as a “slow query sniper” in some circles. The idea is to watch for queries that might be consuming too many
resources, and kill them.
For brevity, we talk about killing queries, but they may just be printed (or some other future action) depending on what
options are given.
Normally pt-kill connects to MySQL to get queries from SHOW PROCESSLIST. Alternatively, it can read SHOW
PROCESSLIST output from files. In this case, pt-kill does not connect to MySQL and --kill has no effect. You
should use --print instead when reading files. The ability to read a file with --test-matching allows you to
capture SHOW PROCESSLIST and test it later with pt-kill to make sure that your matches kill the proper queries.
There are a lot of special rules to follow, such as “don’t kill replication threads,” so be careful not to kill something
important!
Two important options to know are --busy-time and --victims. First, whereas most match/filter options match
their corresponding value from SHOW PROCESSLIST (e.g. --match-command matches a query’s Command
value), the Time value is matched by --busy-time. See also --interval.
Second, --victims controls which matching queries from each class are killed. By default, the matching query
with the highest Time value is killed (the oldest query). See the next section, “GROUP, MATCH AND KILL”, for
more details.
Usually you need to specify at least one --match option, else no queries will match. Or, you can specify
--match-all to match all queries that aren’t ignored by an --ignore option.

2.14.5 GROUP, MATCH AND KILL

Queries pass through several steps to determine which exactly will be killed (or printed–whatever action is specified).
Understanding these steps will help you match precisely the queries you want.
The first step is grouping queries into classes. The --group-by option controls grouping. By default, this option
has no value so all queries are grouped into one default class. All types of matching and filtering (the next step) are
applied per-class. Therefore, you may need to group queries in order to match/filter some classes but not others.
The second step is matching. Matching implies filtering since if a query doesn’t match some criteria, it is removed
from its class. Matching happens for each class. First, queries are filtered from their class by the various Query
Matches options like --match-user. Then, entire classes are filtered by the various Class Matches options
like --query-count.
The third step is victim selection, that is, which matching queries in each class to kill. This is controlled by the
--victims option. Although many queries in a class may match, you may only want to kill the oldest query, or all
queries, etc.
The forth and final step is to take some action on all matching queries from all classes. The Actions options specify
which actions will be taken. At this step, there are no more classes, just a single list of queries to kill, print, etc.

2.14.6 OUTPUT

If only --kill is given, then there is no output. If only --print is given, then a timestamped KILL statement if
printed for every query that would have been killed, like:
# 2009-07-15T15:04:01 KILL 8 (Query 42 sec) SELECT * FROM huge_table

The line shows a timestamp, the query’s Id (8), its Time (42 sec) and its Info (usually the query SQL).

96 Chapter 2. Tools


If both --kill and --print are given, then matching queries are killed and a line for each like the one above is
printed.
Any command executed by --execute-command is responsible for its own output and logging. After being
executed, pt-kill has no control or interaction with the command.

2.14.7 OPTIONS

Specify at least one of --kill, --kill-query, --print, --execute-command or --stop.
--any-busy-time and --each-busy-time are mutually exclusive.
--kill and --kill-query are mutually exclusive.
--daemonize and --test-matching are mutually exclusive.
-ask-pass
-charset
-config
type: Array
-daemonize
-defaults-file
-filter
type: string
Discard events for which this Perl code doesn’t return true.
This option is a string of Perl code or a file containing Perl code that gets compiled into a subroutine with one
argument: $event. This is a hashref. If the given value is a readable file, then pt-kill reads the entire file and
uses its contents as the code. The file should not contain a shebang (#!/usr/bin/perl) line.
If the code returns true, the chain of callbacks continues; otherwise it ends. The code is the last statement in the
subroutine other than return $event. The subroutine template is:
sub { $event = shift; filter && return $event; }

Filters given on the command line are wrapped inside parentheses like like ( filter ). For complex, multi-
line filters, you must put the code inside a file so it will not be wrapped inside parentheses. Either way, the filter
must produce syntactically valid code given the template. For example, an if-else branch given on the command
line would not be valid:
--filter ’if () { } else { }’ # WRONG

2.14. pt-kill 97


Since it’s given on the command line, the if-else branch would be wrapped inside parentheses which is not
syntactically valid. So to accomplish something more complex like this would require putting the code in a file,
for example filter.txt:
my $event_ok; if (...) { $event_ok=1; } else { $event_ok=0; } $event_ok

Then specify --filter filter.txt to read the code from filter.txt.
If the filter code won’t compile, pt-kill will die with an error. If the filter code does compile, an error may still
occur at runtime if the code tries to do something wrong (like pattern match an undefined value). pt-kill does
not provide any safeguards so code carefully!
It is permissible for the code to have side effects (to alter $event).
-group-by
type: string
Apply matches to each class of queries grouped by this SHOW PROCESSLIST column. In addition to
the basic columns of SHOW PROCESSLIST (user, host, command, state, etc.), queries can be matched by
fingerprint which abstracts the SQL query in the Info column.
By default, queries are not grouped, so matches and actions apply to all queries. Grouping allows matches and
actions to apply to classes of similar queries, if any queries in the class match.
For example, detecting cache stampedes (see all-but-oldest under --victims for an explanation of
that term) requires that queries are grouped by the arg attribute. This creates classes of identical queries
(stripped of comments). So queries "SELECT c FROM t WHERE id=1" and "SELECT c FROM t
WHERE id=1" are grouped into the same class, but query c<”SELECT c FROM t WHERE id=3”> is not iden-
tical to the first two queries so it is grouped into another class. Then when --victims all-but-oldest is
specified, all but the oldest query in each class is killed for each class of queries that matches the match criteria.
-help
Show help and exit.
-host
short form: -h; type: string; default: localhost
Connect to host.
-interval
type: time
How often to check for queries to kill. If --busy-time is not given, then the default interval is 30 seconds.
Else the default is half as often as --busy-time. If both --interval and --busy-time are given, then
the explicit --interval value is used.
See also --run-time.
-log
type: string
-password
-pid
type: string

98 Chapter 2. Tools


-port
-run-time
type: time
How long to run before exiting. By default pt-kill runs forever, or until its process is killed or stopped by the
creation of a --sentinel file. If this option is specified, pt-kill runs for the specified amount of time and
sleeps --interval seconds between each check of the PROCESSLIST.
-sentinel
type: string; default: /tmp/pt-kill-sentinel
The presence of the file specified by --sentinel will cause all running instances of pt-kill to exit. You might
find this handy to stop cron jobs gracefully if necessary. See also --stop.
-set-vars
executed.
-socket
-stop
Stop running instances by creating the --sentinel file.
Causes pt-kill to create the sentinel file specified by --sentinel and exit. This should have the effect of
stopping all running instances which are watching the same sentinel file.
-[no]strip-comments
default: yes
Remove SQL comments from queries in the Info column of the PROCESSLIST.
-user
-version
-victims
type: string; default: oldest
Which of the matching queries in each class will be killed. After classes have been matched/filtered, this option
specifies which of the matching queries in each class will be killed (or printed, etc.). The following values are
possible:
oldest
Only kill the single oldest query. This is to prevent killing queries that aren’t really long-running,
they’re just long-waiting. This sorts matching queries by Time and kills the one with the highest
Time value.
all

2.14. pt-kill 99


Kill all queries in the class.
all-but-oldest
Kill all but the oldest query. This is the inverse of the oldest value.
This value can be used to prevent “cache stampedes”, the condition where several identical queries
are executed and create a backlog while the first query attempts to finish. Since all queries are
identical, all but the first query are killed so that it can complete and populate the cache.
-wait-after-kill
type: time
Wait after killing a query, before looking for more to kill. The purpose of this is to give blocked queries a chance
to execute, so we don’t kill a query that’s blocking a bunch of others, and then kill the others immediately
afterwards.
-wait-before-kill
type: time
Wait before killing a query. The purpose of this is to give --execute-command a chance to see the matching
query and gather other MySQL or system information before it’s killed.

2.14.8 QUERY MATCHES

These options filter queries from their classes. If a query does not match, it is removed from its class. The --ignore
options take precedence. The matches for command, db, host, etc. correspond to the columns returned by SHOW
PROCESSLIST: Command, db, Host, etc. All pattern matches are case-sensitive by default, but they can be made
case-insensitive by specifying a regex pattern like (?i-xsm:select).
See also “GROUP, MATCH AND KILL”.
-busy-time
type: time; group: Query Matches
Match queries that have been running for longer than this time. The queries must be in Command=Query status.
This matches a query’s Time value as reported by SHOW PROCESSLIST.
-idle-time
type: time; group: Query Matches
Match queries that have been idle/sleeping for longer than this time. The queries must be in Command=Sleep
status. This matches a query’s Time value as reported by SHOW PROCESSLIST.
-ignore-command
type: string; group: Query Matches
Ignore queries whose Command matches this Perl regex.
See --match-command.
-ignore-db
Ignore queries whose db (database) matches this Perl regex.
See --match-db.
-ignore-host
Ignore queries whose Host matches this Perl regex.
See --match-host.

100 Chapter 2. Tools


-ignore-info
Ignore queries whose Info (query) matches this Perl regex.
See --match-info.
-[no]ignore-self
default: yes; group: Query Matches
Don’t kill pt-kill‘s own connection.
-ignore-state
type: string; group: Query Matches; default: Locked
Ignore queries whose State matches this Perl regex. The default is to keep threads from being killed if they are
locked waiting for another thread.
See --match-state.
-ignore-user
Ignore queries whose user matches this Perl regex.
See --match-user.
-match-all
group: Query Matches
Match all queries that are not ignored. If no ignore options are specified, then every query matches (except
replication threads, unless --replication-threads is also specified). This option allows you to specify
negative matches, i.e. “match every query except...” where the exceptions are defined by specifying various
--ignore options.
This option is not the same as --victims all. This option matches all queries within a class, whereas
--victims all specifies that all matching queries in a class (however they matched) will be killed. Normally,
however, the two are used together because if, for example, you specify --victims oldest, then although
all queries may match, only the oldest will be killed.
-match-command
Match only queries whose Command matches this Perl regex.
Common Command values are:
Query
Sleep
Binlog Dump
Connect
Delayed insert
Execute
Fetch
Init DB
Kill
Prepare
Processlist
Quit
Reset stmt
Table Dump

See http://dev.mysql.com/doc/refman/5.1/en/thread-commands.html for a full list and description of Command
values.

2.14. pt-kill 101


-match-db
Match only queries whose db (database) matches this Perl regex.
-match-host
Match only queries whose Host matches this Perl regex.
The Host value often time includes the port like “host:port”.
-match-info
Match only queries whose Info (query) matches this Perl regex.
The Info column of the processlist shows the query that is being executed or NULL if no query is being executed.
-match-state
Match only queries whose State matches this Perl regex.
Common State values are:
Locked
login
copy to tmp table
Copying to tmp table
Copying to tmp table on disk
Creating tmp table
executing
Reading from net
Sending data
Sorting for order
Sorting result
Table lock
Updating

See http://dev.mysql.com/doc/refman/5.1/en/general-thread-states.html for a full list and description of State
values.
-match-user
Match only queries whose User matches this Perl regex.
-replication-threads
group: Query Matches
Allow matching and killing replication threads.
By default, matches do not apply to replication threads; i.e. replication threads are completely ignored. Speci-
fying this option allows matches to match (and potentially kill) replication threads on masters and slaves.
-test-matching
type: array; group: Query Matches
Files with processlist snapshots to test matching options against. Since the matching options can be complex,
you can save snapshots of processlist in ﬁles, then test matching options against queries in those ﬁles.
This option disables --run-time, --interval, and --[no]ignore-self.



2.14.9 CLASS MATCHES

These matches apply to entire query classes. Classes are created by specifying the --group-by option, else all
queries are members of a single, default class.
-any-busy-time
type: time; group: Class Matches
Match query class if any query has been running for longer than this time. “Longer than” means that if you
specify 10, for example, the class will only match if there’s at least one query that has been running for greater
than 10 seconds.
See --each-busy-time for more details.
-each-busy-time
type: time; group: Class Matches
Match query class if each query has been running for longer than this time. “Longer than” means that if you
specify 10, for example, the class will only match if each and every query has been running for greater than 10
seconds.
See also --any-busy-time (to match a class if ANY query has been running longer than the speciﬁed time)
and --busy-time.
-query-count
type: int; group: Class Matches
Match query class if it has at least this many queries. When queries are grouped into classes by specify-
ing --group-by, this option causes matches to apply only to classes with at least this many queries. If
--group-by is not speciﬁed then this option causes matches to apply only if there are at least this many
queries in the entire SHOW PROCESSLIST.
-verbose
short form: -v
Print information to STDOUT about what is being done.

2.14.10 ACTIONS

These actions are taken for every matching query from all classes. The actions are taken in this order: --print,
--execute-command, --kill”/”--kill-query. This order allows --execute-command to see the out-
put of --print and the query before --kill”/”--kill-query. This may be helpful because pt-kill does not
pass any information to --execute-command.
-execute-command
Execute this command when a query matches.
After the command is executed, pt-kill has no control over it, so the command is responsible for its own info
gathering, logging, interval, etc. The command is executed each time a query matches, so be careful that the
command behaves well when multiple instances are ran. No information from pt-kill is passed to the command.
See also --wait-before-kill.
-kill
group: Actions

2.14. pt-kill 103


Kill the connection for matching queries.
This option makes pt-kill kill the connections (a.k.a. processes, threads) that have matching queries. Use
--kill-query if you only want to kill individual queries and not their connections.
Unless --print is also given, no other information is printed that shows that pt-kill matched and killed a
query.
See also --wait-before-kill and --wait-after-kill.
-kill-query
group: Actions
Kill matching queries.
This option makes pt-kill kill matching queries. This requires MySQL 5.0 or newer. Unlike --kill which
kills the connection for matching queries, this option only kills the query, not its connection.
-print
group: Actions
Print a KILL statement for matching queries; does not actually kill queries.
If you just want to see which queries match and would be killed without actually killing them, specify --print.
To both kill and print matching queries, specify both --kill and --print.

2.14.11 DSN OPTIONS

details.
• A
• D
Default database.
• F
• h
Connect to host.
• p
• P



• S
• u

2.14.12 ENVIRONMENT

PTDEBUG=1 pt-kill ... > FILE 2>&1



Perl.

2.14.14 BUGS

For a list of known bugs, see http://www.percona.com/bugs/pt-kill.
• Tool --version

2.14.15 DOWNLOADING




2.14. pt-kill 105




2.14.16 AUTHORS





welcome.

2.14.19 VERSION

pt-kill 2.1.1

2.15 pt-log-player

2.15.1 NAME

pt-log-player - Replay MySQL query logs.

2.15.2 SYNOPSIS

Usage

pt-log-player [OPTION...] [DSN]



pt-log-player splits and plays slow log files.
Split slow.log on Thread_id into 16 session files, save in ./sessions:
pt-log-player --split Thread_id --session-files 16 --base-dir ./sessions slow.log

Play all those sessions on host1, save results in ./results:
pt-log-player --play ./sessions --base-dir ./results h=host1

Use pt-query-digest to summarize the results:
pt-query-digest ./results/*

2.15.3 RISKS

This tool is meant to load a server as much as possible, for stress-testing purposes. It is not designed to be used on
production servers.
At the time of this release there is a bug which causes pt-log-player to exceed max open files during --split.
will be marked as such. You can see a list of such issues at the following URL: http://www.percona.com/bugs/pt-log-
player.

2.15.4 DESCRIPTION

pt-log-player does two things: it splits MySQL query logs into session files and it plays (executes) queries in session
files on a MySQL server. Only session files can be played; slow logs cannot be played directly without being split.
A session is a group of queries from the slow log that all share a common attribute, usually Thread_id. The common
attribute is specified with --split. Multiple sessions are saved into a single session file. See --session-files,
--max-sessions, --base-file-name and --base-dir. These session files are played with --play.
pt-log-player will --play session files in parallel using N number of --threads. (They’re not technically threads,
but we call them that anyway.) Each thread will play all the sessions in its given session files. The sessions are played
as fast as possible (there are no delays) because the goal is to stress-test and load-test the server. So be careful using
this script on a production server!
Each --play thread writes its results to a separate file. These result files are in slow log format so they can be
aggregated and summarized with pt-query-digest. See “OUTPUT”.

2.15.5 OUTPUT

Both --split and --play have two outputs: status messages printed to STDOUT to let you know what the script
is doing, and session or result files written to separate files saved in --base-dir. You can suppress all output to
STDOUT for each with --quiet, or increase output with --verbose.
The session files written by --split are simple text files containing queries grouped into sessions. For example:

2.15. pt-log-player 107


-- START SESSION 10

use foo

SELECT col FROM foo_tbl

The format of these session files is important: each query must be a single line separated by a single blank line. And
the “– START SESSION” comment tells pt-log-player where individual sessions begin and end so that --play can
correctly fake Thread_id in its result files.
The result files written by --play are in slow log format with a minimal header: the only attributes printed are
Thread_id, Query_time and Schema.

2.15.6 OPTIONS

Specify at least one of --play, --split or --split-random.
--play and --split are mutually exclusive.
-ask-pass
group: Play
-base-dir
type: string; default: ./
Base directory for --split session files and --play result file.
-base-file-name
type: string; default: session
Base file name for --split session files and --play result file.
Each --split session file will be saved as <base-file-name>-N.txt, where N is a four digit, zero-padded
session ID. For example: session-0003.txt.
Each --play result file will be saved as <base-file-name>-results-PID.txt, where PID is the process ID of the
executing thread.
All files are saved in --base-dir.
-charset
short form: -A; type: string; group: Play
-config
type: Array
-defaults-file
Only read mysql options from the given file.
-dry-run
Print which processes play which session files then exit.



-filter
type: string; group: Split
Discard --split events for which this Perl code doesn’t return true.
This option only works with --split.
This option allows you to inject Perl code into the tool to affect how the tool runs. Usually your code should
examine $event to decided whether or not to allow the event. $event is a hashref of attributes and values
of the event being filtered. Or, your code could add new attribute-value pairs to $event for use by other
options that accept event attributes as their value. You can find an explanation of the structure of $event at
http://code.google.com/p/maatkit/wiki/EventAttributes.
There are two ways to supply your code: on the command line or in a file. If you supply your code on the
command line, it is injected into the following subroutine where $filter is your code:
sub {
PTDEBUG && _d(’callback: filter’);
my( $event ) = shift;
( $filter ) && return $event;
}

Therefore you must ensure two things: first, that you correctly escape any special characters that need to be
escaped on the command line for your shell, and two, that your code is syntactically valid when injected into
the subroutine above.
Here’s an example filter supplied on the command line that discards events that are not SELECT statements:
--filter ’$event->{arg} =~ m/^select/i’

The second way to supply your code is in a file. If your code is too complex to be expressed on the command
line that results in valid syntax in the subroutine above, then you need to put the code in a file and give the
file name as the value to --filter. The file should not contain a shebang (#!/usr/bin/perl) line. The
entire contents of the file is injected into the following subroutine:
sub {
PTDEBUG && _d(’callback: filter’);
my( $event ) = shift;
$filter && return $event;
}

That subroutine is almost identical to the one above except your code is not wrapped in parentheses. This allows
you to write multi-line code like:
my $event_ok;
if (...) {
$event_ok = 1;
}
else {
$event_ok = 0;
}
$event_ok

Notice that the last line is not syntactically valid by itself, but it becomes syntactically valid when injected into
the subroutine because it becomes:
$event_ok && return $event;

If your code doesn’t compile, the tool will die with an error. Even if your code compiles, it may crash to
tool during runtime if, for example, it tries a pattern match an undefined value. No safeguards of any kind are
provided so code carefully!



-help
Show help and exit.
-host
short form: -h; type: string; group: Play
Connect to host.
-iterations
type: int; default: 1; group: Play
How many times each thread should play all its session files.
-max-sessions
type: int; default: 5000000; group: Split
Maximum number of sessions to --split.
By default, pt-log-player tries to split every session from the log file. For huge logs, however, this can result in
millions of sessions. This option causes only the first N number of sessions to be saved. All sessions after this
number are ignored, but sessions split before this number will continue to have their queries split even if those
queries appear near the end of the log and after this number has been reached.
-only-select
group: Play
Play only SELECT and USE queries; ignore all others.
-password
short form: -p; type: string; group: Play
-pid
type: string
-play
type: string; group: Play
Play (execute) session files created by --split.
The argument to play must be a comma-separated list of session files created by --split or a directory. If the
argument is a directory, ALL files in that directory will be played.
-port
short form: -P; type: int; group: Play
-print
group: Play
Print queries instead of playing them; requires --play.
You must also specify --play with --print. Although the queries will not be executed, --play is required
to specify which session files to read.



-quiet
short form: -q
Do not print anything; disables --verbose.
-[no]results
default: yes
Print --play results to files in --base-dir.
-session-files
type: int; default: 8; group: Split
Number of session files to create with --split.
The number of session files should either be equal to the number of --threads you intend to --play or be
an even multiple of --threads. This number is important for maximum performance because it:
* allows each thread to have roughly the same amount of sessions to play
* avoids having to open/close many session files
* avoids disk IO overhead by doing large sequential reads

You may want to increase this number beyond --threads if each session file becomes too large. For example,
splitting a 20G log into 8 sessions files may yield roughly eight 2G session files.
See also --max-sessions.
-set-vars
type: string; group: Play; default: wait_timeout=10000
executed.
-socket
short form: -S; type: string; group: Play
-split
Split log by given attribute to create session files.
Valid attributes are any which appear in the log: Thread_id, Schema, etc.
-split-random
group: Split
Split log without an attribute, write queries round-robin to session files.
This option, if specified, overrides --split and causes the log to be split query-by-query, writing each query
to the next session file in round-robin style. If you don’t care about “sessions” and just want to split a lot into N
many session files and the relation or order of the queries does not matter, then use this option.
-threads
type: int; default: 2; group: Play
Number of threads used to play sessions concurrently.
Specifies the number of parallel processes to run. The default is 2. On GNU/Linux machines, the default is the
number of times ‘processor’ appears in /proc/cpuinfo. On Windows, the default is read from the environment.
In any case, the default is at least 2, even when there’s only a single processor.
See also --session-files.



-type
The type of log to --split (default slowlog). The permitted types are
binlog
Split the output of running mysqlbinlog against a binary log file. Currently, splitting binary logs
does not always work well depending on what the binary logs contain. Be sure to check the session
files after splitting to ensure proper “OUTPUT”.
If the binary log contains row-based replication data, you need to run mysqlbinlog with options
--base64-output=decode-rows --verbose, else invalid statements will be written to the
session files.
genlog
Split a general log file.
slowlog
Split a log file in any variation of MySQL slow-log format.
-user
short form: -u; type: string; group: Play
-verbose
short form: -v; cumulative: yes; default: 0
Increase verbosity; can be specified multiple times.
This option is disabled by --quiet.
-version
-[no]warnings
default: no; group: Play
Print warnings about SQL errors such as invalid queries to STDERR.

2.15.7 DSN OPTIONS

details.
• A
• D
Default database.
• F



• h
Connect to host.
• p
• P
• S
• u

2.15.8 ENVIRONMENT

PTDEBUG=1 pt-log-player ... > FILE 2>&1



Perl.

2.15.10 BUGS

For a list of known bugs, see http://www.percona.com/bugs/pt-log-player.
• Tool --version




2.15.11 DOWNLOADING






2.15.12 AUTHORS

Daniel Nichter





2.15.15 VERSION

pt-log-player 2.1.1



2.16 pt-mext

2.16.1 NAME

pt-mext - Look at many samples of MySQL SHOW GLOBAL STATUS side-by-side.

2.16.2 SYNOPSIS

Usage

pt-mext [OPTIONS] -- COMMAND

pt-mext columnizes repeated output from a program like mysqladmin extended.
Get output from mysqladmin:
pt-mext -r -- mysqladmin ext -i10 -c3"

Get output from a file:
pt-mext -r -- cat mysqladmin-output.txt

2.16.3 RISKS

pt-mext is a read-only tool. It should be very low-risk.
will be marked as such. You can see a list of such issues at the following URL: http://www.percona.com/bugs/pt-mext.

2.16.4 DESCRIPTION

pt-mext executes the COMMAND you specify, and reads through the result one line at a time. It places each line into a
temporary file. When it finds a blank line, it assumes that a new sample of SHOW GLOBAL STATUS is starting, and
it creates a new temporary file. At the end of this process, it has a number of temporary files. It joins the temporary
files together side-by-side and prints the result. If the “-r” option is given, it first subtracts each sample from the one
after it before printing results.

2.16.5 OPTIONS

-r Relative: subtract each column from the previous column.

2.16.6 ENVIRONMENT


2.16. pt-mext 115



This tool requires the Bourne shell (/bin/sh) and the seq program.

2.16.8 BUGS

For a list of known bugs, see http://www.percona.com/bugs/pt-mext.
• Tool --version

2.16.9 DOWNLOADING






2.16.10 AUTHORS

Baron Schwartz






This program is copyright 2010 Baron Schwartz, 2011-2012 Percona Inc. Feedback and improvements are welcome.

2.16.13 VERSION

pt-mext 2.1.1

2.17 pt-mysql-summary

2.17.1 NAME

pt-mysql-summary - Summarize MySQL information nicely.

2.17.2 SYNOPSIS

Usage

pt-mysql-summary [OPTIONS] [-- MYSQL OPTIONS]

pt-mysql-summary conveniently summarizes the status and conﬁguration of a MySQL database server so that you
can learn about it at a glance. It is not a tuning tool or diagnosis tool. It produces a report that is easy to diff and can
be pasted into emails without losing the formatting. It should work well on any modern UNIX systems.

2.17.3 RISKS

pt-mysql-summary is a read-only tool. It should be very low-risk.
At the time of this release, we know of no bugs that could harm users.
mysql-summary.

2.17. pt-mysql-summary 117


2.17.4 DESCRIPTION

pt-mysql-summary works by connecting to a MySQL database server and querying it for status and configuration
information. It saves these bits of data into files in a temporary directory, and then formats them neatly with awk and
other scripting languages.
To use, simply execute it. Optionally add a double dash and then the same command-line options you would use to
connect to MySQL, such as the following:
pt-mysql-summary -- --user=root

The tool interacts minimally with the server upon which it runs. It assumes that you’ll run it on the same server you’re
inspecting, and therefore it assumes that it will be able to find the my.cnf configuration file, for example. However, it
should degrade gracefully if this is not the case. Note, however, that its output does not indicate which information
comes from the MySQL database and which comes from the host operating system, so it is possible for confusing
output to be generated if you run the tool on one server and connect to a MySQL database server running on another
server.

2.17.5 OUTPUT

Many of the outputs from this tool are deliberately rounded to show their magnitude but not the exact detail. This is
called fuzzy-rounding. The idea is that it does not matter whether a server is running 918 queries per second or 921
queries per second; such a small variation is insignificant, and only makes the output hard to compare to other servers.
Fuzzy-rounding rounds in larger increments as the input grows. It begins by rounding to the nearest 5, then the nearest
10, nearest 25, and then repeats by a factor of 10 larger (50, 100, 250), and so on, as the input grows.
The following is a sample of the report that the tool produces:
# Percona Toolkit MySQL Summary Report #######################
System time | 2012-03-30 18:46:05 UTC
(local TZ: EDT -0400)
# Instances ##################################################
Port Data Directory Nice OOM Socket
===== ========================== ==== === ======
12345 /tmp/12345/data 0 0 /tmp/12345.sock
12346 /tmp/12346/data 0 0 /tmp/12346.sock
12347 /tmp/12347/data 0 0 /tmp/12347.sock

The first two sections show which server the report was generated on and which MySQL instances are running on the
server. This is detected from the output of ps and does not always detect all instances and parameters, but often works
well. From this point forward, the report will be focused on a single MySQL instance, although several instances may
appear in the above paragraph.
# Report On Port 12345 #######################################
User | msandbox@%
Time | 2012-03-30 14:46:05 (EDT)
Hostname | localhost.localdomain
Version | 5.5.20-log MySQL Community Server (GPL)
Built On | linux2.6 i686
Started | 2012-03-28 23:33 (up 1+15:12:09)
Databases | 4
Datadir | /tmp/12345/data/
Processes | 2 connected, 2 running
Replication | Is not a slave, has 1 slaves connected
Pidfile | /tmp/12345/data/12345.pid (exists)



This section is a quick summary of the MySQL instance: version, uptime, and other very basic parameters. The Time
output is generated from the MySQL server, unlike the system date and time printed earlier, so you can see whether
the database and operating system times match.
# Processlist ################################################

Command COUNT(*) Working SUM(Time) MAX(Time)
------------------------------ -------- ------- --------- ---------
Binlog Dump 1 1 150000 150000
Query 1 1 0 0

User COUNT(*) Working SUM(Time) MAX(Time)
------------------------------ -------- ------- --------- ---------
msandbox 2 2 150000 150000

Host COUNT(*) Working SUM(Time) MAX(Time)
------------------------------ -------- ------- --------- ---------
localhost 2 2 150000 150000

db COUNT(*) Working SUM(Time) MAX(Time)
------------------------------ -------- ------- --------- ---------
NULL 2 2 150000 150000

State COUNT(*) Working SUM(Time) MAX(Time)
------------------------------ -------- ------- --------- ---------
Master has sent all binlog to 1 1 150000 150000
NULL 1 1 0 0

This section is a summary of the output from SHOW PROCESSLIST. Each sub-section is aggregated by a differ-
ent item, which is shown as the first column heading. When summarized by Command, every row in SHOW PRO-
CESSLIST is included, but otherwise, rows whose Command is Sleep are excluded from the SUM and MAX columns,
so they do not skew the numbers too much. In the example shown, the server is idle except for this tool itself, and one
connected replica, which is executing Binlog Dump.
The columns are the number of rows included, the number that are not in Sleep status, the sum of the Time column,
and the maximum Time column. The numbers are fuzzy-rounded.
# Status Counters (Wait 10 Seconds) ##########################
Variable Per day Per second 10 secs
Binlog_cache_disk_use 4
Binlog_cache_use 80
Bytes_received 15000000 175 200
Bytes_sent 15000000 175 2000
Com_admin_commands 1
...................(many lines omitted)............................
Threads_created 40 1
Uptime 90000 1 1

This section shows selected counters from two snapshots of SHOW GLOBAL STATUS, gathered approximately 10
seconds apart and fuzzy-rounded. It includes only items that are incrementing counters; it does not include absolute
numbers such as the Threads_running status variable, which represents a current value, rather than an accumulated
number over time.
The first column is the variable name, and the second column is the counter from the first snapshot divided by 86400
(the number of seconds in a day), so you can see the magnitude of the counter’s change per day. 86400 fuzzy-rounds
to 90000, so the Uptime counter should always be about 90000.
The third column is the value from the first snapshot, divided by Uptime and then fuzzy-rounded, so it represents
approximately how quickly the counter is growing per-second over the uptime of the server.



The third column is the incremental difference from the ﬁrst and second snapshot, divided by the difference in uptime
and then fuzzy-rounded. Therefore, it shows how quickly the counter is growing per second at the time the report was
generated.
# Table cache ################################################
Size | 400
Usage | 15%

This section shows the size of the table cache, followed by the percentage of the table cache in use. The usage is
fuzzy-rounded.
# Key Percona Server features ################################
Table & Index Stats | Not Supported
Multiple I/O Threads | Enabled
Corruption Resilient | Not Supported
Durable Replication | Not Supported
Import InnoDB Tables | Not Supported
Fast Server Restarts | Not Supported
Enhanced Logging | Not Supported
Replica Perf Logging | Not Supported
Response Time Hist. | Not Supported
Smooth Flushing | Not Supported
HandlerSocket NoSQL | Not Supported
Fast Hash UDFs | Unknown

This section shows features that are available in Percona Server and whether they are enabled or not. In the example
shown, the server is standard MySQL, not Percona Server, so the features are generally not supported.
# Plugins ####################################################
InnoDB compression | ACTIVE

This feature shows speciﬁc plugins and whether they are enabled.
# Query cache ################################################
query_cache_type | ON
Size | 0.0
Usage | 0%
HitToInsertRatio | 0%

This section shows whether the query cache is enabled and its size, followed by the percentage of the cache in use and
the hit-to-insert ratio. The latter two are fuzzy-rounded.
# Schema #####################################################
Would you like to mysqldump -d the schema and analyze it? y/n y
There are 4 databases. Would you like to dump all, or just one?
Type the name of the database, or press Enter to dump all of them.

Database Tables Views SPs Trigs Funcs FKs Partn
mysql 24
performance_schema 17
sakila 16 7 3 6 3 22

Database MyISAM CSV PERFORMANCE_SCHEMA InnoDB
mysql 22 2
performance_schema 17
sakila 8 15

Database BTREE FULLTEXT
mysql 31



performance_schema
sakila 63 1

c t s e l d i t
m v s
h i e n o a n i
e a m
a m t u n t t n
d r a
r e m g e y
i c l
s b t i
u h l
t l i n
m a i
a o m t
t r n
m b e e t
p x
t
Database === === === === === === === === === === ===
mysql 61 10 6 78 5 4 26 3 4 5 3
performance_schema 5 16 33
sakila 1 15 1 3 4 3 19 42 26

If you select to dump the schema and analyze it, the tool will print the above section. This summarizes the number
and type of objects in the database. It is generated by running mysqldump --no-data, not by querying the
INFORMATION_SCHEMA, which can freeze a busy server. You can use the --databases option to specify
which databases to examine. If you do not, and you run the tool interactively, it will prompt you as shown.
You can choose not to dump the schema, to dump all of the databases, or to dump only a single named one, by
specifying the appropriate options. In the example above, we are dumping all databases.
The first sub-report in the section is the count of objects by type in each database: tables, views, and so on. The second
one shows how many tables use various storage engines in each database. The third sub-report shows the number of
each type of indexes in each database.
The last section shows the number of columns of various data types in each database. For compact display, the column
headers are formatted vertically, so you need to read downwards from the top. In this example, the first column is
char and the second column is timestamp. This example is truncated so it does not wrap on a terminal.
All of the numbers in this portion of the output are exact, not fuzzy-rounded.
# Noteworthy Technologies ####################################
Full Text Indexing | Yes
Geospatial Types | No
Foreign Keys | Yes
Partitioning | No
InnoDB Compression | Yes
SSL | No
Explicit LOCK TABLES | No
Delayed Insert | No
XA Transactions | No
NDB Cluster | No
Prepared Statements | No
Prepared statement count | 0

This section shows some specific technologies used on this server. Some of them are detected from the schema dump
performed for the previous sections; others can be detected by looking at SHOW GLOBAL STATUS.
# InnoDB #####################################################
Version | 1.1.8
Buffer Pool Size | 16.0M
Buffer Pool Fill | 100%
Buffer Pool Dirty | 0%
File Per Table | OFF



Page Size | 16k
Log File Size | 2 * 5.0M = 10.0M
Log Buffer Size | 8M
Flush Method |
Flush Log At Commit | 1
XA Support | ON
Checksums | ON
Doublewrite | ON
R/W I/O Threads | 4 4
I/O Capacity | 200
Thread Concurrency | 0
Concurrency Tickets | 500
Commit Concurrency | 0
Txn Isolation Level | REPEATABLE-READ
Adaptive Flushing | ON
Adaptive Checkpoint |
Checkpoint Age 0 |
InnoDB Queue |
0 queries inside InnoDB, 0 queries in queue
Oldest Transaction |
0 Seconds
History List Len 209 |
Read Views 1 |
Undo Log Entries |
1 transactions, 1 total undo, 1 max undo
Pending I/O Reads |
0 buf pool reads, 0 normal AIO,
0 ibuf AIO, 0 preads
Pending I/O Writes | 0 buf pool (0 LRU, 0 flush list, 0 page);
0 AIO, 0 sync, 0 log IO (0 log, 0 chkp);
0 pwrites
Pending I/O Flushes | 0 buf pool, 0 log
Transaction States | 1xnot started

This section shows important configuration variables for the InnoDB storage engine. The buffer pool fill percent and
dirty percent are fuzzy-rounded. The last few lines are derived from the output of SHOW INNODB STATUS. It is
likely that this output will change in the future to become more useful.
# MyISAM #####################################################
Key Cache | 16.0M
Pct Used | 10%
Unflushed | 0%

This section shows the size of the MyISAM key cache, followed by the percentage of the cache in use and percentage
unflushed (fuzzy-rounded).
# Security ###################################################
Users | 2 users, 0 anon, 0 w/o pw, 0 old pw
Old Passwords | OFF

This section is generated from queries to tables in the mysql system database. It shows how many users exist, and
various potential security risks such as old-style passwords and users without passwords.
# Binary Logging #############################################
Binlogs | 1
Zero-Sized | 0
Total Size | 21.8M
binlog_format | STATEMENT
expire_logs_days | 0
sync_binlog | 0
server_id | 12345
binlog_do_db |
binlog_ignore_db |



This section shows conﬁguration and status of the binary logs. If there are zero-sized binary logs, then it is possible
that the binlog index is out of sync with the binary logs that actually exist on disk.
# Noteworthy Variables #######################################
Auto-Inc Incr/Offset | 1/1
default_storage_engine | InnoDB
flush_time | 0
init_connect |
init_file |
sql_mode |
join_buffer_size | 128k
sort_buffer_size | 2M
read_buffer_size | 128k
read_rnd_buffer_size | 256k
bulk_insert_buffer | 0.00
max_heap_table_size | 16M
tmp_table_size | 16M
max_allowed_packet | 1M
thread_stack | 192k
log | OFF
log_error | /tmp/12345/data/mysqld.log
log_warnings | 1
log_slow_queries | ON
log_queries_not_using_indexes | OFF
log_slave_updates | ON

This section shows several noteworthy server conﬁguration variables that might be important to know about when
working with this server.
# Configuration File #########################################
Config File | /tmp/12345/my.sandbox.cnf
[client]
user = msandbox
password = msandbox
port = 12345
socket = /tmp/12345/mysql_sandbox12345.sock
[mysqld]
port = 12345
socket = /tmp/12345/mysql_sandbox12345.sock
pid-file = /tmp/12345/data/mysql_sandbox12345.pid
basedir = /home/baron/5.5.20
datadir = /tmp/12345/data
key_buffer_size = 16M
innodb_buffer_pool_size = 16M
innodb_data_home_dir = /tmp/12345/data
innodb_log_group_home_dir = /tmp/12345/data
innodb_data_file_path = ibdata1:10M:autoextend
innodb_log_file_size = 5M
log-bin = mysql-bin
relay_log = mysql-relay-bin
log_slave_updates
server-id = 12345
report-host = 127.0.0.1
report-port = 12345
log-error = mysqld.log
innodb_lock_wait_timeout = 3
# The End ####################################################



This section shows a pretty-printed version of the my.cnf file, with comments removed and with whitespace added to
align things for easy reading. The tool tries to detect the my.cnf file by looking at the output of ps, and if it does not
find the location of the file there, it tries common locations until it finds a file. Note that this file might not actually
correspond with the server from which the report was generated. This can happen when the tool isn’t run on the same
server it’s reporting on, or when detecting the location of the configuration file fails.

2.17.6 OPTIONS

All options after – are passed to mysql.
-config
type: string
Read this comma-separated list of config files. If specified, this must be the first option on the command line.
-help
-save-samples
type: string
Save the data files used to generate the summary in this directory.
-read-samples
type: string
Create a report from the files found in this directory.
-databases
type: string
Names of databases to summarize. If you want all of them, you can use the value --all-databases; you
can also pass in a comma-separated list of database names. If not provided, the program will ask you for manual
input.
-sleep
Seconds to sleep when gathering status counters.
-version
Print tool’s version and exit.

2.17.7 ENVIRONMENT



This tool requires Bash v3 or newer, Perl 5.8 or newer, and binutils. These are generally already provided by most
distributions. On BSD systems, it may require a mounted procfs.

2.17.9 BUGS

For a list of known bugs, see http://www.percona.com/bugs/pt-mysql-summary.



• Tool --version

2.17.10 DOWNLOADING






2.17.11 AUTHORS

Baron Schwartz, Brian Fraser, and Daniel Nichter.




welcome.



2.17.14 VERSION

pt-mysql-summary 2.1.1

2.18 pt-online-schema-change

2.18.1 NAME

pt-online-schema-change - ALTER tables without locking them.

2.18.2 SYNOPSIS

Usage

pt-online-schema-change [OPTIONS] DSN

pt-online-schema-change alters a table’s structure without blocking reads or writes. Specify the database and table
in the DSN. Do not use this tool before reading its documentation and checking your backups carefully.
Add a column to sakila.actor:
pt-online-schema-change --alter "ADD COLUMN c1 INT" D=sakila,t=actor

Change sakila.actor to InnoDB, effectively performing OPTIMIZE TABLE in a non-blocking fashion because it is
already an InnoDB table:
pt-online-schema-change --alter "ENGINE=InnoDB" D=sakila,t=actor

2.18.3 RISKS

pt-online-schema-change modiﬁes data and structures. You should be careful with it, and test it before using it in
production. You should also ensure that you have recoverable backups before using this tool.
At the time of this release, we know of no bugs that could cause harm to users.
online-schema-change.

2.18.4 DESCRIPTION

pt-online-schema-change emulates the way that MySQL alters tables internally, but it works on a copy of the table
you wish to alter. This means that the original table is not locked, and clients may continue to read and change data in
it.



pt-online-schema-change works by creating an empty copy of the table to alter, modifying it as desired, and then
copying rows from the original table into the new table. When the copy is complete, it moves away the original table
and replaces it with the new one. By default, it also drops the original table.
The data copy process is performed in small chunks of data, which are varied to attempt to make them execute in
a specific amount of time (see --chunk-time). This process is very similar to how other tools, such as pt-table-
checksum, work. Any modifications to data in the original tables during the copy will be reflected in the new table,
because the tool creates triggers on the original table to update the corresponding rows in the new table. The use of
triggers means that the tool will not work if any triggers are already defined on the table.
When the tool finishes copying data into the new table, it uses an atomic RENAME TABLE operation to simultaneously
rename the original and new tables. After this is complete, the tool drops the original table.
Foreign keys complicate the tool’s operation and introduce additional risk. The technique of atomically renaming the
original and new tables does not work when foreign keys refer to the table. The tool must update foreign keys to refer
to the new table after the schema change is complete. The tool supports two methods for accomplishing this. You can
read more about this in the documentation for --alter-foreign-keys-method.
Foreign keys also cause some side effects. The final table will have the same foreign keys and indexes as the original
table (unless you specify differently in your ALTER statement), but the names of the objects may be changed slightly
to avoid object name collisions in MySQL and InnoDB.
For safety, the tool does not modify the table unless you specify the --execute option, which is not enabled
by default. The tool supports a variety of other measures to prevent unwanted load or other problems, including
automatically detecting replicas, connecting to them, and using the following safety checks:
• The tool refuses to operate if it detects replication filters. See --[no]check-replication-filters for
details.
• The tool pauses the data copy operation if it observes any replicas that are delayed in replication. See
--max-lag for details.
• The tool pauses or aborts its operation if it detects too much load on the server. See --max-load and
--critical-load for details.
• The tool sets its lock wait timeout to 1 second so that it is more likely to be the victim of any lock contention,
and less likely to disrupt other transactions. See --lock-wait-timeout for details.
• The tool refuses to alter the table if foreign key constraints reference it, unless you specify
--alter-foreign-keys-method.

2.18.5 OUTPUT

The tool prints information about its activities to STDOUT so that you can see what it is doing. During the data copy
phase, it prints progress reports to STDERR. You can get additional information with the --print option.

2.18.6 OPTIONS

--dry-run and --execute are mutually exclusive.
-alter
type: string
The schema modification, without the ALTER TABLE keywords. You can perform multiple modifications to the
table by specifying them with commas. Please refer to the MySQL manual for the syntax of ALTER TABLE.
You cannot use the RENAME clause to ALTER TABLE, or the tool will fail.

2.18. pt-online-schema-change 127


-alter-foreign-keys-method
type: string
How to modify foreign keys so they reference the new table. Foreign keys that reference the table to be altered
must be treated specially to ensure that they continue to reference the correct table. When the tool renames the
original table to let the new one take its place, the foreign keys “follow” the renamed table, and must be changed
to reference the new table instead.
The tool supports two techniques to achieve this. It automatically ﬁnds “child tables” that reference the table to
be altered.
auto
Automatically determine which method is best. The tool uses rebuild_constraints if possible
(see the description of that method for details), and if not, then it uses drop_swap.
rebuild_constraints
This method uses ALTER TABLE to drop and re-add foreign key constraints that reference the new
table. This is the preferred technique, unless one or more of the “child” tables is so large that the
ALTER would take too long. The tool determines that by comparing the number of rows in the child
table to the rate at which the tool is able to copy rows from the old table to the new table. If the tool
estimates that the child table can be altered in less time than the --chunk-time, then it will use
this technique. For purposes of estimating the time required to alter the child table, the tool multiplies
the row-copying rate by --chunk-size-limit, because MySQL’s ALTER TABLE is typically
much faster than the external process of copying rows.
Due to a limitation in MySQL, foreign keys will not have the same names after the ALTER that they
did prior to it. The tool has to rename the foreign key when it redeﬁnes it, which adds a leading
underscore to the name. In some cases, MySQL also automatically renames indexes required for the
foreign key.
drop_swap
Disable foreign key checks (FOREIGN_KEY_CHECKS=0), then drop the original table before re-
naming the new table into its place. This is different from the normal method of swapping the old
and new table, which uses an atomic RENAME that is undetectable to client applications.
This method is faster and does not block, but it is riskier for two reasons. First, for a short time
between dropping the original table and renaming the temporary table, the table to be altered simply
does not exist, and queries against it will result in an error. Secondly, if there is an error and the new
table cannot be renamed into the place of the old one, then it is too late to abort, because the old table
is gone permanently.
none
This method is like drop_swap without the “swap”. Any foreign keys that referenced the original
table will now reference a nonexistent table. This will typically cause foreign key violations that are
visible in SHOW ENGINE INNODB STATUS, similar to the following:
Trying to add to index ‘idx_fk_staff_id‘ tuple:
DATA TUPLE: 2 fields;
0: len 1; hex 05; asc ;;
1: len 4; hex 80000001; asc ;;
But the parent table ‘sakila‘.‘staff_old‘
or its .ibd file does not currently exist!

This is because the original table (in this case, sakila.staff) was renamed to sakila.staff_old and then
dropped. This method of handling foreign key constraints is provided so that the database adminis-
trator can disable the tool’s built-in functionality if desired.



-ask-pass
-charset
-check-interval
Sleep time between checks for --max-lag.
-[no]check-replication-filters
default: yes
Abort if any replication filter is set on any server. The tool looks for server options that filter replication, such
as binlog_ignore_db and replicate_do_db. If it finds any such filters, it aborts with an error.
If the replicas are configured with any filtering options, you should be careful not to modify any databases
or tables that exist on the master and not the replicas, because it could cause replication to fail. For more
information on replication rules, see http://dev.mysql.com/doc/en/replication-rules.html.
-check-slave-lag
type: string
Pause the data copy until this replica’s lag is less than --max-lag. The value is a DSN that inherits prop-
erties from the the connection options (--port, --user, etc.). This option overrides the normal behavior
of finding and continually monitoring replication lag on ALL connected replicas. If you don’t want to mon-
itor ALL replicas, but you want more than just one replica to be monitored, then use the DSN option to the
--recursion-method option instead of this option.
-chunk-index
type: string
Prefer this index for chunking tables. By default, the tool chooses the most appropriate index for chunking.
This option lets you specify the index that you prefer. If the index doesn’t exist, then the tool will fall back to
its default behavior of choosing an index. The tool adds the index to the SQL statements in a FORCE INDEX
clause. Be careful when using this option; a poor choice of index could cause bad performance.
-chunk-size
type: size; default: 1000
Number of rows to select for each chunk copied. Allowable suffixes are k, M, G.
This option can override the default behavior, which is to adjust chunk size dynamically to try to make chunks
run in exactly --chunk-time seconds. When this option isn’t set explicitly, its default value is used as a
starting point, but after that, the tool ignores this option’s value. If you set this option explicitly, however, then
it disables the dynamic adjustment behavior and tries to make all chunks exactly the specified number of rows.
There is a subtlety: if the chunk index is not unique, then it’s possible that chunks will be larger than desired.
For example, if a table is chunked by an index that contains 10,000 of a given value, there is no way to write a
WHERE clause that matches only 1,000 of the values, and that chunk will be at least 10,000 rows large. Such a
chunk will probably be skipped because of --chunk-size-limit.
-chunk-size-limit
Do not copy chunks this much larger than the desired chunk size.



When a table has no unique indexes, chunk sizes can be inaccurate. This option speciﬁes a maximum tolerable
limit to the inaccuracy. The tool uses <EXPLAIN> to estimate how many rows are in the chunk. If that estimate
exceeds the desired chunk size times the limit, then the tool skips the chunk.
The minimum value for this option is 1, which means that no chunk can be larger than --chunk-size. You
probably don’t want to specify 1, because rows reported by EXPLAIN are estimates, which can be different
from the real number of rows in the chunk. You can disable oversized chunk checking by specifying a value of
0.
The tool also uses this option to determine how to handle foreign keys that reference the table to be altered. See
--alter-foreign-keys-method for details.
-chunk-time
Adjust the chunk size dynamically so each data-copy query takes this long to execute. The tool tracks the copy
rate (rows per second) and adjusts the chunk size after each data-copy query, so that the next query takes this
amount of time (in seconds) to execute. It keeps an exponentially decaying moving average of queries per
second, so that if the server’s performance changes due to changes in server load, the tool adapts quickly.
If this option is set to zero, the chunk size doesn’t auto-adjust, so query times will vary, but query chunk sizes
will not. Another way to do the same thing is to specify a value for --chunk-size explicitly, instead of
leaving it at the default.
-config
type: Array
-critical-load
type: Array; default: Threads_running=50
Examine SHOW GLOBAL STATUS after every chunk, and abort if the load is too high. The option accepts a
comma-separated list of MySQL status variables and thresholds. An optional =MAX_VALUE (or :MAX_VALUE)
can follow each variable. If not given, the tool determines a threshold by examining the current value at startup
and doubling it.
See --max-load for further details. These options work similarly, except that this option will abort the tool’s
operation instead of pausing it, and the default value is computed differently if you specify no threshold. The
reason for this option is as a safety check in case the triggers on the original table add so much load to the server
that it causes downtime. There is probably no single value of Threads_running that is wrong for every server,
but a default of 50 seems likely to be unacceptably high for most servers, indicating that the operation should be
canceled immediately.
-defaults-file
-[no]drop-old-table
default: yes
Drop the original table after renaming it. After the original table has been successfully renamed to let the new
table take its place, and if there are no errors, the tool drops the original table by default. If there are any errors,
the tool leaves the original table in place.
-dry-run
Create and alter the new table, but do not create triggers, copy data, or replace the original table.
-execute
Indicate that you have read the documentation and want to alter the table. You must specify this option to alter
the table. If you do not, then the tool will only perform some safety checks and exit. This helps ensure that you



have read the documentation and understand how to use this tool. If you have not read the documentation, then
do not specify this option.
-help
Show help and exit.
-host
Connect to host.
-lock-wait-timeout
Set the session value of innodb_lock_wait_timeout. This option helps guard against long lock waits
if the data-copy queries become slow for some reason. Setting this option dynamically requires the InnoDB
plugin, so this works only on newer InnoDB and MySQL versions. If the setting’s current value is greater than
the speciﬁed value, and the tool cannot set the value as desired, then it prints a warning. If the tool cannot set
the value but the current value is less than or equal to the desired value, there is no error.
-max-lag
Pause the data copy until all replicas’ lag is less than this value. After each data-copy query (each chunk),
the tool looks at the replication lag of all replicas to which it connects, using Seconds_Behind_Master. If any
replica is lagging more than the value of this option, then the tool will sleep for --check-interval seconds,
then check all replicas again. If you specify --check-slave-lag, then the tool only examines that server
for lag, not all servers. If you want to control exactly which servers the tool monitors, use the DSN value to
--recursion-method.
The tool waits forever for replicas to stop lagging. If any replica is stopped, the tool waits forever until the
replica is started. The data copy continues when all replicas are running and not lagging too much.
The tool prints progress reports while waiting. If a replica is stopped, it prints a progress report immediately,
then again at every progress report interval.
-max-load
type: Array; default: Threads_running=25
Examine SHOW GLOBAL STATUS after every chunk, and pause if any status variables are higher than their
thresholds. The option accepts a comma-separated list of MySQL status variables. An optional =MAX_VALUE
(or :MAX_VALUE) can follow each variable. If not given, the tool determines a threshold by examining the
current value and increasing it by 20%.
For example, if you want the tool to pause when Threads_connected gets too high, you can specify
“Threads_connected”, and the tool will check the current value when it starts working and add 20% to that
value. If the current value is 100, then the tool will pause when Threads_connected exceeds 120, and resume
working when it is below 120 again. If you want to specify an explicit threshold, such as 110, you can use either
“Threads_connected:110” or “Threads_connected=110”.
The purpose of this option is to prevent the tool from adding too much load to the server. If the data-copy queries
are intrusive, or if they cause lock waits, then other queries on the server will tend to block and queue. This will
typically cause Threads_running to increase, and the tool can detect that by running SHOW GLOBAL STATUS
immediately after each query ﬁnishes. If you specify a threshold for this variable, then you can instruct the tool
to wait until queries are running normally again. This will not prevent queueing, however; it will only give the
server a chance to recover from the queueing. If you notice queueing, it is best to decrease the chunk time.
-password



-pid
type: string
Create the given PID file. The file contains the process ID of the tool’s instance. The PID file is removed when
the tool exits. The tool checks for the existence of the PID file when starting; if it exists and the process with the
matching PID exists, the tool exits.
-port
-print
Print SQL statements to STDOUT. Specifying this option allows you to see most of the statements that the tool
executes. You can use this option with --dry-run, for example.
-progress
Print progress reports to STDERR while copying rows. The value is a comma-separated list with two parts. The
first part can be percentage, time, or iterations; the second part specifies how often an update should be printed,
in percentage, seconds, or number of iterations.
-quiet
short form: -q
Do not print messages to STDOUT. Errors and warnings are still printed to STDERR.
-recurse
type: int
Number of levels to recurse in the hierarchy when discovering replicas. Default is infinite. See also
--recursion-method.
-recursion-method
type: string
Preferred recursion method for discovering replicas. Possible methods are:
METHOD USES
=========== ==================
dsn=DSN DSNs from a table

The processlist method is the default, because SHOW SLAVE HOSTS is not reliable. However, the hosts
method can work better if the server uses a non-standard port (not 3306). The tool usually does the right thing
and finds all replicas, but you may give a preferred method and it will be used first.
The hosts method requires replicas to be configured with report_host, report_port, etc.
The dsn method is special: it specifies a table from which other DSN strings are read. The specified DSN must
specify a D and t, or a database-qualified t. The DSN table should have the following structure:
CREATE TABLE ‘dsns‘ (
‘id‘ int(11) NOT NULL AUTO_INCREMENT,
‘parent_id‘ int(11) DEFAULT NULL,
‘dsn‘ varchar(255) NOT NULL,
PRIMARY KEY (‘id‘)
);



To make the tool monitor only the hosts 10.10.1.16 and 10.10.1.17 for replication lag, insert the values
h=10.10.1.16 and h=10.10.1.17 into the table. Currently, the DSNs are ordered by id, but id and
parent_id are otherwise ignored.
-retries
Retry a chunk this many times when there is a nonfatal error. Nonfatal errors are problems such as a lock wait
timeout or the query being killed. This option applies to the data copy operation.
-set-vars
executed.
-socket
-[no]swap-tables
default: yes
Swap the original table and the new, altered table. This step completes the online schema change process by
making the table with the new schema take the place of the original table. The original table becomes the “old
table,” and the tool drops it unless you disable --[no]drop-old-table.
-user
-version

2.18.7 DSN OPTIONS

details.
• A
• D
Database for the old and new table.
• F
• h



Connect to host.
• p
• P
• S
• t
dsn: table; copy: no
Table to alter.
• u

2.18.8 ENVIRONMENT

PTDEBUG=1 pt-online-schema-change ... > FILE 2>&1



Perl.
This tool works only on MySQL 5.0.2 and newer versions, because earlier versions do not support triggers.

2.18.10 BUGS

For a list of known bugs, see http://www.percona.com/bugs/pt-online-schema-change.
• Tool --version




2.18.11 DOWNLOADING






2.18.12 AUTHORS

Daniel Nichter and Baron Schwartz


The “online schema change” concept was ﬁrst implemented by Shlomi Noach in his tool
oak-online-alter-table, part of http://code.google.com/p/openarkkit/. Engineers at Facebook then built
another version called OnlineSchemaChange.php as explained by their blog post: http://tinyurl.com/32zeb86.
This tool is a hybrid of both approaches, with additional features and functionality not present in either.







2.18.16 VERSION

pt-online-schema-change 2.1.1

2.19 pt-pmp

2.19.1 NAME

pt-pmp - Aggregate GDB stack traces for a selected program.

2.19.2 SYNOPSIS

Usage

pt-pmp [OPTIONS] [FILES]

pt-pmp is a poor man’s profiler, inspired by http://poormansprofiler.org. It can create and summarize full stack traces
of processes on Linux. Summaries of stack traces can be an invaluable tool for diagnosing what a process is waiting
for.

2.19.3 RISKS

pt-pmp is a read-only tool. However, collecting GDB stacktraces is achieved by attaching GDB to the program and
printing stack traces from all threads. This will freeze the program for some period of time, ranging from a second or
so to much longer on very busy systems with a lot of memory and many threads in the program. In the tool’s default
usage as a MySQL profiling tool, this means that MySQL will be unresponsive while the tool runs, although if you are
using the tool to diagnose an unresponsive server, there is really no reason not to do this. In addition to freezing the
server, there is also some risk of the server crashing or performing badly after GDB detaches from it.
will be marked as such. You can see a list of such issues at the following URL: http://www.percona.com/bugs/pt-pmp.

2.19.4 DESCRIPTION

pt-pmp performs two tasks: it gets a stack trace, and it summarizes the stack trace. If a file is given on the command
line, the tool skips the first step and just aggregates the file.
To summarize the stack trace, the tool extracts the function name (symbol) from each level of the stack, and combines
them with commas. It does this for each thread in the output. Afterwards, it sorts similar threads together and counts
how many of each one there are, then sorts them most-frequent first.



2.19.5 OPTIONS

Options must precede files on the command line.
-b BINARY Which binary to trace (default mysqld)
-i ITERATIONS How many traces to gather and aggregate (default 1)
-k KEEPFILE Keep the raw traces in this file after aggregation
-l NUMBER Aggregate only first NUMBER functions; 0=infinity (default 0)
-p PID Process ID of the process to trace; overrides -b
-s SLEEPTIME Number of seconds to sleep between iterations (default 0)

2.19.6 ENVIRONMENT



This tool requires Bash v3 or newer.

2.19.8 BUGS

For a list of known bugs, see http://www.percona.com/bugs/pt-pmp.
• Tool --version

2.19.9 DOWNLOADING






2.19. pt-pmp 137


2.19.10 AUTHORS

Baron Schwartz, based on a script by Domas Mituzas (http://poormansproﬁler.org/)




welcome.

2.19.13 VERSION

pt-pmp 2.1.1

2.20 pt-query-advisor

2.20.1 NAME

pt-query-advisor - Analyze queries and advise on possible problems.

2.20.2 SYNOPSIS

Usage

pt-query-advisor [OPTION...] [FILE]

pt-query-advisor analyzes queries and advises on possible problems. Queries are given either by specifying slowlog
ﬁles, –query, or –review.
Analyze all queries in a slow log:
pt-query-advisor /path/to/slow-query.log

Analyze all queries in a general log:



pt-query-advisor --type genlog mysql.log

Get queries from tcpdump using pt-query-digest:
pt-query-digest --type tcpdump.txt --print --no-report | pt-query-advisor

2.20.3 RISKS

pt-query-advisor simply reads queries and examines them, and is thus very low risk.
At the time of this release there is a bug that may cause an infinite (or very long) loop when parsing very large queries.
query-advisor.

2.20.4 DESCRIPTION

pt-query-advisor examines queries and applies rules to them, trying to find queries that look bad according to the
rules. It reports on queries that match the rules, so you can find bad practices or hidden problems in your SQL. By
default, it accepts a MySQL slow query log as input.

2.20.5 RULES

These are the rules that pt-query-advisor will apply to the queries it examines. Each rule has three bits of information:
an ID, a severity and a description.
The rule’s ID is its identifier. We use a seven-character ID, and the naming convention is three characters, a period,
and a three-digit number. The first three characters are sort of an abbreviation of the general class of the rule. For
example, ALI.001 is some rule related to how the query uses aliases.
The rule’s severity is an indication of how important it is that this rule matched a query. We use NOTE, WARN, and
CRIT to denote these levels.
The rule’s description is a textual, human-readable explanation of what it means when a query matches this rule.
Depending on the verbosity of the report you generate, you will see more of the text in the description. By default,
you’ll see only the first sentence, which is sort of a terse synopsis of the rule’s meaning. At a higher verbosity, you’ll
see subsequent sentences.
ALI.001
severity: note
Aliasing without the AS keyword. Explicitly using the AS keyword in column or table aliases, such as
“tbl AS alias,” is more readable than implicit aliases such as “tbl alias”.
ALI.002
severity: warn

2.20. pt-query-advisor 139


Aliasing the ‘*’ wildcard. Aliasing a column wildcard, such as “SELECT tbl.* col1, col2” probably
indicates a bug in your SQL. You probably meant for the query to retrieve col1, but instead it renames the
last column in the *-wildcarded list.
ALI.003
severity: note
Aliasing without renaming. The table or column’s alias is the same as its real name, and the alias just
makes the query harder to read.
ARG.001
severity: warn
Argument with leading wildcard. An argument has a leading wildcard character, such as “%foo”. The
predicate with this argument is not sargable and cannot use an index if one exists.
ARG.002
severity: note
LIKE without a wildcard. A LIKE pattern that does not include a wildcard is potentially a bug in the SQL.
CLA.001
severity: warn
SELECT without WHERE. The SELECT statement has no WHERE clause.
CLA.002
severity: note
ORDER BY RAND(). ORDER BY RAND() is a very inefﬁcient way to retrieve a random row from the
results.
CLA.003
severity: note
LIMIT with OFFSET. Paginating a result set with LIMIT and OFFSET is O(n^2) complexity, and will
cause performance problems as the data grows larger.
CLA.004
severity: note
Ordinal in the GROUP BY clause. Using a number in the GROUP BY clause, instead of an expression or
column name, can cause problems if the query is changed.
CLA.005
severity: warn
ORDER BY constant column.
CLA.006
severity: warn
GROUP BY or ORDER BY different tables will force a temp table and ﬁlesort.
CLA.007
severity: warn
ORDER BY different directions prevents index from being used. All tables in the ORDER BY clause
must be either ASC or DESC, else MySQL cannot use an index.



COL.001
severity: note
SELECT *. Selecting all columns with the * wildcard will cause the query’s meaning and behavior to
change if the table’s schema changes, and might cause the query to retrieve too much data.
COL.002
severity: note
Blind INSERT. The INSERT or REPLACE query doesn’t specify the columns explicitly, so the query’s
behavior will change if the table’s schema changes; use “INSERT INTO tbl(col1, col2) VALUES...”
instead.
LIT.001
severity: warn
Storing an IP address as characters. The string literal looks like an IP address, but is not an argument to
INET_ATON(), indicating that the data is stored as characters instead of as integers. It is more efficient
to store IP addresses as integers.
LIT.002
severity: warn
Unquoted date/time literal. A query such as “WHERE col<2010-02-12” is valid SQL but is probably a
bug; the literal should be quoted.
KWR.001
severity: note
SQL_CALC_FOUND_ROWS is inefficient. SQL_CALC_FOUND_ROWS can cause performance prob-
lems because it does not scale well; use alternative strategies to build functionality such as paginated result
screens.
JOI.001
severity: crit
Mixing comma and ANSI joins. Mixing comma joins and ANSI joins is confusing to humans, and the
behavior differs between some MySQL versions.
JOI.002
severity: crit
A table is joined twice. The same table appears at least twice in the FROM clause.
JOI.003
severity: warn
Reference to outer table column in WHERE clause prevents OUTER JOIN, implicitly converts to INNER
JOIN.
JOI.004
severity: warn
Exclusion join uses wrong column in WHERE. The exclusion join (LEFT OUTER JOIN with a WHERE
clause that is satisfied only if there is no row in the right-hand table) seems to use the wrong column in the
WHERE clause. A query such as ”... FROM l LEFT OUTER JOIN r ON l.l=r.r WHERE r.z IS NULL”
probably ought to list r.r in the WHERE IS NULL clause.



RES.001
severity: warn
Non-deterministic GROUP BY. The SQL retrieves columns that are neither in an aggregate function nor
the GROUP BY expression, so these values will be non-deterministic in the result.
RES.002
severity: warn
LIMIT without ORDER BY. LIMIT without ORDER BY causes non-deterministic results, depending on
the query execution plan.
STA.001
severity: note
!= is non-standard. Use the <> operator to test for inequality.
SUB.001
severity: crit
IN() and NOT IN() subqueries are poorly optimized. MySQL executes the subquery as a dependent
subquery for each row in the outer query. This is a frequent cause of serious performance problems. This
might change version 6.0 of MySQL, but for versions 5.1 and older, the query should be rewritten as a
JOIN or a LEFT OUTER JOIN, respectively.

2.20.6 OPTIONS

--query and --review are mutually exclusive.
-ask-pass
-charset
-config
type: Array
-[no]continue-on-error
default: yes
Continue working even if there is an error.
-daemonize
-database
Connect to this database. This is also used as the default database for --[no]show-create-table if a
query does not use database-qualiﬁed tables.



-defaults-file
-group-by
type: string; default: rule_id
Group items in the report by this attribute. Possible attributes are:
ATTRIBUTE GROUPS
========= ==========================================================
rule_id Items matching the same rule ID
query_id Queries with the same ID (the same fingerprint)
none No grouping, report each query and its advice individually

-help
Show help and exit.
-host
Connect to host.
-ignore-rules
type: hash
Ignore these rule IDs.
Specify a comma-separated list of rule IDs (e.g. LIT.001,RES.002,etc.) to ignore. Currently, the rule IDs are
case-sensitive and must be uppercase.
-password
-pid
type: string
-port
-print-all
Print all queries, even those that do not match any rules. With --group-by none, non-matching queries are
printed in the main report and profile. For other --group-by values, non-matching queries are only printed
in the profile. Non-matching queries have zeros for NOTE, WARN and CRIT in the profile.
-query
type: string
Analyze this single query and ignore files and STDIN. This option allows you to supply a single query on the
command line. Any files also specified on the command line are ignored.
-report-format
type: string; default: compact



Type of report format: full or compact. In full mode, every query’s report contains the description of the rules
it matched, even if this information was previously displayed. In compact mode, the repeated information is
suppressed, and only the rule ID is displayed.
-review
type: DSN
Analyze queries from this pt-query-digest query review table.
-sample
How many samples of the query to show.
-set-vars
executed.
-[no]show-create-table
default: yes
Get SHOW CREATE TABLE for each query’s table.
If host connection options are given (like --host, --port, etc.) then the tool will also get SHOW CREATE
TABLE for each query. This information is needed for some rules like JOI.004. If this option is disabled by
specifying --no-show-create-table then some rules may not be checked.
-socket
-type
type: Array
The type of input to parse (default slowlog). The permitted types are slowlog and genlog.
-user
-verbose
Increase verbosity of output. At the default level of verbosity, the program prints only the ﬁrst sentence of each
rule’s description. At higher levels, the program prints more of the description. See also --report-format.
-version
-where
type: string
Apply this WHERE clause to the SELECT query on the --review table.

2.20.7 DSN OPTIONS




details.
• A
• D
Database that contains the query review table.
• F
• h
Connect to host.
• p
• P
• S
• t
Table to use as the query review table.
• u

2.20.8 ENVIRONMENT

PTDEBUG=1 pt-query-advisor ... > FILE 2>&1





Perl.

2.20.10 BUGS

For a list of known bugs, see http://www.percona.com/bugs/pt-query-advisor.
• Tool --version

2.20.11 DOWNLOADING






2.20.12 AUTHORS








2.20.15 VERSION

pt-query-advisor 2.1.1

2.21 pt-query-digest

2.21.1 NAME

pt-query-digest - Analyze query execution logs and generate a query report, ﬁlter, replay, or transform queries for
MySQL, PostgreSQL, memcached, and more.

2.21.2 SYNOPSIS

Usage

pt-query-digest [OPTION...] [FILE]

pt-query-digest parses and analyzes MySQL log ﬁles. With no FILE, or when FILE is -, it read standard input.
Analyze, aggregate, and report on a slow query log:
pt-query-digest /path/to/slow.log

Review a slow log, saving results to the test.query_review table in a MySQL server running on host1. See --review
for more on reviewing queries:
pt-query-digest --review h=host1,D=test,t=query_review /path/to/slow.log

Filter out everything but SELECT queries, replay the queries against another server, then use the timings from replay-
ing them to analyze their performance:
pt-query-digest /path/to/slow.log --execute h=another_server
--filter ’$event->{fingerprint} =~ m/^select/’

Print the structure of events so you can construct a complex --filter:
pt-query-digest /path/to/slow.log --no-report
--filter ’print Dumper($event)’

2.21. pt-query-digest 147


Watch SHOW FULL PROCESSLIST and output a log in slow query log format:
pt-query-digest --processlist h=host1 --print --no-report

The default aggregation and analysis is CPU and memory intensive. Disable it if you don’t need the default report:
pt-query-digest <arguments> --no-report

2.21.3 RISKS

By default pt-query-digest merely collects and aggregates data from the files specified. It is designed to be as efficient
as possible, but depending on the input you give it, it can use a lot of CPU and memory. Practically speaking, it is safe
to run even on production systems, but you might want to monitor it until you are satisfied that the input you give it
does not cause undue load.
Various options will cause pt-query-digest to insert data into tables, execute SQL queries, and so on. These include
the --execute option and --review.
query-digest.

2.21.4 DESCRIPTION

pt-query-digest is a framework for doing things with events from a query source such as the slow query log or
PROCESSLIST. By default it acts as a very sophisticated log analysis tool. You can group and sort queries in many
different ways simultaneously and find the most expensive queries, or create a timeline of queries in the log, for
example. It can also do a “query review,” which means to save a sample of each type of query into a MySQL table so
you can easily see whether you’ve reviewed and analyzed a query before. The benefit of this is that you can keep track
of changes to your server’s queries and avoid repeated work. You can also save other information with the queries,
such as comments, issue numbers in your ticketing system, and so on.
Note that this is a work in very active progress and you should expect incompatible changes in the future.

2.21.5 ATTRIBUTES

pt-query-digest works on events, which are a collection of key/value pairs called attributes. You’ll recognize most of
the attributes right away: Query_time, Lock_time, and so on. You can just look at a slow log and see them. However,
there are some that don’t exist in the slow log, and slow logs may actually include different kinds of attributes (for
example, you may have a server with the Percona patches).
For a full list of attributes, see http://code.google.com/p/maatkit/wiki/EventAttributes.
With creative use of --filter, you can create new attributes derived from existing attributes. For example, to create
an attribute called Row_ratio for examining the ratio of Rows_sent to Rows_examined, specify a filter like:
--filter ’($event->{Row_ratio} = $event->{Rows_sent} / ($event->{Rows_examined})) && 1’



The && 1 trick is needed to create a valid one-line syntax that is always true, even if the assignment happens to
evaluate false. The new attribute will automatically appears in the output:
# Row ratio 1.00 0.00 1 0.50 1 0.71 0.50

Attributes created this way can be specified for --order-by or any option that requires an attribute.

2.21.6 memcached

memcached events have additional attributes related to the memcached protocol: cmd, key, res (result) and val. Also,
boolean attributes are created for the various commands, misses and errors: Memc_CMD where CMD is a memcached
command (get, set, delete, etc.), Memc_error and Memc_miss.
These attributes are no different from slow log attributes, so you can use them with --[no]report, --group-by,
in a --filter, etc.
These attributes and more are documented at http://code.google.com/p/maatkit/wiki/EventAttributes.

2.21.7 OUTPUT

The default output is a query analysis report. The --[no]report option controls whether or not this report is
printed. Sometimes you may wish to parse all the queries but suppress the report, for example when using --print
or --review.
There is one paragraph for each class of query analyzed. A “class” of queries all have the same value for the
--group-by attribute which is “fingerprint” by default. (See “ATTRIBUTES”.) A fingerprint is an abstracted ver-
sion of the query text with literals removed, whitespace collapsed, and so forth. The report is formatted so it’s easy to
paste into emails without wrapping, and all non-query lines begin with a comment, so you can save it to a .sql file and
open it in your favorite syntax-highlighting text editor. There is a response-time profile at the beginning.
The output described here is controlled by --report-format. That option allows you to specify what to print and
in what order. The default output in the default order is described here.
The report, by default, begins with a paragraph about the entire analysis run The information is very similar to what
you’ll see for each class of queries in the log, but it doesn’t have some information that would be too expensive to keep
globally for the analysis. It also has some statistics about the code’s execution itself, such as the CPU and memory
usage, the local date and time of the run, and a list of input file read/parsed.
Following this is the response-time profile over the events. This is a highly summarized view of the unique events in
the detailed query report that follows. It contains the following columns:
Column Meaning
============ ==========================================================
Rank The query’s rank within the entire set of queries analyzed
Query ID The query’s fingerprint
Response time The total response time, and percentage of overall total
Calls The number of times this query was executed
R/Call The mean response time per execution
Apdx The Apdex score; see --apdex-threshold for details
V/M The Variance-to-mean ratio of response time
EXPLAIN If --explain was specified, a sparkline; see --explain
Item The distilled query

A final line whose rank is shown as MISC contains aggregate statistics on the queries that were not included in the
report, due to options such as --limit and --outliers. For details on the variance-to-mean ratio, please see
http://en.wikipedia.org/wiki/Index_of_dispersion.



Next, the detailed query report is printed. Each query appears in a paragraph. Here is a sample, slightly reformatted
so ‘perldoc’ will not wrap lines in a terminal. The following will all be one paragraph, but we’ll break it up for
commentary.
# Query 2: 0.01 QPS, 0.02x conc, ID 0xFDEA8D2993C9CAF3 at byte 160665

This line identifies the sequential number of the query in the sort order specified by --order-by. Then there’s the
queries per second, and the approximate concurrency for this query (calculated as a function of the timespan and total
Query_time). Next there’s a query ID. This ID is a hex version of the query’s checksum in the database, if you’re
using --review. You can select the reviewed query’s details from the database with a query like SELECT ....
WHERE checksum=0xFDEA8D2993C9CAF3.
If you are investigating the report and want to print out every sample of a particular query, then the following
--filter may be helpful: ‘‘pt-query-digest slow-log.log –no-report –print –filter ‘$event-‘‘{fingerprint} &&
make_checksum($event->{fingerprint}) eq “FDEA8D2993C9CAF3”’>.
Notice that you must remove the 0x prefix from the checksum in order for this to work.
Finally, in case you want to find a sample of the query in the log file, there’s the byte offset where you can look. (This
is not always accurate, due to some silly anomalies in the slow-log format, but it’s usually right.) The position refers
to the worst sample, which we’ll see more about below.
Next is the table of metrics about this class of queries.
# pct total min max avg 95% stddev median
# Count 0 2
# Exec time 13 1105s 552s 554s 553s 554s 2s 553s
# Lock time 0 216us 99us 117us 108us 117us 12us 108us
# Rows sent 20 6.26M 3.13M 3.13M 3.13M 3.13M 12.73 3.13M
# Rows exam 0 6.26M 3.13M 3.13M 3.13M 3.13M 12.73 3.13M

The first line is column headers for the table. The percentage is the percent of the total for the whole analysis run,
and the total is the actual value of the specified metric. For example, in this case we can see that the query executed 2
times, which is 13% of the total number of queries in the file. The min, max and avg columns are self-explanatory. The
95% column shows the 95th percentile; 95% of the values are less than or equal to this value. The standard deviation
shows you how tightly grouped the values are. The standard deviation and median are both calculated from the 95th
percentile, discarding the extremely large values.
The stddev, median and 95th percentile statistics are approximate. Exact statistics require keeping every value seen,
sorting, and doing some calculations on them. This uses a lot of memory. To avoid this, we keep 1000 buckets,
each of them 5% bigger than the one before, ranging from .000001 up to a very big number. When we see a value
we increment the bucket into which it falls. Thus we have fixed memory per class of queries. The drawback is the
imprecision, which typically falls in the 5 percent range.
Next we have statistics on the users, databases and time range for the query.
# Users 1 user1
# Databases 2 db1(1), db2(1)
# Time range 2008-11-26 04:55:18 to 2008-11-27 00:15:15

The users and databases are shown as a count of distinct values, followed by the values. If there’s only one, it’s shown
alone; if there are many, we show each of the most frequent ones, followed by the number of times it appears.
# Query_time distribution
# 1us
# 10us
# 100us
# 1ms
# 10ms
# 100ms



# 1s
# 10s+ #############################################################

The execution times show a logarithmic chart of time clustering. Each query goes into one of the “buckets” and is
counted up. The buckets are powers of ten. The first bucket is all values in the “single microsecond range” – that is,
less than 10us. The second is “tens of microseconds,” which is from 10us up to (but not including) 100us; and so on.
The charted attribute can be changed by specifying --report-histogram but is limited to time-based attributes.
# Tables
# SHOW TABLE STATUS LIKE ’table1’G
# SHOW CREATE TABLE ‘table1‘G
# EXPLAIN
SELECT * FROM table1G

This section is a convenience: if you’re trying to optimize the queries you see in the slow log, you probably want to
examine the table structure and size. These are copy-and-paste-ready commands to do that.
Finally, we see a sample of the queries in this class of query. This is not a random sample. It is the query that performed
the worst, according to the sort order given by --order-by. You will normally see a commented # EXPLAIN line
just before it, so you can copy-paste the query to examine its EXPLAIN plan. But for non-SELECT queries that isn’t
possible to do, so the tool tries to transform the query into a roughly equivalent SELECT query, and adds that below.
If you want to find this sample event in the log, use the offset mentioned above, and something like the following:
tail -c +<offset> /path/to/file | head

See also --report-format.

2.21.8 SPARKLINES

The output also contains sparklines. Sparklines are “data-intense, design-simple, word-sized graphics”
(http://en.wikipedia.org/wiki/Sparkline).There is a sparkline for --report-histogram and for --explain. See
each of those options for details about interpreting their sparklines.

2.21.9 QUERY REVIEWS

A “query review” is the process of storing all the query fingerprints analyzed. This has several benefits:
• You can add meta-data to classes of queries, such as marking them for follow-up, adding notes to queries, or
marking them with an issue ID for your issue tracking system.
• You can refer to the stored values on subsequent runs so you’ll know whether you’ve seen a query before. This
can help you cut down on duplicated work.
• You can store historical data such as the row count, query times, and generally anything you can see in the
report.
To use this feature, you run pt-query-digest with the --review option. It will store the fingerprints and other
information into the table you specify. Next time you run it with the same option, it will do the following:
• It won’t show you queries you’ve already reviewed. A query is considered to be already reviewed if you’ve
set a value for the reviewed_by column. (If you want to see queries you’ve already reviewed, use the
--report-all option.)
• Queries that you’ve reviewed, and don’t appear in the output, will cause gaps in the query number sequence in
the first line of each paragraph. And the value you’ve specified for --limit will still be honored. So if you’ve
reviewed all queries in the top 10 and you ask for the top 10, you won’t see anything in the output.



• If you want to see the queries you’ve already reviewed, you can specify --report-all. Then you’ll see the
normal analysis output, but you’ll also see the information from the review table, just below the execution time
graph. For example,
# Review information
# comments: really bad IN() subquery, fix soon!
# first_seen: 2008-12-01 11:48:57
# jira_ticket: 1933
# last_seen: 2008-12-18 11:49:07
# priority: high
# reviewed_by: xaprb
# reviewed_on: 2008-12-18 15:03:11

You can see how useful this meta-data is – as you analyze your queries, you get your comments integrated right
into the report.
If you add the --review-history option, it will also store information into a separate database table, so
you can keep historical trending information on classes of queries.

2.21.10 FINGERPRINTS

A query fingerprint is the abstracted form of a query, which makes it possible to group similar queries together.
Abstracting a query removes literal values, normalizes whitespace, and so on. For example, consider these two queries:
SELECT name, password FROM user WHERE id=’12823’;
select name, password from user
where id=5;

Both of those queries will fingerprint to
select name, password from user where id=?

Once the query’s fingerprint is known, we can then talk about a query as though it represents all similar queries.
What pt-query-digest does is analogous to a GROUP BY statement in SQL. (But note that “multiple columns” doesn’t
define a multi-column grouping; it defines multiple reports!) If your command-line looks like this,
pt-query-digest /path/to/slow.log --select Rows_read,Rows_sent
--group-by fingerprint --order-by Query_time:sum --limit 10

The corresponding pseudo-SQL looks like this:
SELECT WORST(query BY Query_time), SUM(Query_time), ...
FROM /path/to/slow.log
GROUP BY FINGERPRINT(query)
ORDER BY SUM(Query_time) DESC
LIMIT 10

You can also use the value distill, which is a kind of super-fingerprint. See --group-by for more.
When parsing memcached input (--type memcached), the fingerprint is an abstracted version of the com-
mand and key, with placeholders removed. For example, get user_123_preferences fingerprints to get
user_?_preferences. There is also a key_print which a fingerprinted version of the key. This example’s
key_print is user_?_preferences.
Query fingerprinting accommodates a great many special cases, which have proven necessary in the real world. For
example, an IN list with 5 literals is really equivalent to one with 4 literals, so lists of literals are collapsed to a single
one. If you want to understand more about how and why all of these cases are handled, please review the test cases in
the Subversion repository. If you find something that is not fingerprinted properly, please submit a bug report with a
reproducible test case. Here is a list of transformations during fingerprinting, which might not be exhaustive:



• Group all SELECT queries from mysqldump together, even if they are against different tables. Ditto for all of
pt-table-checksum’s checksum queries.
• Shorten multi-value INSERT statements to a single VALUES() list.
• Strip comments.
• Abstract the databases in USE statements, so all USE statements are grouped together.
• Replace all literals, such as quoted strings. For efficiency, the code that replaces literal numbers is somewhat
non-selective, and might replace some things as numbers when they really are not. Hexadecimal literals are
also replaced. NULL is treated as a literal. Numbers embedded in identifiers are also replaced, so tables named
similarly will be fingerprinted to the same values (e.g. users_2009 and users_2010 will fingerprint identically).
• Collapse all whitespace into a single space.
• Lowercase the entire query.
• Replace all literals inside of IN() and VALUES() lists with a single placeholder, regardless of cardinality.
• Collapse multiple identical UNION queries into a single one.

2.21.11 OPTIONS

DSN values in --review-history default to values in --review if COPY is yes.
-apdex-threshold
Set Apdex target threshold (T) for query response time. The Application Performance Index (Apdex) Technical
Specification V1.1 defines T as “a positive decimal value in seconds, having no more than two significant digits
of granularity.” This value only applies to query response time (Query_time).
Options can be abbreviated so specifying --apdex-t also works.
See http://www.apdex.org/.
-ask-pass
-attribute-aliases
type: array; default: db|Schema
List of attribute|alias,etc.
Certain attributes have multiple names, like db and Schema. If an event does not have the primary attribute,
pt-query-digest looks for an alias attribute. If it finds an alias, it creates the primary attribute with the alias
attribute’s value and removes the alias attribute.
If the event has the primary attribute, all alias attributes are deleted.
This helps simplify event attributes so that, for example, there will not be report lines for both db and Schema.
-attribute-value-limit
A sanity limit for attribute values.
This option deals with bugs in slow-logging functionality that causes large values for attributes. If the attribute’s
value is bigger than this, the last-seen value for that class of query is used instead.



-aux-dsn
type: DSN
Auxiliary DSN used for special options.
The following options may require a DSN even when only parsing a slow log file:
* --since
* --until

See each option for why it might require a DSN.
-charset
-check-attributes-limit
Stop checking for new attributes after this many events.
For better speed, pt-query-digest stops checking events for new attributes after a certain number of events. Any
new attributes after this number will be ignored and will not be reported.
One special case is new attributes for pre-existing query classes (see --group-by about query classes).
New attributes will not be added to pre-existing query classes even if the attributes are detected before the
--check-attributes-limit limit.
-config
type: Array
default: yes
Continue parsing even if there is an error.
-create-review-history-table
Create the --review-history table if it does not exist.
This option causes the table specified by --review-history to be created with the default structure shown
in the documentation for that option.
-create-review-table
Create the --review table if it does not exist.
This option causes the table specified by --review to be created with the default structure shown in the
documentation for that option.
-daemonize
-defaults-file
-embedded-attributes
type: array
Two Perl regex patterns to capture pseudo-attributes embedded in queries.



Embedded attributes might be special attribute-value pairs that you’ve hidden in comments. The first regex
should match the entire set of attributes (in case there are multiple). The second regex should match and capture
attribute-value pairs from the first regex.
For example, suppose your query looks like the following:
SELECT * from users -- file: /login.php, line: 493;

You might run pt-query-digest with the following option:
:program:‘pt-query-digest‘ --embedded-attributes ’ -- .*’,’(w+): ([^,]+)’

The first regular expression captures the whole comment:
" -- file: /login.php, line: 493;"

The second one splits it into attribute-value pairs and adds them to the event:
ATTRIBUTE VALUE
========= ==========
file /login.php
line 493

NOTE: All commas in the regex patterns must be escaped with otherwise the pattern will break.
-execute
type: DSN
Execute queries on this DSN.
Adds a callback into the chain, after filters but before the reports. Events are executed on this DSN. If they are
successful, the time they take to execute overwrites the event’s Query_time attribute and the original Query_time
value (from the log) is saved as the Exec_orig_time attribute. If unsuccessful, the callback returns false and
terminates the chain.
If the connection fails, pt-query-digest tries to reconnect once per second.
See also --mirror and --execute-throttle.
-execute-throttle
type: array
Throttle values for --execute.
By default --execute runs without any limitations or concerns for the amount of time that it takes to execute
the events. The --execute-throttle allows you to limit the amount of time spent doing --execute
relative to the other processes that handle events. This works by marking some events with a Skip_exec
attribute when --execute begins to take too much time. --execute will not execute an event if this
attribute is true. This indirectly decreases the time spent doing --execute.
The --execute-throttle option takes at least two comma-separated values: max allowed --execute
time as a percentage and a check interval time. An optional third value is a percentage step for increasing and
decreasing the probability that an event will be marked Skip_exec true. 5 (percent) is the default step.
For example: --execute-throttle 70,60,10. This will limit --execute to 70% of total event
processing time, checked every minute (60 seconds) and probability stepped up and down by 10%. When
--execute exceeds 70%, the probability that events will be marked Skip_exec true increases by 10%.
--execute time is checked again after another minute. If it’s still above 70%, then the probability will in-
crease another 10%. Or, if it’s dropped below 70%, then the probability will decrease by 10%.
-expected-range
type: array; default: 5,10



Explain items when there are more or fewer than expected.
Defines the number of items expected to be seen in the report given by --[no]report, as controlled by
--limit and --outliers. If there are more or fewer items in the report, each one will explain why it was
included.
-explain
type: DSN
Run EXPLAIN for the sample query with this DSN and print results.
This works only when --group-by includes fingerprint. It causes pt-query-digest to run EXPLAIN and
include the output into the report. For safety, queries that appear to have a subquery that EXPLAIN will execute
won’t be EXPLAINed. Those are typically “derived table” queries of the form
select ... from ( select .... ) der;

The EXPLAIN results are printed in three places: a sparkline in the event header, a full vertical format in the
event report, and a sparkline in the profile.
The full format appears at the end of each event report in vertical style (G) just like MySQL prints it.
The sparklines (see “SPARKLINES”) are compact representations of the access type for each table and whether
or not “Using temporary” or “Using filesort” appear in EXPLAIN. The sparklines look like:
nr>TF

That sparkline means that there are two tables, the first uses a range (n) access, the second uses a ref access,
and both “Using temporary” (T) and “Using filesort” (F) appear. The greater-than character just separates table
access codes from T and/or F.
The abbreviated table access codes are:
a ALL
c const
e eq_ref
f fulltext
i index
m index_merge
n range
o ref_or_null
r ref
s system
u unique_subquery

A capitalized access code means that “Using index” appears in EXPLAIN for that table.
-filter
type: string
argument: $event. This is a hashref. If the given value is a readable file, then pt-query-digest reads the entire
file and uses its contents as the code. The file should not contain a shebang (#!/usr/bin/perl) line.






If the filter code won’t compile, pt-query-digest will die with an error. If the filter code does compile, an
error may still occur at runtime if the code tries to do something wrong (like pattern match an undefined value).
pt-query-digest does not provide any safeguards so code carefully!
An example filter that discards everything but SELECT statements:

This is compiled into a subroutine like the following:
sub { $event = shift; ( $event->{arg} =~ m/^select/i ) && return $event; }

You can find an explanation of the structure of $event at http://code.google.com/p/maatkit/wiki/EventAttributes.
Here are more examples of filter code:
Host/IP matches domain.com
–filter ‘($event->{host} || $event->{ip} || “”) =~ m/domain.com/’
Sometimes MySQL logs the host where the IP is expected. Therefore, we check both.
User matches john
–filter ‘($event->{user} || “”) =~ m/john/’
More than 1 warning
–filter ‘($event->{Warning_count} || 0) > 1’
Query does full table scan or full join
–filter ‘(($event->{Full_scan} || “”) eq “Yes”) || (($event->{Full_join} || “”) eq “Yes”)’
Query was not served from query cache
–filter ‘($event->{QC_Hit} || “”) eq “No”’
Query is 1 MB or larger
–filter ‘$event->{bytes} >= 1_048_576’
Since --filter allows you to alter $event, you can use it to do other things, like create new attributes. See
“ATTRIBUTES” for an example.
-fingerprints
Add query fingerprints to the standard query analysis report. This is mostly useful for debugging purposes.
-[no]for-explain
default: yes
Print extra information to make analysis easy.



This option adds code snippets to make it easy to run SHOW CREATE TABLE and SHOW TABLE STATUS for
the query’s tables. It also rewrites non-SELECT queries into a SELECT that might be helpful for determining
the non-SELECT statement’s index usage.
-group-by
type: Array; default: fingerprint
Which attribute of the events to group by.
In general, you can group queries into classes based on any attribute of the query, such as user or db, which
will by default show you which users and which databases get the most Query_time. The default attribute,
fingerprint, groups similar, abstracted queries into classes; see below and see also “FINGERPRINTS”.
A report is printed for each --group-by value (unless --no-report is given). Therefore, --group-by
user,db means “report on queries with the same user and report on queries with the same db”; it does not
mean “report on queries with the same user and db.” See also “OUTPUT”.
Every value must have a corresponding value in the same position in --order-by. However, adding values
to --group-by will automatically add values to --order-by, for your convenience.
There are several magical values that cause some extra data mining to happen before the grouping takes place:
fingerprint
This causes events to be fingerprinted to abstract queries into a canonical form, which is then used to
group events together into a class. See “FINGERPRINTS” for more about fingerprinting.
tables
This causes events to be inspected for what appear to be tables, and then aggregated by that. Note
that a query that contains two or more tables will be counted as many times as there are tables; so a
join against two tables will count the Query_time against both tables.
distill
This is a sort of super-fingerprint that collapses queries down into a suggestion of what they do, such
as INSERT SELECT table1 table2.
If parsing memcached input (--type memcached), there are other attributes which you can group by: key_print
(see memcached section in “FINGERPRINTS”), cmd, key, res and val (see memcached section in “AT-
TRIBUTES”).
-help
Show help and exit.
-host
Connect to host.
-ignore-attributes
type: array; default: arg, cmd, insert_id, ip, port, Thread_id, timestamp, exptime, flags, key, res, val, server_id,
offset, end_log_pos, Xid
Do not aggregate these attributes when auto-detecting --select.
If you do not specify --select then pt-query-digest auto-detects and aggregates every attribute that it finds
in the slow log. Some attributes, however, should not be aggregated. This option allows you to specify a list of
attributes to ignore. This only works when no explicit --select is given.
-inherit-attributes
type: array; default: db,ts
If missing, inherit these attributes from the last event that had them.



This option sets which attributes are inherited or carried forward to events which do not have them. For example,
if one event has the db attribute equal to “foo”, but the next event doesn’t have the db attribute, then it inherits
“foo” for its db attribute.
Inheritance is usually desirable, but in some cases it might confuse things. If a query inherits a database that it
doesn’t actually use, then this could confuse --execute.
-interval
type: float; default: .1
How frequently to poll the processlist, in seconds.
-iterations
How many times to iterate through the collect-and-report cycle. If 0, iterate to infinity. Each iteration runs for
--run-time amount of time. An iteration is usually determined by an amount of time and a report is printed
when that amount of time elapses. With --run-time-mode interval, an interval is instead determined
by the interval time you specify with --run-time. See --run-time and --run-time-mode for more
information.
-limit
type: Array; default: 95%:20
Limit output to the given percentage or count.
If the argument is an integer, report only the top N worst queries. If the argument is an integer followed by the
% sign, report that percentage of the worst queries. If the percentage is followed by a colon and another integer,
report the top percentage or the number specified by that integer, whichever comes first.
The value is actually a comma-separated array of values, one for each item in --group-by. If you don’t
specify a value for any of those items, the default is the top 95%.
See also --outliers.
-log
type: string
-mirror
type: float
How often to check whether connections should be moved, depending on read_only. Requires
--processlist and --execute.
This option causes pt-query-digest to check every N seconds whether it is reading from a read-write server and
executing against a read-only server, which is a sensible way to set up two servers if you’re doing something like
master-master replication. The http://code.google.com/p/mysql-master-master/ master-master toolkit does this.
The aim is to keep the passive server ready for failover, which is impossible without putting it under a realistic
workload.
-order-by
type: Array; default: Query_time:sum
Sort events by this attribute and aggregate function.
This is a comma-separated list of order-by expressions, one for each --group-by attribute. The default
Query_time:sum is used for --group-by attributes without explicitly given --order-by attributes
(that is, if you specify more --group-by attributes than corresponding --order-by attributes). The syntax
is attribute:aggregate. See “ATTRIBUTES” for valid attributes. Valid aggregates are:



Aggregate Meaning
========= ============================
sum Sum/total attribute value
min Minimum attribute value
max Maximum attribute value
cnt Frequency/count of the query

For example, the default Query_time:sum means that queries in the query analysis report will be ordered
(sorted) by their total query execution time (“Exec time”). Query_time:max orders the queries by their
maximum query execution time, so the query with the single largest Query_time will be list first. cnt refers
more to the frequency of the query as a whole, how often it appears; “Count” is its corresponding line in the
query analysis report. So any attribute and cnt should yield the same report wherein queries are sorted by the
number of times they appear.
When parsing general logs (--type genlog), the default --order-by becomes Query_time:cnt.
General logs do not report query times so only the cnt aggregate makes sense because all query times are
zero.
If you specify an attribute that doesn’t exist in the events, then pt-query-digest falls back to the default
Query_time:sum and prints a notice at the beginning of the report for each query class. You can create
attributes with --filter and order by them; see “ATTRIBUTES” for an example.
-outliers
type: array; default: Query_time:1:10
Report outliers by attribute:percentile:count.
The syntax of this option is a comma-separated list of colon-delimited strings. The first field is the attribute by
which an outlier is defined. The second is a number that is compared to the attribute’s 95th percentile. The third
is optional, and is compared to the attribute’s cnt aggregate. Queries that pass this specification are added to the
report, regardless of any limits you specified in --limit.
For example, to report queries whose 95th percentile Query_time is at least 60 seconds and which are seen at
least 5 times, use the following argument:
--outliers Query_time:60:5

You can specify an –outliers option for each value in --group-by.
-password
-pid
type: string
-pipeline-profile
Print a profile of the pipeline processes.
-port
-print
Print log events to STDOUT in standard slow-query-log format.



-print-iterations
Print the start time for each --iterations.
This option causes a line like the following to be printed at the start of each --iterations report:
# Iteration 2 started at 2009-11-24T14:39:48.345780

This line will print even if --no-report is specified. If --iterations 0 is specified, each iteration
number will be 0.
-processlist
type: DSN
Poll this DSN’s processlist for queries, with --interval sleep between.
If the connection fails, pt-query-digest tries to reopen it once per second. See also --mirror.
-progress
-read-timeout
Wait this long for an event from the input; 0 to wait forever.
This option sets the maximum time to wait for an event from the input. It applies to all types of input except
--processlist. If an event is not received after the specified time, the script stops reading the input and
prints its reports. If --iterations is 0 or greater than 1, the next iteration will begin, else the script will exit.
This option requires the Perl POSIX module.
-[no]report
default: yes
Print out reports on the aggregate results from --group-by.
This is the standard slow-log analysis functionality. See “OUTPUT” for the description of what this does and
what the results look like.
-report-all
Include all queries, even if they have already been reviewed.
-report-format
type: Array; default: rusage,date,hostname,files,header,profile,query_report,prepared
Print these sections of the query analysis report.
SECTION PRINTS
============ ======================================================
rusage CPU times and memory usage reported by ps
date Current local date and time
hostname Hostname of machine on which :program:‘pt-query-digest‘ was run
files Input files read/parse
header Summary of the entire analysis run
profile Compact table of queries for an overview of the report
query_report Detailed information about each unique query
prepared Prepared statements

The sections are printed in the order specified. The rusage, date, files and header sections are grouped together
if specified together; other sections are separated by blank lines.



See “OUTPUT” for more information on the various parts of the query report.
-report-histogram
type: string; default: Query_time
Chart the distribution of this attribute’s values.
The distribution chart is limited to time-based attributes, so charting Rows_examined, for example, will
produce a useless chart. Charts look like:
# Query_time distribution
# 1us
# 10us
# 100us
# 1ms
# 10ms ################################
# 100ms ################################################################
# 1s ########
# 10s+

A sparkline (see “SPARKLINES”) of the full chart is also printed in the header for each query event. The
sparkline of that full chart is:
# Query_time sparkline: | .^_ |

The sparkline itself is the 8 characters between the pipes (|), one character for each of the 8 buckets (1us, 10us,
etc.) Four character codes are used to represent the approximate relation between each bucket’s value:
_ . - ^

The caret ^ represents peaks (buckets with the most values), and the underscore _ represents lows (buckets with
the least or at least one value). The period . and the hyphen - represent buckets with values between these two
extremes. If a bucket has no values, a space is printed. So in the example above, the period represents the 10ms
bucket, the caret the 100ms bucket, and the underscore the 1s bucket.
See “OUTPUT” for more information.
-review
type: DSN
Store a sample of each class of query in this DSN.
The argument specifies a table to store all unique query fingerprints in. The table must have at least the
following columns. You can add more columns for your own special purposes, but they won’t be used by
pt-query-digest. The following CREATE TABLE definition is also used for --create-review-table.
MAGIC_create_review:
CREATE TABLE query_review (
checksum BIGINT UNSIGNED NOT NULL PRIMARY KEY,
fingerprint TEXT NOT NULL,
first_seen DATETIME,
last_seen DATETIME,
reviewed_by VARCHAR(20),
reviewed_on DATETIME,
comments TEXT
)

The columns are as follows:



COLUMN MEANING
=========== ===============
checksum A 64-bit checksum of the query fingerprint
fingerprint The abstracted version of the query; its primary key
sample The query text of a sample of the class of queries
first_seen The smallest timestamp of this class of queries
last_seen The largest timestamp of this class of queries
reviewed_by Initially NULL; if set, query is skipped thereafter
reviewed_on Initially NULL; not assigned any special meaning
comments Initially NULL; not assigned any special meaning

Note that the fingerprint column is the true primary key for a class of queries. The checksum is just a
cryptographic hash of this value, which provides a shorter value that is very likely to also be unique.
After parsing and aggregating events, your table should contain a row for each fingerprint. This option depends
on --group-by fingerprint (which is the default). It will not work otherwise.
-review-history
type: DSN
The table in which to store historical values for review trend analysis.
Each time you review queries with --review, pt-query-digest will save information into this table so you
can see how classes of queries have changed over time.
This DSN inherits unspecified values from --review. It should mention a table in which to store statistics
about each class of queries. pt-query-digest verifies the existence of the table, and your privileges to insert,
delete and update on that table.
pt-query-digest then inspects the columns in the table. The table must have at least the following columns:
CREATE TABLE query_review_history (
checksum BIGINT UNSIGNED NOT NULL,
sample TEXT NOT NULL
);

Any columns not mentioned above are inspected to see if they follow a certain naming convention. The column
is special if the name ends with an underscore followed by any of these MAGIC_history_cols values:
pct|avt|cnt|sum|min|max|pct_95|stddev|median|rank

If the column ends with one of those values, then the prefix is interpreted as the event attribute to store in that
column, and the suffix is interpreted as the metric to be stored. For example, a column named Query_time_min
will be used to store the minimum Query_time for the class of events. The presence of this column will also add
Query_time to the --select list.
The table should also have a primary key, but that is up to you, depending on how you want to store the historical
data. We suggest adding ts_min and ts_max columns and making them part of the primary key along with the
checksum. But you could also just add a ts_min column and make it a DATE type, so you’d get one row per
class of queries per day.
The default table structure follows. The following MAGIC_create_review_history table definition is used for
--create-review-history-table:
CREATE TABLE query_review_history (
checksum BIGINT UNSIGNED NOT NULL,
ts_min DATETIME,
ts_max DATETIME,
ts_cnt FLOAT,
Query_time_sum FLOAT,



Query_time_min FLOAT,
Query_time_max FLOAT,
Query_time_pct_95 FLOAT,
Query_time_stddev FLOAT,
Query_time_median FLOAT,
Lock_time_sum FLOAT,
Lock_time_min FLOAT,
Lock_time_max FLOAT,
Lock_time_pct_95 FLOAT,
Lock_time_stddev FLOAT,
Lock_time_median FLOAT,
Rows_sent_sum FLOAT,
Rows_sent_min FLOAT,
Rows_sent_max FLOAT,
Rows_sent_pct_95 FLOAT,
Rows_sent_stddev FLOAT,
Rows_sent_median FLOAT,
Rows_examined_sum FLOAT,
Rows_examined_min FLOAT,
Rows_examined_max FLOAT,
Rows_examined_pct_95 FLOAT,
Rows_examined_stddev FLOAT,
Rows_examined_median FLOAT,
-- Percona extended slowlog attributes
-- http://www.percona.com/docs/wiki/patches:slow_extended
Rows_affected_sum FLOAT,
Rows_affected_min FLOAT,
Rows_affected_max FLOAT,
Rows_affected_pct_95 FLOAT,
Rows_affected_stddev FLOAT,
Rows_affected_median FLOAT,
Rows_read_sum FLOAT,
Rows_read_min FLOAT,
Rows_read_max FLOAT,
Rows_read_pct_95 FLOAT,
Rows_read_stddev FLOAT,
Rows_read_median FLOAT,
Merge_passes_sum FLOAT,
Merge_passes_min FLOAT,
Merge_passes_max FLOAT,
Merge_passes_pct_95 FLOAT,
Merge_passes_stddev FLOAT,
Merge_passes_median FLOAT,
InnoDB_IO_r_ops_min FLOAT,
InnoDB_IO_r_ops_max FLOAT,
InnoDB_IO_r_ops_pct_95 FLOAT,
InnoDB_IO_r_ops_stddev FLOAT,
InnoDB_IO_r_ops_median FLOAT,
InnoDB_IO_r_bytes_min FLOAT,
InnoDB_IO_r_bytes_max FLOAT,
InnoDB_IO_r_bytes_pct_95 FLOAT,
InnoDB_IO_r_bytes_stddev FLOAT,
InnoDB_IO_r_bytes_median FLOAT,
InnoDB_IO_r_wait_min FLOAT,
InnoDB_IO_r_wait_max FLOAT,
InnoDB_IO_r_wait_pct_95 FLOAT,
InnoDB_IO_r_wait_stddev FLOAT,
InnoDB_IO_r_wait_median FLOAT,



InnoDB_rec_lock_wait_min FLOAT,
InnoDB_rec_lock_wait_max FLOAT,
InnoDB_rec_lock_wait_pct_95 FLOAT,
InnoDB_rec_lock_wait_stddev FLOAT,
InnoDB_rec_lock_wait_median FLOAT,
InnoDB_queue_wait_min FLOAT,
InnoDB_queue_wait_max FLOAT,
InnoDB_queue_wait_pct_95 FLOAT,
InnoDB_queue_wait_stddev FLOAT,
InnoDB_queue_wait_median FLOAT,
InnoDB_pages_distinct_min FLOAT,
InnoDB_pages_distinct_max FLOAT,
InnoDB_pages_distinct_pct_95 FLOAT,
InnoDB_pages_distinct_stddev FLOAT,
InnoDB_pages_distinct_median FLOAT,
-- Boolean (Yes/No) attributes. Only the cnt and sum are needed for these.
-- cnt is how many times is attribute was recorded and sum is how many of
-- those times the value was Yes. Therefore sum/cnt * 100 = % of recorded
-- times that the value was Yes.
QC_Hit_cnt FLOAT,
QC_Hit_sum FLOAT,
Full_scan_cnt FLOAT,
Full_scan_sum FLOAT,
Full_join_cnt FLOAT,
Full_join_sum FLOAT,
Tmp_table_cnt FLOAT,
Tmp_table_sum FLOAT,
Tmp_table_on_disk_cnt FLOAT,
Tmp_table_on_disk_sum FLOAT,
Filesort_cnt FLOAT,
Filesort_sum FLOAT,
Filesort_on_disk_cnt FLOAT,
Filesort_on_disk_sum FLOAT,
PRIMARY KEY(checksum, ts_min, ts_max)
);

Note that we store the count (cnt) for the ts attribute only; it will be redundant to store this for other attributes.
-run-time
type: time
How long to run for each --iterations. The default is to run forever (you can interrupt with CTRL-C). Be-
cause --iterations defaults to 1, if you only specify --run-time, pt-query-digest runs for that amount
of time and then exits. The two options are specified together to do collect-and-report cycles. For example, spec-
ifying --iterations 4 --run-time 15m with a continuous input (like STDIN or --processlist) will
cause pt-query-digest to run for 1 hour (15 minutes x 4), reporting four times, once at each 15 minute interval.
-run-time-mode
type: string; default: clock
Set what the value of --run-time operates on. Following are the possible values for this option:
clock
--run-time specifies an amount of real clock time during which the tool should run for each
--iterations.
event
--run-time specifies an amount of log time. Log time is determined by timestamps in the log.
The first timestamp seen is remembered, and each timestamp after that is compared to the first to



determine how much log time has passed. For example, if the first timestamp seen is 12:00:00 and
the next is 12:01:30, that is 1 minute and 30 seconds of log time. The tool will read events until
the log time is greater than or equal to the specified --run-time value.
Since timestamps in logs are not always printed, or not always printed frequently, this mode varies in
accuracy.
interval
--run-time specifies interval boundaries of log time into which events are divided and reports
are generated. This mode is different from the others because it doesn’t specify how long to run.
The value of --run-time must be an interval that divides evenly into minutes, hours or days. For
example, 5m divides evenly into hours (60/5=12, so 12 5 minutes intervals per hour) but 7m does not
(60/7=8.6).
Specifying --run-time-mode interval --run-time 30m --iterations 0 is simi-
lar to specifying --run-time-mode clock --run-time 30m --iterations 0. In the
latter case, pt-query-digest will run forever, producing reports every 30 minutes, but this only works
effectively with continuous inputs like STDIN and the processlist. For fixed inputs, like log files,
the former example produces multiple reports by dividing the log into 30 minutes intervals based on
timestamps.
Intervals are calculated from the zeroth second/minute/hour in which a timestamp occurs, not from
whatever time it specifies. For example, with 30 minute intervals and a timestamp of 12:10:30, the
interval is not 12:10:30 to 12:40:30, it is 12:00:00 to 12:29:59. Or, with 1 hour intervals,
it is 12:00:00 to 12:59:59. When a new timestamp exceeds the interval, a report is printed, and
the next interval is recalculated based on the new timestamp.
Since --iterations is 1 by default, you probably want to specify a new value else pt-query-
digest will only get and report on the first interval from the log since 1 interval = 1 iteration. If you
want to get and report every interval in a log, specify --iterations 0.
-sample
type: int
Filter out all but the first N occurrences of each query. The queries are filtered on the first value in
--group-by, so by default, this will filter by query fingerprint. For example, --sample 2 will permit
two sample queries for each fingerprint. Useful in conjunction with --print to print out the queries. You
probably want to set --no-report to avoid the overhead of aggregating and reporting if you’re just using this
to print out samples of queries. A complete example:
:program:‘pt-query-digest‘ --sample 2 --no-report --print slow.log

-select
type: Array
Compute aggregate statistics for these attributes.
By default pt-query-digest auto-detects, aggregates and prints metrics for every query attribute that it finds in
the slow query log. This option specifies a list of only the attributes that you want. You can specify an alternative
attribute with a colon. For example, db:Schema uses db if it’s available, and Schema if it’s not.
Previously, pt-query-digest only aggregated these attributes:
Query_time,Lock_time,Rows_sent,Rows_examined,user,db:Schema,ts

Attributes specified in the --review-history table will always be selected even if you do not specify
--select.
See also --ignore-attributes and “ATTRIBUTES”.



-set-vars
executed.
-shorten
Shorten long statements in reports.
Shortens long statements, replacing the omitted portion with a /*... omitted ...*/ comment. This
applies only to the output in reports, not to information stored for --review or other places. It prevents
a large statement from causing difficulty in a report. The argument is the preferred length of the shortened
statement. Not all statements can be shortened, but very large INSERT and similar statements often can; and so
can IN() lists, although only the first such list in the statement will be shortened.
If it shortens something beyond recognition, you can find the original statement in the log, at the offset shown
in the report header (see “OUTPUT”).
-show-all
type: Hash
Show all values for these attributes.
By default pt-query-digest only shows as many of an attribute’s value that fit on a single line. This option
allows you to specify attributes for which all values will be shown (line width is ignored). This only works for
attributes with string values like user, host, db, etc. Multiple attributes can be specified, comma-separated.
-since
type: string
Parse only queries newer than this value (parse queries since this date).
This option allows you to ignore queries older than a certain value and parse only those queries which are more
recent than the value. The value can be several types:
* Simple time value N with optional suffix: N[shmd], where
s=seconds, h=hours, m=minutes, d=days (default s if no suffix
given); this is like saying "since N[shmd] ago"
* Full date with optional hours:minutes:seconds:
YYYY-MM-DD [HH:MM::SS]
* Short, MySQL-style date:
YYMMDD [HH:MM:SS]
* Any time expression evaluated by MySQL:
CURRENT_DATE - INTERVAL 7 DAY

If you give a MySQL time expression, then you must also specify a DSN so that pt-query-digest can con-
nect to MySQL to evaluate the expression. If you specify --execute, --explain, --processlist,
--review or --review-history, then one of these DSNs will be used automatically. Otherwise, you
must specify an --aux-dsn or pt-query-digest will die saying that the value is invalid.
The MySQL time expression is wrapped inside a query like “SELECT UNIX_TIMESTAMP(<expression>)”,
so be sure that the expression is valid inside this query. For example, do not use UNIX_TIMESTAMP() because
UNIX_TIMESTAMP(UNIX_TIMESTAMP()) returns 0.
Events are assumed to be in chronological: older events at the beginning of the log and newer events at the end
of the log. --since is strict: it ignores all queries until one is found that is new enough. Therefore, if the
query events are not consistently timestamped, some may be ignored which are actually new enough.
See also --until.



-socket
-statistics
Print statistics about internal counters. This option is mostly for development and debugging. The statistics
report is printed for each iteration after all other reports, even if no events are processed or --no-report is
specified. The statistics report looks like:
# No events processed.

# Statistic Count %/Events
# ================================================ ====== ========
# events_read 142030 100.00
# events_parsed 50430 35.51
# events_aggregated 0 0.00
# ignored_midstream_server_response 18111 12.75
# no_tcp_data 91600 64.49
# pipeline_restarted_after_MemcachedProtocolParser 142030 100.00
# pipeline_restarted_after_TcpdumpParser 1 0.00
# unknown_client_command 1 0.00
# unknown_client_data 32318 22.75

The first column is the internal counter name; the second column is counter’s count; and the third column is the
count as a percentage of events_read.
In this case, it shows why no events were processed/aggregated: 100% of events were rejected by the
MemcachedProtocolParser. Of those, 35.51% were data packets, but of these 12.75% of ignored mid-
stream server response, one was an unknown client command, and 22.75% were unknown client data. The other
64.49% were TCP control packets (probably most ACKs).
Since pt-query-digest is complex, you will probably need someone familiar with its code to decipher the statis-
tics report.
-table-access
Print a table access report.
The table access report shows which tables are accessed by all the queries and if the access is a read or write.
The report looks like:
write ‘baz‘.‘tbl‘
read ‘baz‘.‘new_tbl‘
write ‘baz‘.‘tbl3‘
write ‘db6‘.‘tbl6‘

If you pipe the output to sort, the read and write tables will be grouped together and sorted alphabetically:
read ‘baz‘.‘new_tbl‘
write ‘baz‘.‘tbl‘
write ‘baz‘.‘tbl3‘
write ‘db6‘.‘tbl6‘

-tcpdump-errors
type: string
Write the tcpdump data to this file on error. If pt-query-digest doesn’t parse the stream correctly for some
reason, the session’s packets since the last query event will be written out to create a usable test case. If this
happens, pt-query-digest will not raise an error; it will just discard the session’s saved state and permit the tool
to continue working. See “tcpdump” for more information about parsing tcpdump output.



-timeline
Show a timeline of events.
This option makes pt-query-digest print another kind of report: a timeline of the events. Each query is still
grouped and aggregate into classes according to --group-by, but then they are printed in chronological
order. The timeline report prints out the timestamp, interval, count and value of each classes.
If all you want is the timeline report, then specify --no-report to suppress the default query analy-
sis report. Otherwise, the timeline report will be printed at the end before the response-time profile (see
--report-format and “OUTPUT”).
For example, this:
:program:‘pt-query-digest‘ /path/to/log --group-by distill --timeline

will print something like:
# ########################################################
# distill report
# ########################################################
# 2009-07-25 11:19:27 1+00:00:01 2 SELECT foo
# 2009-07-27 11:19:30 00:01 2 SELECT bar
# 2009-07-27 11:30:00 1+06:30:00 2 SELECT foo

-type
type: Array
The type of input to parse (default slowlog). The permitted types are
binlog
Parse a binary log file.
genlog
Parse a MySQL general log file. General logs lack a lot of “ATTRIBUTES”, notably Query_time.
The default --order-by for general logs changes to Query_time:cnt.
http
Parse HTTP traffic from tcpdump.
pglog
Parse a log file in PostgreSQL format. The parser will automatically recognize logs sent to syslog
and transparently parse the syslog format, too. The recommended configuration for logging in your
postgresql.conf is as follows.
The log_destination setting can be set to either syslog or stderr. Syslog has the added benefit of not
interleaving log messages from several sessions concurrently, which the parser cannot handle, so this
might be better than stderr. CSV-formatted logs are not supported at this time.
The log_min_duration_statement setting should be set to 0 to capture all statements with their dura-
tions. Alternatively, the parser will also recognize and handle various combinations of log_duration
and log_statement.
You may enable log_connections and log_disconnections, but this is optional.
It is highly recommended to set your log_line_prefix to the following:
log_line_prefix = ’%m c=%c,u=%u,D=%d ’

This lets the parser find timestamps with milliseconds, session IDs, users, and databases from the
log. If these items are missing, you’ll simply get less information to analyze. For compatibility with



other log analysis tools such as PQA and pgfouine, various log line prefix formats are supported. The
general format is as follows: a timestamp can be detected and extracted (the syslog timestamp is NOT
parsed), and a name=value list of properties can also. Although the suggested format is as shown
above, any name=value list will be captured and interpreted by using the first letter of the ‘name’
part, lowercased, to determine the meaning of the item. The lowercased first letter is interpreted
to mean the same thing as PostgreSQL’s built-in %-codes for the log_line_prefix format string. For
example, u means user, so unicorn=fred will be interpreted as user=fred; d means database, so D=john
will be interpreted as database=john. The pgfouine-suggested formatting is user=%u and db=%d, so
it should Just Work regardless of which format you choose. The main thing is to add as much
information as possible into the log_line_prefix to permit richer analysis.
Currently, only English locale messages are supported, so if your server’s locale is set to something
else, the log won’t be parsed properly. (Log messages with “duration:” and “statement:” won’t be
recognized.)
slowlog
Parse a log file in any variation of MySQL slow-log format.
tcpdump
Inspect network packets and decode the MySQL client protocol, extracting queries and responses
from it.
pt-query-digest does not actually watch the network (i.e. it does NOT “sniff packets”). Instead, it’s
just parsing the output of tcpdump. You are responsible for generating this output; pt-query-digest
does not do it for you. Then you send this to pt-query-digest as you would any log file: as files on
the command line or to STDIN.
The parser expects the input to be formatted with the following options: -x -n -q -tttt. For
example, if you want to capture output from your local machine, you can do something like the
following (the port must come last on FreeBSD):
tcpdump -s 65535 -x -nn -q -tttt -i any -c 1000 port 3306
> mysql.tcp.txt
:program:‘pt-query-digest‘ --type tcpdump mysql.tcp.txt

The other tcpdump parameters, such as -s, -c, and -i, are up to you. Just make sure the output looks
like this (there is a line break in the first line to avoid man-page problems):
2009-04-12 09:50:16.804849 IP 127.0.0.1.42167
> 127.0.0.1.3306: tcp 37
0x0000: 4508 0059 6eb2 4000 4006 cde2 7f00 0001
0x0010: ....

Remember tcpdump has a handy -c option to stop after it captures some number of packets! That’s
very useful for testing your tcpdump command. Note that tcpdump can’t capture traffic on a Unix
socket. Read http://bugs.mysql.com/bug.php?id=31577 if you’re confused about this.
Devananda Van Der Veen explained on the MySQL Performance Blog how to capture traffic without
dropping packets on busy servers. Dropped packets cause pt-query-digest to miss the response to
a request, then see the response to a later request and assign the wrong execution time to the query.
You can change the filter to something like the following to help capture a subset of the queries. (See
http://www.mysqlperformanceblog.com/?p=6092 for details.)
tcpdump -i any -s 65535 -x -n -q -tttt
’port 3306 and tcp[1] & 7 == 2 and tcp[3] & 7 == 2’

All MySQL servers running on port 3306 are automatically detected in the tcpdump output.
Therefore, if the tcpdump out contains packets from multiple servers on port 3306 (for example,



10.0.0.1:3306, 10.0.0.2:3306, etc.), all packets/queries from all these servers will be analyzed to-
gether as if they were one server.
If you’re analyzing traffic for a MySQL server that is not running on port 3306, see
--watch-server.
Also note that pt-query-digest may fail to report the database for queries when parsing tcpdump out-
put. The database is discovered only in the initial connect events for a new client or when <USE db>
is executed. If the tcpdump output contains neither of these, then pt-query-digest cannot discover
the database.
Server-side prepared statements are supported. SSL-encrypted traffic cannot be inspected and de-
coded.
memcached
Similar to tcpdump, but the expected input is memcached packets instead of MySQL packets. For
example:
tcpdump -i any port 11211 -s 65535 -x -nn -q -tttt
> memcached.tcp.txt
:program:‘pt-query-digest‘ --type memcached memcached.tcp.txt

memcached uses port 11211 by default.
-until
type: string
Parse only queries older than this value (parse queries until this date).
This option allows you to ignore queries newer than a certain value and parse only those queries which are older
than the value. The value can be one of the same types listed for --since.
Unlike --since, --until is not strict: all queries are parsed until one has a timestamp that is equal to or
greater than --until. Then all subsequent queries are ignored.
-user
-variations
type: Array
Report the number of variations in these attributes’ values.
Variations show how many distinct values an attribute had within a class. The usual value for this option is arg
which shows how many distinct queries were in the class. This can be useful to determine a query’s cacheability.
Distinct values are determined by CRC32 checksums of the attributes’ values. These checksums are reported in
the query report for attributes specified by this option, like:
# arg crc 109 (1/25%), 144 (1/25%)... 2 more

In that class there were 4 distinct queries. The checksums of the first two variations are shown, and each one
occurred once (or, 25% of the time).
The counts of distinct variations is approximate because only 1,000 variations are saved. The mod (%) 1000 of
the full CRC32 checksum is saved, so some distinct checksums are treated as equal.
-version



-watch-server
type: string
This option tells pt-query-digest which server IP address and port (like “10.0.0.1:3306”) to watch when parsing
tcpdump (for --type tcpdump and memcached); all other servers are ignored. If you don’t specify it, pt-
query-digest watches all servers by looking for any IP address using port 3306 or “mysql”. If you’re watching
a server with a non-standard port, this won’t work, so you must specify the IP address and port to watch.
If you want to watch a mix of servers, some running on standard port 3306 and some running on non-standard
ports, you need to create separate tcpdump outputs for the non-standard port servers and then specify this option
for each. At present pt-query-digest cannot auto-detect servers on port 3306 and also be told to watch a server
on a non-standard port.
-[no]zero-admin
default: yes
Zero out the Rows_XXX properties for administrator command events.
-[no]zero-bool
default: yes
Print 0% boolean values in report.

2.21.12 DSN OPTIONS

details.
• A
• D
Database that contains the query review table.
• F
• h
Connect to host.
• p
• P



• S
• t
Table to use as the query review table.
• u

2.21.13 ENVIRONMENT

PTDEBUG=1 pt-query-digest ... > FILE 2>&1



Perl.

2.21.15 BUGS

For a list of known bugs, see http://www.percona.com/bugs/pt-query-digest.
• Tool --version

2.21.16 DOWNLOADING








2.21.17 AUTHORS






2.21.20 VERSION

pt-query-digest 2.1.1

2.22 pt-show-grants

2.22.1 NAME

pt-show-grants - Canonicalize and print MySQL grants so you can effectively replicate, compare and version-control
them.

2.22.2 SYNOPSIS

Usage

pt-show-grants [OPTION...] [DSN]

pt-show-grants shows grants (user privileges) from a MySQL server.



Examples

pt-show-grants

pt-show-grants --separate --revoke | diff othergrants.sql -

2.22.3 RISKS

pt-show-grants is read-only by default, and very low-risk. If you specify --flush, it will execute FLUSH
PRIVILEGES.
show-grants.

2.22.4 DESCRIPTION

pt-show-grants extracts, orders, and then prints grants for MySQL user accounts.
Why would you want this? There are several reasons.
The first is to easily replicate users from one server to another; you can simply extract the grants from the first server
and pipe the output directly into another server.
The second use is to place your grants into version control. If you do a daily automated grant dump into version
control, you’ll get lots of spurious changesets for grants that don’t change, because MySQL prints the actual grants
out in a seemingly random order. For instance, one day it’ll say
GRANT DELETE, INSERT, UPDATE ON ‘test‘.* TO ’foo’@’%’;

And then another day it’ll say
GRANT INSERT, DELETE, UPDATE ON ‘test‘.* TO ’foo’@’%’;

The grants haven’t changed, but the order has. This script sorts the grants within the line, between ‘GRANT’ and
‘ON’. If there are multiple rows from SHOW GRANTS, it sorts the rows too, except that it always prints the row
with the user’s password first, if it exists. This removes three kinds of inconsistency you’ll get from running SHOW
GRANTS, and avoids spurious changesets in version control.
Third, if you want to diff grants across servers, it will be hard without “canonicalizing” them, which pt-show-grants
does. The output is fully diff-able.
With the --revoke, --separate and other options, pt-show-grants also makes it easy to revoke specific privi-
leges from users. This is tedious otherwise.

2.22.5 OPTIONS


2.22. pt-show-grants 175


-ask-pass
-charset
-config
type: Array
-database
-defaults-file
-drop
Add DROP USER before each user in the output.
-flush
Add FLUSH PRIVILEGES after output.
You might need this on pre-4.1.1 servers if you want to drop a user completely.
-[no]header
default: yes
Print dump header.
The header precedes the dumped grants. It looks like:
-- Grants dumped by :program:‘pt-show-grants‘ 1.0.19
-- Dumped from server Localhost via UNIX socket, MySQL 5.0.82-log at 2009-10-26 10:01:04

See also --[no]timestamp.
-help
Show help and exit.
-host
Connect to host.
-ignore
type: array
Ignore this comma-separated list of users.
-only
type: array
Only show grants for this comma-separated list of users.
-password



-pid
type: string
-port
-revoke
Add REVOKE statements for each GRANT statement.
-separate
List each GRANT or REVOKE separately.
The default output from MySQL’s SHOW GRANTS command lists many privileges on a single line. With
--flush, places a FLUSH PRIVILEGES after each user, instead of once at the end of all the output.
-set-vars
executed.
-socket
-[no]timestamp
default: yes
Add timestamp to the dump header.
See also --[no]header.
-user
-version

2.22.6 DSN OPTIONS

details.
• A



• D
Default database.
• F
• h
Connect to host.
• p
• P
• S
• u

2.22.7 ENVIRONMENT

PTDEBUG=1 pt-show-grants ... > FILE 2>&1



Perl.

2.22.9 BUGS

For a list of known bugs, see http://www.percona.com/bugs/pt-show-grants.



• Tool --version

2.22.10 DOWNLOADING






2.22.11 AUTHORS

Baron Schwartz




welcome.



2.22.14 VERSION

pt-show-grants 2.1.1

2.23 pt-sift

2.23.1 NAME

pt-sift - Browses files created by pt-collect.

2.23.2 SYNOPSIS

Usage

pt-sift FILE|PREFIX|DIRECTORY

pt-sift browses the files created by pt-collect. If you specify a FILE or PREFIX, it browses only files with that prefix.
If you specify a DIRECTORY, then it browses all files within that directory.

2.23.3 RISKS

pt-sift is a read-only tool. It should be very low-risk.
will be marked as such. You can see a list of such issues at the following URL: http://www.percona.com/bugs/pt-sift.

2.23.4 DESCRIPTION

pt-sift downloads other tools that it might need, such as pt-diskstats, and then makes a list of the unique timestamp
prefixes of all the files in the directory, as written by the pt-collect tool. If the user specified a timestamp on the
command line, then it begins with that sample of data; otherwise it begins by showing a list of the timestamps and
prompting for a selection. Thereafter, it displays a summary of the selected sample, and the user can navigate and
inspect with keystrokes. The keystroke commands you can use are as follows:
d
Sets the action to start the pt-diskstats tool on the sample’s disk performance statistics.
i
Sets the action to view the first INNODB STATUS sample in less.
m
Displays the first 4 samples of SHOW STATUS counters side by side with the pt-mext tool.
n



Summarizes the first sample of netstat data in two ways: by originating host, and by connection state.
j
Select the next timestamp as the active sample.
k
Select the previous timestamp as the active sample.
q
Quit the program.
1
Sets the action for each sample to the default, which is to view a summary of the sample.
0
Sets the action to just list the files in the sample.
• Sets the action to view all of the sample’s files in the less program.

2.23.5 OPTIONS

This tool does not have any command-line options.

2.23.6 ENVIRONMENT



This tool requires Bash v3 and the following programs: pt-diskstats, pt-pmp, pt-mext, and pt-align. If these programs
are not in your PATH, they will be fetched from the Internet if curl is available.

2.23.8 BUGS

For a list of known bugs, see http://www.percona.com/bugs/pt-sift.
• Tool --version

2.23. pt-sift 181


2.23.9 DOWNLOADING






2.23.10 AUTHORS

Baron Schwartz




welcome.

2.23.13 VERSION

pt-sift 2.1.1



2.24 pt-slave-delay

2.24.1 NAME

pt-slave-delay - Make a MySQL slave server lag behind its master.

2.24.2 SYNOPSIS

Usage

pt-slave-delay [OPTION...] SLAVE-HOST [MASTER-HOST]

pt-slave-delay starts and stops a slave server as needed to make it lag behind the master. The SLAVE-HOST and
MASTER-HOST use DSN syntax, and values are copied from the SLAVE-HOST to the MASTER-HOST if omitted.
To hold slavehost one minute behind its master for ten minutes:
pt-slave-delay --delay 1m --interval 15s --run-time 10m slavehost

2.24.3 RISKS

pt-slave-delay is generally very low-risk. It simply starts and stops the replication SQL thread. This might cause
monitoring systems to think the slave is having trouble.
slave-delay.

2.24.4 DESCRIPTION

pt-slave-delay watches a slave and starts and stops its replication SQL thread as necessary to hold it at least as
far behind the master as you request. In practice, it will typically cause the slave to lag between --delay and
--delay”+”--interval behind the master.
It bases the delay on binlog positions in the slave’s relay logs by default, so there is no need to connect to the master.
This works well if the IO thread doesn’t lag the master much, which is typical in most replication setups; the IO thread
lag is usually milliseconds on a fast network. If your IO thread’s lag is too large for your purposes, pt-slave-delay can
also connect to the master for information about binlog positions.
If the slave’s I/O thread reports that it is waiting for the SQL thread to free some relay log space, pt-slave-delay will
automatically connect to the master to ﬁnd binary log positions. If --ask-pass and --daemonize are given, it
is possible that this could cause it to ask for a password while daemonized. In this case, it exits. Therefore, if you
think your slave might encounter this condition, you should be sure to either specify --use-master explicitly when
daemonizing, or don’t specify --ask-pass.
The SLAVE-HOST and optional MASTER-HOST are both DSNs. See “DSN OPTIONS”. Missing MASTER-HOST
values are ﬁlled in with values from SLAVE-HOST, so you don’t need to specify them in both places. pt-slave-delay

2.24. pt-slave-delay 183


reads all normal MySQL option ﬁles, such as ~/.my.cnf, so you may not need to specify username, password and other
common options at all.
pt-slave-delay tries to exit gracefully by trapping signals such as Ctrl-C. You cannot bypass --[no]continue with
a trappable signal.

2.24.5 PRIVILEGES

pt-slave-delay requires the following privileges: PROCESS, REPLICATION CLIENT, and SUPER.

2.24.6 OUTPUT

If you specify --quiet, there is no output. Otherwise, the normal output is a status message consisting of a timestamp
and information about what pt-slave-delay is doing: starting the slave, stopping the slave, or just observing.

2.24.7 OPTIONS

-ask-pass
-charset
-config
type: Array
-[no]continue
default: yes
Continue replication normally on exit. After exiting, restart the slave’s SQL thread with no UNTIL condition,
so it will run as usual and catch up to the master. This is enabled by default and works even if you terminate
pt-slave-delay with Control-C.
-daemonize
-defaults-file
-delay
type: time; default: 1h
How far the slave should lag its master.
-help
Show help and exit.



-host
Connect to host.
-interval
type: time; default: 1m
How frequently pt-slave-delay should check whether the slave needs to be started or stopped.
-log
type: string
-password
-pid
type: string
-port
-quiet
short form: -q
Don’t print informational messages about operation. See OUTPUT for details.
-run-time
type: time
How long pt-slave-delay should run before exiting. The default is to run forever.
-set-vars
executed.
-socket
-use-master
Get binlog positions from master, not slave. Don’t trust the binlog positions in the slave’s relay log. Connect
to the master and get binlog positions instead. If you specify this option without giving a MASTER-HOST on
the command line, pt-slave-delay examines the slave’s SHOW SLAVE STATUS to determine the hostname and
port for connecting to the master.
pt-slave-delay uses only the MASTER_HOST and MASTER_PORT values from SHOW SLAVE STATUS for
the master connection. It does not use the MASTER_USER value. If you want to specify a different username
for the master than the one you use to connect to the slave, you should specify the MASTER-HOST option
explicitly on the command line.



-user
-version

2.24.8 DSN OPTIONS

details.
• A
• D
Default database.
• F
• h
Connect to host.
• p
• P
• S
• u



2.24.9 ENVIRONMENT

PTDEBUG=1 pt-slave-delay ... > FILE 2>&1



Perl.

2.24.11 BUGS

For a list of known bugs, see http://www.percona.com/bugs/pt-slave-delay.
• Tool --version

2.24.12 DOWNLOADING






2.24.13 AUTHORS

Sergey Zhuravlev and Baron Schwartz






This program is copyright 2007-2011 Sergey Zhuravle and Baron Schwartz, 2011-2012 Percona Inc. Feedback and
improvements are welcome.

2.24.16 VERSION

pt-slave-delay 2.1.1

2.25 pt-slave-find

2.25.1 NAME

pt-slave-find - Find and print replication hierarchy tree of MySQL slaves.

2.25.2 SYNOPSIS

Usage

pt-slave-find [OPTION...] MASTER-HOST

pt-slave-find finds and prints a hierarchy tree of MySQL slaves.

Examples

pt-slave-find --host master-host



2.25.3 RISKS

pt-slave-find is read-only and very low-risk.
slave-find.

2.25.4 DESCRIPTION

pt-slave-find connects to a MySQL replication master and finds its slaves. Currently the only thing it can do is print a
tree-like view of the replication hierarchy.
The master host can be specified using one of two methods. The first method is to use the standard connection-related
command line options: --defaults-file, --password, --host, --port, --socket or --user.
The second method to specify the master host is a DSN. A DSN is a special syntax that can be either just a hostname
(like server.domain.com or 1.2.3.4), or a key=value,key=value string. Keys are a single letter:
KEY MEANING
=== =======
h Connect to host
P Port number to use for connection
S Socket file to use for connection
u User for login if not current user
p Password to use when connecting
F Only read default options from the given file

pt-slave-find reads all normal MySQL option files, such as ~/.my.cnf, so you may not need to specify username,
password and other common options at all.

2.25.5 EXIT STATUS

An exit status of 0 (sometimes also called a return value or return code) indicates success. Any other value represents
the exit status of the Perl process itself.

2.25.6 OPTIONS

-ask-pass
-charset

2.25. pt-slave-find 189


-config
type: Array
-database
type: string; short form: -D
Database to use.
-defaults-file
-help
Show help and exit.
-host
Connect to host.
-password
-pid
type: string
-port
-recurse
type: int
Number of levels to recurse in the hierarchy. Default is infinite.
See --recursion-method.
-recursion-method
type: string
METHOD USES
=========== ================

is required if the server uses a non-standard port (not 3306). Usually pt-slave-find does the right thing and finds



-report-format
type: string; default: summary
Set what information about the slaves is printed. The report format can be one of the following:
•hostname

Print just the hostname name of the slaves. It looks like:
127.0.0.1:12345
+- 127.0.0.1:12346
+- 127.0.0.1:12347

•summary

Print a summary of each slave’s settings. This report shows more information about each slave, like:
127.0.0.1:12345
Version 5.1.34-log
Server ID 12345
Uptime 04:56 (started 2010-06-17T11:21:22)
Replication Is not a slave, has 1 slaves connected
Filters
Binary logging STATEMENT
Slave status
Slave mode STRICT
Auto-increment increment 1, offset 1
+- 127.0.0.1:12346
Version 5.1.34-log
Server ID 12346
Uptime 04:54 (started 2010-06-17T11:21:24)
Replication Is a slave, has 1 slaves connected
Filters
Binary logging STATEMENT
Slave status 0 seconds behind, running, no errors
Slave mode STRICT
Auto-increment increment 1, offset 1

-set-vars
executed.
-socket
-user
-version



2.25.7 DSN OPTIONS

details.
• A
• D
Default database.
• F
• h
Connect to host.
• p
• P
• S
• u

2.25.8 ENVIRONMENT

PTDEBUG=1 pt-slave-find ... > FILE 2>&1





Perl.

2.25.10 BUGS

For a list of known bugs, see http://www.percona.com/bugs/pt-slave-ﬁnd.
• Tool --version

2.25.11 DOWNLOADING






2.25.12 AUTHORS







welcome.

2.25.15 VERSION

pt-slave-ﬁnd 2.1.1

2.26 pt-slave-restart

2.26.1 NAME

pt-slave-restart - Watch and restart MySQL replication after errors.

2.26.2 SYNOPSIS

Usage

pt-slave-restart [OPTION...] [DSN]

pt-slave-restart watches one or more MySQL replication slaves for errors, and tries to restart replication if it stops.

2.26.3 RISKS

pt-slave-restart is a brute-force way to try to keep a slave server running when it is having problems with replication.
Don’t be too hasty to use it unless you need to. If you use this tool carelessly, you might miss the chance to really
solve the slave server’s problems.
At the time of this release there is a bug that causes an invalid CHANGE MASTER TO statement to be executed.
slave-restart.



2.26.4 DESCRIPTION

pt-slave-restart watches one or more MySQL replication slaves and tries to skip statements that cause errors. It polls
slaves intelligently with an exponentially varying sleep time. You can specify errors to skip and run the slaves until a
certain binlog position.
Note: it has come to my attention that Yahoo! had or has an internal tool called fix_repl, described to me by a past
Yahoo! employee and mentioned in the first edition of High Performance MySQL. Apparently this tool does the same
thing. Make no mistake, though: this is not a way to “fix replication.” In fact I would not even encourage its use on a
regular basis; I use it only when I have an error I know I just need to skip past.

2.26.5 OUTPUT

If you specify --verbose, pt-slave-restart prints a line every time it sees the slave has an error. See --verbose
for details.

2.26.6 SLEEP

pt-slave-restart sleeps intelligently between polling the slave. The current sleep time varies.
• The initial sleep time is given by --sleep.
• If it checks and finds an error, it halves the previous sleep time.
• If it finds no error, it doubles the previous sleep time.
• The sleep time is bounded below by --min-sleep and above by --max-sleep.
• Immediately after finding an error, pt-slave-restart assumes another error is very likely to happen next, so it
sleeps the current sleep time or the initial sleep time, whichever is less.

2.26.7 EXIT STATUS

An exit status of 0 (sometimes also called a return value or return code) indicates success. Any other value represents
the exit status of the Perl process itself, or of the last forked process that exited if there were multiple servers to
monitor.

2.26.8 COMPATIBILITY

pt-slave-restart should work on many versions of MySQL. Lettercase of many output columns from SHOW SLAVE
STATUS has changed over time, so it treats them all as lowercase.

2.26.9 OPTIONS

-always
Start slaves even when there is no error. With this option enabled, pt-slave-restart will not let you stop the slave
manually if you want to!
-ask-pass

2.26. pt-slave-restart 195


-charset
-[no]check-relay-log
default: yes
Check the last relay log file and position before checking for slave errors.
By default pt-slave-restart will not doing anything (it will just sleep) if neither the relay log file nor the relay
log position have changed since the last check. This prevents infinite loops (i.e. restarting the same error in the
same relay log file at the same relay log position).
For certain slave errors, however, this check needs to be disabled by specifying --no-check-relay-log.
Do not do this unless you know what you are doing!
-config
type: Array
-daemonize
-database
Database to use.
-defaults-file
-error-length
type: int
Max length of error message to print. When --verbose is set high enough to print the error, this option will
truncate the error text to the specified length. This can be useful to prevent wrapping on the terminal.
-error-numbers
type: hash
Only restart this comma-separated list of errors. Makes pt-slave-restart only try to restart if the error number
is in this comma-separated list of errors. If it sees an error not in the list, it will exit.
The error number is in the last_errno column of SHOW SLAVE STATUS.
-error-text
type: string
Only restart errors that match this pattern. A Perl regular expression against which the error text, if any, is
matched. If the error text exists and matches, pt-slave-restart will try to restart the slave. If it exists but doesn’t
match, pt-slave-restart will exit.
The error text is in the last_error column of SHOW SLAVE STATUS.
-help
Show help and exit.



-host
Connect to host.
-log
type: string
-max-sleep
type: float; default: 64
Maximum sleep seconds.
The maximum time pt-slave-restart will sleep before polling the slave again. This is also the time that pt-slave-
restart will wait for all other running instances to quit if both --stop and --monitor are specified.
See “SLEEP”.
-min-sleep
The minimum time pt-slave-restart will sleep before polling the slave again. See “SLEEP”.
-monitor
Whether to monitor the slave (default). Unless you specify –monitor explicitly, --stop will disable it.
-password
-pid
type: string
-port
-quiet
short form: -q
Suppresses normal output (disables --verbose).
-recurse
Watch slaves of the specified server, up to the specified number of servers deep in the hierarchy. The default
depth of 0 means “just watch the slave specified.”
pt-slave-restart examines SHOW PROCESSLIST and tries to determine which connections are from slaves,
then connect to them. See --recursion-method.
Recursion works by finding all slaves when the program starts, then watching them. If there is more than one
slave, pt-slave-restart uses fork() to monitor them.
This also works if you have configured your slaves to show up in SHOW SLAVE HOSTS. The minimal config-
uration for this is the report_host parameter, but there are other “report” parameters as well for the port,
username, and password.



-recursion-method
type: string
METHOD USES
=========== ================

is required if the server uses a non-standard port (not 3306). Usually pt-slave-restart does the right thing and
finds the slaves, but you may give a preferred method and it will be used first. If it doesn’t find any slaves, the
other methods will be tried.
-run-time
type: time
Time to run before exiting. Causes pt-slave-restart to stop after the specified time has elapsed. Optional suffix:
s=seconds, m=minutes, h=hours, d=days; if no suffix, s is used.
-sentinel
type: string; default: /tmp/pt-slave-restart-sentinel
-set-vars
executed.
-skip-count
Number of statements to skip when restarting the slave.
-sleep
Initial sleep seconds between checking the slave.
See “SLEEP”.
-socket
-stop
Causes pt-slave-restart to create the sentinel file specified by --sentinel. This should have the effect of
stopping all running instances which are watching the same sentinel file. If --monitor isn’t specified, pt-
slave-restart will exit after creating the file. If it is specified, pt-slave-restart will wait the interval given by
--max-sleep, then remove the file and continue working.
You might find this handy to stop cron jobs gracefully if necessary, or to replace one running instance with
another. For example, if you want to stop and restart pt-slave-restart every hour (just to make sure that it is
restarted every hour, in case of a server crash or some other problem), you could use a crontab line like this:



0 * * * * :program:‘pt-slave-restart‘ --monitor --stop --sentinel /tmp/pt-slave-restartup

The non-default --sentinel will make sure the hourly cron job stops only instances previously started with
the same options (that is, from the same cron job).
See also --sentinel.
-until-master
type: string
Run until this master log file and position. Start the slave, and retry if it fails, until it reaches the given repli-
cation coordinates. The coordinates are the logfile and position on the master, given by relay_master_log_file,
exec_master_log_pos. The argument must be in the format “file,pos”. Separate the filename and position with a
single comma and no space.
This will also cause an UNTIL clause to be given to START SLAVE.
After reaching this point, the slave should be stopped and pt-slave-restart will exit.
-until-relay
type: string
Run until this relay log file and position. Like --until-master, but in the slave’s relay logs instead. The
coordinates are given by relay_log_file, relay_log_pos.
-user
-verbose
Be verbose; can specify multiple times. Verbosity 1 outputs connection information, a timestamp, relay_log_file,
relay_log_pos, and last_errno. Verbosity 2 adds last_error. See also --error-length. Verbosity 3 prints
the current sleep time each time pt-slave-restart sleeps.
-version

2.26.10 DSN OPTIONS

details.
• A
• D
Default database.
• F



• h
Connect to host.
• p
• P
• S
• u

2.26.11 ENVIRONMENT

PTDEBUG=1 pt-slave-restart ... > FILE 2>&1



Perl.

2.26.13 BUGS

For a list of known bugs, see http://www.percona.com/bugs/pt-slave-restart.
• Tool --version




2.26.14 DOWNLOADING






2.26.15 AUTHORS

Baron Schwartz




welcome.

2.26.18 VERSION

pt-slave-restart 2.1.1



2.27 pt-stalk

2.27.1 NAME

pt-stalk - Gather forensic data about MySQL when a problem occurs.

2.27.2 SYNOPSIS

Usage

pt-stalk [OPTIONS] [-- MYSQL OPTIONS]

pt-stalk watches for a trigger condition to become true, and then collects data to help in diagnosing problems. It
is designed to run as a daemon with root privileges, so that you can diagnose intermittent problems that you cannot
observe directly. You can also use it to execute a custom command, or to gather the data on demand without waiting
for the trigger to happen.

2.27.3 RISKS

pt-stalk is a read-write tool; it collects data from the system and writes it into a series of files. It should be very
low-risk. Some of the options can cause intrusive data collection to be performed, however, so if you enable any
non-default options, you should read their documentation carefully.
will be marked as such. You can see a list of such issues at the following URL: http://www.percona.com/bugs/pt-stalk.

2.27.4 DESCRIPTION

Sometimes a problem happens infrequently and for a short time, giving you no chance to see the system when it
happens. How do you solve intermittent MySQL problems when you can’t observe them? That’s why pt-stalk exists.
In addition to using it when there’s a known problem on your servers, it is a good idea to run pt-stalk all the time, even
when you think nothing is wrong. You will appreciate the data it gathers when a problem occurs, because problems
such as MySQL lockups or spikes of activity typically leave no evidence to use in root cause analysis.
This tool does two things: it watches a server (typically MySQL) for a trigger to occur, and it gathers diagnostic data.
To use it effectively, you need to define a good trigger condition. A good trigger is sensitive enough to fire reliably
when a problem occurs, so that you don’t miss a chance to solve problems. On the other hand, a good trigger isn’t
prone to false positives, so you don’t gather information when the server is functioning normally.
The most reliable triggers for MySQL tend to be the number of connections to the server, and the number of queries
running concurrently. These are available in the SHOW GLOBAL STATUS command as Threads_connected and
Threads_running. Sometimes Threads_connected is not a reliable indicator of trouble, but Threads_running usually
is. Your job, as the tool’s user, is to define an appropriate trigger condition for the tool. Choose carefully, because the
quality of your results will depend on the trigger you choose.



You can define the trigger with the --function, --variable, and --threshold options, among others. Please
read the documentation for –function to learn how to do this.
The pt-stalk tool, by default, simply watches MySQL repeatedly until the trigger becomes true. It then gathers
diagnostics for a while, and sleeps afterwards for some time to prevent repeatedly gathering data if the condition
remains true. In crude pseudocode, omitting some subtleties,
while true; do
if --variable from --function is greater than --threshold; then
observations++
if observations is greater than --cycles; then
capture diagnostics for --run-time seconds
exit if --iterations is exceeded
sleep for --sleep seconds
done
done
clean up data that’s older than --retention-time
sleep for --interval seconds
done

The diagnostic data is written to files whose names begin with a timestamp, so you can distinguish samples from each
other in case the tool collects data multiple times. The pt-sift tool is designed to help you browse and analyze the
resulting samples of data.
Although this sounds simple enough, in practice there are a number of subtleties, such as detecting when the disk is
beginning to fill up so that the tool doesn’t cause the server to run out of disk space. This tool handles these types
of potential problems, so it’s a good idea to use this tool instead of writing something from scratch and possibly
experiencing some of the hazards this tool is designed to prevent.

2.27.5 CONFIGURING

You can use standard Percona Toolkit configuration files to set commandline options.
You will probably want to run the tool as a daemon and customize at least the diagnostic threshold. Here’s a sample
configuration file for triggering when there are more than 20 queries running at once:
daemonize
threshold=20

If you’re not running the tool as it’s designed (as a root user, daemonized) then you’ll need to set several options, such
as --dest, to locations that are writable by non-root users.

2.27.6 OPTIONS

-collect
default: yes; negatable: yes
Collect system information. You can negate this option to make the tool watch the system but not actually gather
any diagnostic data.
See also --stalk.
-collect-gdb
Collect GDB stacktraces. This is achieved by attaching to MySQL and printing stack traces from all threads.
This will freeze the server for some period of time, ranging from a second or so to much longer on very busy
systems with a lot of memory and many threads in the server. For this reason, it is disabled by default. However,
if you are trying to diagnose a server stall or lockup, freezing the server causes no additional harm, and the stack
traces can be vital for diagnosis.

2.27. pt-stalk 203


In addition to freezing the server, there is also some risk of the server crashing or performing badly after GDB
detaches from it.
-collect-oprofile
Collect oprofile data. This is achieved by starting an oprofile session, letting it run for the collection time, and
then stopping and saving the resulting profile data in the system’s default location. Please read your system’s
oprofile documentation to learn more about this.
-collect-strace
Collect strace data. This is achieved by attaching strace to the server, which will make it run very slowly until
strace detaches. The same cautions apply as those listed in –collect-gdb. You should not enable this option
together with –collect-gdb, because GDB and strace can’t attach to the server process simultaneously.
-collect-tcpdump
Collect tcpdump data. This option causes tcpdump to capture all traffic on all interfaces for the port on which
MySQL is listening. You can later use pt-query-digest to decode the MySQL protocol and extract a log of query
traffic from it.
-config
type: string
-cycles
The number of times the trigger condition must be true before collecting data. This helps prevent false positives,
and makes the trigger condition less likely to fire when the problem recovers quickly.
-daemonize
Daemonize the tool. This causes the tool to fork into the background and log its output as specified in –log.
-dest
type: string; default: /var/lib/pt-stalk
Where to store the diagnostic data. Each time the tool collects data, it writes to a new set of files, which are
named with the current system timestamp.
-disk-bytes-free
type: size; default: 100M
Don’t collect data if the disk has less than this much free space. This prevents the tool from filling up the disk
with diagnostic data.
If the --dest directory contains a previously captured sample of data, the tool will measure its size and use
that as an estimate of how much data is likely to be gathered this time, too. It will then be even more pessimistic,
and will refuse to collect data unless the disk has enough free space to hold the sample and still have the desired
amount of free space. For example, if you’d like 100MB of free space and the previous diagnostic sample
consumed 100MB, the tool won’t collect any data unless the disk has 200MB free.
Valid size value suffixes are k, M, G, and T.
-disk-pct-free
Don’t collect data if the disk has less than this percent free space. This prevents the tool from filling up the disk
with diagnostic data.
This option works similarly to --disk-bytes-free but specifies a percentage margin of safety instead of
a bytes margin of safety. The tool honors both options, and will not collect any data unless both margins are
satisfied.



-function
type: string; default: status
Specifies what to watch for a diagnostic trigger. The default value watches SHOW GLOBAL STATUS, but
you can also watch SHOW PROCESSLIST or supply a plugin file with your own custom code. This function
supplies the value of --variable, which is then compared against --threshold to see if the trigger
condition is met. Additional options may be required as well; see below. Possible values:
•status

This value specifies that the source of data for the diagnostic trigger is SHOW GLOBAL STATUS.
The value of --variable then defines which status counter is the trigger.

•processlist

This value specifies that the data for the diagnostic trigger comes from SHOW FULL PROCESSLIST.
The trigger value is the count of processes whose --variable column matches the --match
option. For example, to trigger when more than 10 processes are in the “statistics” state, use the
following options:
--function processlist --variable State
--match statistics --threshold 10

In addition, you can specify a file that contains your custom trigger function, written in Unix shell script. This
can be a wrapper that executes anything you wish. If the argument to –function is a file, then it takes precedence
over builtin functions, so if there is a file in the working directory named “status” or “processlist” then the tool
will use that file as a plugin, even though those are otherwise recognized as reserved words for this option.
The plugin file works by providing a function called trg_plugin, and the tool simply sources the file and
executes the function. For example, the function might look like the following:
trg_plugin() {
mysql $EXT_ARGV -e "SHOW ENGINE INNODB STATUS"
| grep -c "has waited at"
}

This snippet will count the number of mutex waits inside of InnoDB. It illustrates the general principle: the
function must output a number, which is then compared to the threshold as usual. The $EXT_ARGV variable
contains the MySQL options mentioned in the “SYNOPSIS” above.
The plugin should not alter the tool’s existing global variables. Prefix any plugin-specific global variables with
“PLUGIN_” or make them local.
-help
-interval
Interval between checks for the diagnostic trigger.
-iterations
type: int
Exit after collecting diagnostics this many times. By default, the tool will continue to watch the server forever,
but this is useful for scenarios where you want to capture once and then exit, for example.
-log
type: string; default: /var/log/pt-stalk.log

2.27. pt-stalk 205


-match
type: string
The pattern to use when watching SHOW PROCESSLIST. See the documentation for --function for details.
-notify-by-email
type: string
Send mail to this list of addresses when data is collected.
-pid
type: string; default: /var/run/pt-stalk.pid
Create a PID file when daemonized.
-prefix
type: string
The filename prefix for diagnostic samples. By default, samples have a timestamp prefix based on the current
local time, such as 2011_12_06_14_02_02, which is December 6, 2011 at 14:02:02.
-retention-time
Number of days to retain collected samples. Any samples that are older will be purged.
-run-time
How long the tool will collect data when it triggers. This should not be longer than --sleep. It is usually not
necessary to change this; if the default 30 seconds hasn’t gathered enough diagnostic data, running longer is not
likely to do so. In fact, in many cases a shorter collection period is appropriate.
-sleep
How long to sleep after collecting data. This prevents the tool from triggering continuously, which might be a
problem if the collection process is intrusive. It also prevents filling up the disk or gathering too much data to
analyze reasonably.
-stalk
Watch the server and wait for the trigger to occur. You can negate this option to make the tool immediately
gather any diagnostic data once and exit. This is useful if a problem is already happening, but pt-stalk is not
running, so you only want to collect diagnostic data.
If this option is negate, --daemonize, --log, --pid, and other stalking-related options have no ef-
fect; the tool simply collects diagnostic data and exits. Safeguard options, like --disk-bytes-free and
--disk-pct-free, are still respected.
See also --collect.
-threshold
The threshold at which the diagnostic trigger should fire. See --function for details.
-variable
type: string; default: Threads_running
The variable to compare against the threshold. See --function for details.
-version



2.27.7 ENVIRONMENT

This tool does not use any environment variables for conﬁguration.


This tool requires Bash v3 or newer.

2.27.9 BUGS

For a list of known bugs, see http://www.percona.com/bugs/pt-stalk.
• Tool --version

2.27.10 DOWNLOADING






2.27.11 AUTHORS

Baron Schwartz, Justin Swanhart, Fernando Ipar, and Daniel Nichter



2.27. pt-stalk 207



welcome.

2.27.14 VERSION

pt-stalk 2.1.1

2.28 pt-summary

2.28.1 NAME

pt-summary - Summarize system information nicely.

2.28.2 SYNOPSIS

Usage

pt-summary

pt-summary conveniently summarizes the status and conﬁguration of a server. It is not a tuning tool or diagnosis tool.
It produces a report that is easy to diff and can be pasted into emails without losing the formatting. This tool works
well on many types of Unix systems.
Download and run:
wget http://percona.com/get/pt-summary
bash ./pt-summary

2.28.3 RISKS

pt-summary is a read-only tool. It should be very low-risk.



summary.

2.28.4 DESCRIPTION

pt-summary runs a large variety of commands to inspect system status and configuration, saves the output into files
in a temporary directory, and then runs Unix commands on these results to format them nicely. It works best when
executed as a privileged user, but will also work without privileges, although some output might not be possible to
generate without root.

2.28.5 OUTPUT

Many of the outputs from this tool are deliberately rounded to show their magnitude but not the exact detail. This
is called fuzzy-rounding. The idea is that it doesn’t matter whether a particular counter is 918 or 921; such a small
variation is insignificant, and only makes the output hard to compare to other servers. Fuzzy-rounding rounds in larger
increments as the input grows. It begins by rounding to the nearest 5, then the nearest 10, nearest 25, and then repeats
by a factor of 10 larger (50, 100, 250), and so on, as the input grows.
The following is a simple report generated from a CentOS virtual machine, broken into sections with commentary
following each section. Some long lines are reformatted for clarity when reading this documentation as a manual page
in a terminal.
# Percona Toolkit System Summary Report ######################
Date | 2012-03-30 00:58:07 UTC (local TZ: EDT -0400)
Hostname | localhost.localdomain
Uptime | 20:58:06 up 1 day, 20 min, 1 user,
load average: 0.14, 0.18, 0.18
System | innotek GmbH; VirtualBox; v1.2 ()
Service Tag | 0
Platform | Linux
Release | CentOS release 5.5 (Final)
Kernel | 2.6.18-194.el5
Architecture | CPU = 32-bit, OS = 32-bit
Threading | NPTL 2.5
Compiler | GNU CC version 4.1.2 20080704 (Red Hat 4.1.2-48).
SELinux | Enforcing
Virtualized | VirtualBox

This section shows the current date and time, and a synopsis of the server and operating system.
# Processor ##################################################
Processors | physical = 1, cores = 0, virtual = 1, hyperthreading = no
Speeds | 1x2510.626
Models | 1xIntel(R) Core(TM) i5-2400S CPU @ 2.50GHz
Caches | 1x6144 KB

This section is derived from /proc/cpuinfo.
# Memory #####################################################
Total | 503.2M
Free | 29.0M
Used | physical = 474.2M, swap allocated = 1.0M,
swap used = 16.0k, virtual = 474.3M

2.28. pt-summary 209


Buffers | 33.9M
Caches | 262.6M
Dirty | 396 kB
UsedRSS | 201.9M
Swappiness | 60
DirtyPolicy | 40, 10
Locator Size Speed Form Factor Type Type Detail
======= ==== ===== =========== ==== ===========

Information about memory is gathered from free. The Used statistic is the total of the rss sizes displayed by ps. The
Dirty statistic for the cached value comes from /proc/meminfo. On Linux, the swappiness settings are gathered from
sysctl. The final portion of this section is a table of the DIMMs, which comes from dmidecode. In this example
there is no output.
# Mounted Filesystems ########################################
Filesystem Size Used Type Opts Mountpoint
/dev/mapper/VolGroup00-LogVol00 15G 17% ext3 rw /
/dev/sda1 99M 13% ext3 rw /boot
tmpfs 252M 0% tmpfs rw /dev/shm

The mounted filesystem section is a combination of information from mount and df. This section is skipped if you
disable --summarize-mounts.
# Disk Schedulers And Queue Size #############################
dm-0 | UNREADABLE
dm-1 | UNREADABLE
hdc | [cfq] 128
md0 | UNREADABLE
sda | [cfq] 128

The disk scheduler information is extracted from the /sys filesystem in Linux.
# Disk Partioning ############################################
Device Type Start End Size
============ ==== ========== ========== ==================
/dev/sda Disk 17179869184
/dev/sda1 Part 1 13 98703360
/dev/sda2 Part 14 2088 17059230720

Information about disk partitioning comes from fdisk -l.
# Kernel Inode State #########################################
dentry-state | 10697 8559 45 0 0 0
file-nr | 960 0 50539
inode-nr | 14059 8139

These lines are from the files of the same name in the /proc/sys/fs directory on Linux. Read the proc man page to
learn about the meaning of these files on your system.
# LVM Volumes ################################################
LV VG Attr LSize Origin Snap% Move Log Copy% Convert
LogVol00 VolGroup00 -wi-ao 269.00G
LogVol01 VolGroup00 -wi-ao 9.75G

This section shows the output of lvs.
# RAID Controller ############################################
Controller | No RAID controller detected



The tool can detect a variety of RAID controllers by examining lspci and dmesg information. If the controller
software is installed on the system, in many cases it is able to execute status commands and show a summary of the
RAID controller’s status and configuration. If your system is not supported, please file a bug report.
# Network Config #############################################
Controller | Intel Corporation 82540EM Gigabit Ethernet Controller
FIN Timeout | 60
Port Range | 61000

The network controllers attached to the system are detected from lspci. The TCP/IP protocol configuration param-
eters are extracted from sysctl. You can skip this section by disabling the --summarize-network option.
# Interface Statistics #######################################
interface rx_bytes rx_packets rx_errors tx_bytes tx_packets tx_errors
========= ======== ========== ========= ======== ========== =========
lo 60000000 12500 0 60000000 12500 0
eth0 15000000 80000 0 1500000 10000 0
sit0 0 0 0 0 0 0

Interface statistics are gathered from ip -s link and are fuzzy-rounded. The columns are received and transmitted
bytes, packets, and errors. You can skip this section by disabling the --summarize-network option.
# Network Connections ########################################
Connections from remote IP addresses
127.0.0.1 2
Connections to local IP addresses
127.0.0.1 2
Connections to top 10 local ports
38346 1
60875 1
States of connections
ESTABLISHED 5
LISTEN 8

This section shows a summary of network connections, retrieved from netstat and “fuzzy-rounded” to make them
easier to compare when the numbers grow large. There are two sub-sections showing how many connections there
are per origin and destination IP address, and a sub-section showing the count of ports in use. The section ends with
the count of the network connections’ states. You can skip this section by disabling the --summarize-network
option.
# Top Processes ##############################################
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
1 root 15 0 2072 628 540 S 0.0 0.1 0:02.55 init
2 root RT -5 0 0 0 S 0.0 0.0 0:00.00 migration/0
3 root 34 19 0 0 0 S 0.0 0.0 0:00.03 ksoftirqd/0
4 root RT -5 0 0 0 S 0.0 0.0 0:00.00 watchdog/0
5 root 10 -5 0 0 0 S 0.0 0.0 0:00.97 events/0
6 root 10 -5 0 0 0 S 0.0 0.0 0:00.00 khelper
7 root 10 -5 0 0 0 S 0.0 0.0 0:00.00 kthread
10 root 10 -5 0 0 0 S 0.0 0.0 0:00.13 kblockd/0
11 root 20 -5 0 0 0 S 0.0 0.0 0:00.00 kacpid
# Notable Processes ##########################################
PID OOM COMMAND
2028 +0 sshd

This section shows the first few lines of top so that you can see what processes are actively using CPU time. The
notable processes include the SSH daemon and any process whose out-of-memory-killer priority is set to 17. You can
skip this section by disabling the --summarize-processes option.



# Simplified and fuzzy rounded vmstat (wait please) ##########
procs ---swap-- -----io---- ---system---- --------cpu--------
r b si so bi bo ir cs us sy il wa st
2 0 0 0 3 15 30 125 0 0 99 0 0
0 0 0 0 0 0 1250 800 6 10 84 0 0
0 0 0 0 0 0 1000 125 0 0 100 0 0
0 0 0 0 0 0 1000 125 0 0 100 0 0
0 0 0 0 0 450 1000 125 0 1 88 11 0
# The End ####################################################

This section is a trimmed-down sample of vmstat 1 5, so you can see the general status of the system at present.
The values in the table are fuzzy-rounded, except for the CPU columns. You can skip this section by disabling the
--summarize-processes option.

2.28.6 OPTIONS

-config
type: string
-help
-save-samples
type: string
Save the collected data in this directory.
-read-samples
type: string
Create a report from the files in this directory.
-summarize-mounts
Report on mounted filesystems and disk usage.
-summarize-network
Report on network controllers and configuration.
-summarize-processes
Report on top processes and vmstat output.
-sleep
How long to sleep when gathering samples from vmstat.
-version


This tool requires the Bourne shell (/bin/sh).



2.28.8 BUGS

For a list of known bugs, see http://www.percona.com/bugs/pt-summary.
• Tool --version

2.28.9 DOWNLOADING






2.28.10 AUTHORS

Baron Schwartz and Kevin van Zonneveld (http://kevin.vanzonneveld.net)




welcome.




2.28.13 VERSION

pt-summary 2.1.1

2.29 pt-table-checksum

2.29.1 NAME

pt-table-checksum - Verify MySQL replication integrity.

2.29.2 SYNOPSIS

Usage

pt-table-checksum [OPTION...] [DSN]

pt-table-checksum performs an online replication consistency check by executing checksum queries on the master,
which produces different results on replicas that are inconsistent with the master. The optional DSN specifies the
master host. The tool’s exit status is nonzero if any differences are found, or if any warnings or errors occur.
The following command will connect to the replication master on localhost, checksum every table, and report the
results on every detected replica:
pt-table-checksum

This tool is focused on finding data differences efficiently. If any data is different, you can resolve the problem with
pt-table-sync.

2.29.3 RISKS

pt-table-checksum can add load to the MySQL server, although it has many safeguards to prevent this. It inserts a
small amount of data into a table that contains checksum results. It has checks that, if disabled, can potentially cause
replication to fail when unsafe replication options are used. In short, it is safe by default, but it permits you to turn off
its safety checks.
The tool presumes that schemas and tables are identical on the master and all replicas. Replication will break if, for
example, a replica does not have a schema that exists on the master (and that schema is checksummed), or if the
structure of a table on a replica is different than on the master.
At the time of this release, we know of no bugs that could cause harm to users.



will be marked as such. You can see a list of such issues at the following URL: http://www.percona.com/bugs/pt-table-
checksum.

2.29.4 DESCRIPTION

pt-table-checksum is designed to do the right thing by default in almost every case. When in doubt, use --explain
to see how the tool will checksum a table. The following is a high-level overview of how the tool functions.
In contrast to older versions of pt-table-checksum, this tool is focused on a single purpose, and does not have a lot
of complexity or support many different checksumming techniques. It executes checksum queries on only one server,
and these flow through replication to re-execute on replicas. If you need the older behavior, you can use Percona
Toolkit version 1.0.
pt-table-checksum connects to the server you specify, and finds databases and tables that match the filters you specify
(if any). It works one table at a time, so it does not accumulate large amounts of memory or do a lot of work before
beginning to checksum. This makes it usable on very large servers. We have used it on servers with hundreds of
thousands of databases and tables, and trillions of rows. No matter how large the server is, pt-table-checksum works
equally well.
One reason it can work on very large tables is that it divides each table into chunks of rows, and checksums each chunk
with a single REPLACE..SELECT query. It varies the chunk size to make the checksum queries run in the desired
amount of time. The goal of chunking the tables, instead of doing each table with a single big query, is to ensure that
checksums are unintrusive and don’t cause too much replication lag or load on the server. That’s why the target time
for each chunk is 0.5 seconds by default.
The tool keeps track of how quickly the server is able to execute the queries, and adjusts the chunks as it learns more
about the server’s performance. It uses an exponentially decaying weighted average to keep the chunk size stable,
yet remain responsive if the server’s performance changes during checksumming for any reason. This means that the
tool will quickly throttle itself if your server becomes heavily loaded during a traffic spike or a background task, for
example.
Chunking is accomplished by a technique that we used to call “nibbling” in other tools in Percona Toolkit. It is the
same technique used for pt-archiver, for example. The legacy chunking algorithms used in older versions of pt-table-
checksum are removed, because they did not result in predictably sized chunks, and didn’t work well on many tables.
All that is required to divide a table into chunks is an index of some sort (preferably a primary key or unique index). If
there is no index, and the table contains a suitably small number of rows, the tool will checksum the table in a single
chunk.
pt-table-checksum has many other safeguards to ensure that it does not interfere with any server’s operation, including
replicas. To accomplish this, pt-table-checksum detects replicas and connects to them automatically. (If this fails,
you can give it a hint with the --recursion-method option.)
The tool monitors replicas continually. If any replica falls too far behind in replication, pt-table-checksum pauses to
allow it to catch up. If any replica has an error, or replication stops, pt-table-checksum pauses and waits. In addition,
pt-table-checksum looks for common causes of problems, such as replication filters, and refuses to operate unless you
force it to. Replication filters are dangerous, because the queries that pt-table-checksum executes could potentially
conflict with them and cause replication to fail.
pt-table-checksum verifies that chunks are not too large to checksum safely. It performs an EXPLAIN query on each
chunk, and skips chunks that might be larger than the desired number of rows. You can configure the sensitivity of
this safeguard with the --chunk-size-limit option. If a table will be checksummed in a single chunk because
it has a small number of rows, then pt-table-checksum additionally verifies that the table isn’t oversized on replicas.
This avoids the following scenario: a table is empty on the master but is very large on a replica, and is checksummed
in a single large query, which causes a very long delay in replication.

2.29. pt-table-checksum 215


There are several other safeguards. For example, pt-table-checksum sets its session-level innodb_lock_wait_timeout
to 1 second, so that if there is a lock wait, it will be the victim instead of causing other queries to time out. Another
safeguard checks the load on the database server, and pauses if the load is too high. There is no single right answer for
how to do this, but by default pt-table-checksum will pause if there are more than 25 concurrently executing queries.
You should probably set a sane value for your server with the --max-load option.
Checksumming usually is a low-priority task that should yield to other work on the server. However, a tool that must
be restarted constantly is difficult to use. Thus, pt-table-checksum is very resilient to errors. For example, if the
database administrator needs to kill pt-table-checksum‘s queries for any reason, that is not a fatal error. Users often
run pt-kill to kill any long-running checksum queries. The tool will retry a killed query once, and if it fails again, it will
move on to the next chunk of that table. The same behavior applies if there is a lock wait timeout. The tool will print
a warning if such an error happens, but only once per table. If the connection to any server fails, pt-table-checksum
will attempt to reconnect and continue working.
If pt-table-checksum encounters a condition that causes it to stop completely, it is easy to resume it with the
--resume option. It will begin from the last chunk of the last table that it processed. You can also safely stop
the tool with CTRL-C. It will finish the chunk it is currently processing, and then exit. You can resume it as usual
afterwards.
After pt-table-checksum finishes checksumming all of the chunks in a table, it pauses and waits for all detected
replicas to finish executing the checksum queries. Once that is finished, it checks all of the replicas to see if they have
the same data as the master, and then prints a line of output with the results. You can see a sample of its output later in
this documentation.
The tool prints progress indicators during time-consuming operations. It prints a progress indicator as each table is
checksummed. The progress is computed by the estimated number of rows in the table. It will also print a progress
report when it pauses to wait for replication to catch up, and when it is waiting to check replicas for differences from
the master. You can make the output less verbose with the --quiet option.
If you wish, you can query the checksum tables manually to get a report of which tables and chunks have differences
from the master. The following query will report every database and table with differences, along with a summary of
the number of chunks and rows possibly affected:
SELECT db, tbl, SUM(this_cnt) AS total_rows, COUNT(*) AS chunks
FROM percona.checksums
WHERE (
master_cnt <> this_cnt
OR master_crc <> this_crc
OR ISNULL(master_crc) <> ISNULL(this_crc))
GROUP BY db, tbl;

The table referenced in that query is the checksum table, where the checksums are stored. Each row in the table
contains the checksum of one chunk of data from some table in the server.
Version 2.0 of pt-table-checksum is not backwards compatible with pt-table-sync version 1.0. In some cases this is not
a serious problem. Adding a “boundaries” column to the table, and then updating it with a manually generated WHERE
clause, may suffice to let pt-table-sync version 1.0 interoperate with pt-table-checksum version 2.0. Assuming an
integer primary key named ‘id’, You can try something like the following:
ALTER TABLE checksums ADD boundaries VARCHAR(500);
UPDATE checksums
SET boundaries = COALESCE(CONCAT(’id BETWEEN ’, lower_boundary,
’ AND ’, upper_boundary), ’1=1’);

2.29.5 OUTPUT

The tool prints tabular results, one line per table:



TS ERRORS DIFFS ROWS CHUNKS SKIPPED TIME TABLE
10-20T08:36:50 0 0 200 1 0 0.005 db1.tbl1
10-20T08:36:50 0 0 603 7 0 0.035 db1.tbl2
10-20T08:36:50 0 0 16 1 0 0.003 db2.tbl3
10-20T08:36:50 0 0 600 6 0 0.024 db2.tbl4

Errors, warnings, and progress reports are printed to standard error. See also --quiet.
Each table’s results are printed when the tool finishes checksumming the table. The columns are as follows:
TS The timestamp (without the year) when the tool finished checksumming the table.
ERRORS The number of errors and warnings that occurred while checksumming the table. Errors and warnings are
printed to standard error while the table is in progress.
DIFFS The number of chunks that differ from the master on one or more replicas. If --no-replicate-check is
specified, this column will always have zeros. If --replicate-check-only is specified, then only tables
with differences are printed.
ROWS The number of rows selected and checksummed from the table. It might be different from the number of rows
in the table if you use the –where option.
CHUNKS The number of chunks into which the table was divided.
SKIPPED The number of chunks that were skipped due to errors or warnings, or because they were oversized.
TIME The time elapsed while checksumming the table.
TABLE The database and table that was checksummed.
If --replicate-check-only is specified, only checksum differences on detected replicas are printed. The
output is different: one paragraph per replica, one checksum difference per line, and values are separted by spaces:
Differences on h=127.0.0.1,P=12346
TABLE CHUNK CNT_DIFF CRC_DIFF CHUNK_INDEX LOWER_BOUNDARY UPPER_BOUNDARY
db1.tbl1 1 0 1 PRIMARY 1 100

Differences on h=127.0.0.1,P=12347
TABLE CHUNK CNT_DIFF CRC_DIFF CHUNK_INDEX LOWER_BOUNDARY UPPER_BOUNDARY

The first line of a paragraph indicates the replica with differences. In this example there are two: h=127.0.0.1,P=12346
and h=127.0.0.1,P=12347. The columns are as follows:
TABLE The database and table that differs from the master.
CHUNK The chunk number of the table that differs from the master.
CNT_DIFF The number of chunk rows on the replica minus the number of chunk rows on the master.
CRC_DIFF 1 if the CRC of the chunk on the replica is different than the CRC of the chunk on the master, else 0.
CHUNK_INDEX The index used to chunk the table.
LOWER_BOUNDARY The index values that define the lower boundary of the chunk.
UPPER_BOUNDARY The index values that define the upper boundary of the chunk.

2.29.6 EXIT STATUS

A non-zero exit status indicates errors, warnings, or checksum differences.



2.29.7 OPTIONS

-ask-pass
group: Connection
-check-interval
type: time; default: 1; group: Throttle
Sleep time between checks for --max-lag.
-[no]check-replication-filters
default: yes; group: Safety
Do not checksum if any replication filters are set on any replicas. The tool looks for server options that filter
replication, such as binlog_ignore_db and replicate_do_db. If it finds any such filters, it aborts with an error.
If the replicas are configured with any filtering options, you should be careful not to checksum any databases or
tables that exist on the master and not the replicas. Changes to such tables might normally be skipped on the
replicas because of the filtering options, but the checksum queries modify the contents of the table that stores
the checksums, not the tables whose data you are checksumming. Therefore, these queries will be executed on
the replica, and if the table or database you’re checksumming does not exist, the queries will cause replication
to fail. For more information on replication rules, see http://dev.mysql.com/doc/en/replication-rules.html.
Replication filtering makes it impossible to be sure that the checksum queries won’t break replication (or simply
fail to replicate). If you are sure that it’s OK to run the checksum queries, you can negate this option to disable
the checks. See also --replicate-database.
-check-slave-lag
type: string; group: Throttle
Pause checksumming until this replica’s lag is less than --max-lag. The value is a DSN that inherits proper-
ties from the master host and the connection options (--port, --user, etc.). This option overrides the normal
behavior of finding and continually monitoring replication lag on ALL connected replicas. If you don’t want to
monitor ALL replicas, but you want more than just one replica to be monitored, then use the DSN option to the
--recursion-method option instead of this option.
-chunk-index
type: string
Prefer this index for chunking tables. By default, pt-table-checksum chooses the most appropriate index for
chunking. This option lets you specify the index that you prefer. If the index doesn’t exist, then pt-table-
checksum will fall back to its default behavior of choosing an index. pt-table-checksum adds the index to the
checksum SQL statements in a FORCE INDEX clause. Be careful when using this option; a poor choice of
index could cause bad performance. This is probably best to use when you are checksumming only a single
table, not an entire server.
-chunk-size
type: size; default: 1000
Number of rows to select for each checksum query. Allowable suffixes are k, M, G.
This option can override the default behavior, which is to adjust chunk size dynamically to try to make chunks
run in exactly --chunk-time seconds. When this option isn’t set explicitly, its default value is used as a
starting point, but after that, the tool ignores this option’s value. If you set this option explicitly, however, then
it disables the dynamic adjustment behavior and tries to make all chunks exactly the specified number of rows.
There is a subtlety: if the chunk index is not unique, then it’s possible that chunks will be larger than desired.
For example, if a table is chunked by an index that contains 10,000 of a given value, there is no way to write a



WHERE clause that matches only 1,000 of the values, and that chunk will be at least 10,000 rows large. Such a
chunk will probably be skipped because of --chunk-size-limit.
-chunk-size-limit
type: float; default: 2.0; group: Safety
Do not checksum chunks this much larger than the desired chunk size.
When a table has no unique indexes, chunk sizes can be inaccurate. This option specifies a maximum tolerable
limit to the inaccuracy. The tool uses <EXPLAIN> to estimate how many rows are in the chunk. If that estimate
exceeds the desired chunk size times the limit (twice as large, by default), then the tool skips the chunk.
The minimum value for this option is 1, which means that no chunk can be larger than --chunk-size. You
probably don’t want to specify 1, because rows reported by EXPLAIN are estimates, which can be different
from the real number of rows in the chunk. If the tool skips too many chunks because they are oversized, you
might want to specify a value larger than the default of 2.
You can disable oversized chunk checking by specifying a value of 0.
-chunk-time
Adjust the chunk size dynamically so each checksum query takes this long to execute.
The tool tracks the checksum rate (rows per second) for all tables and each table individually. It uses these rates
to adjust the chunk size after each checksum query, so that the next checksum query takes this amount of time
(in seconds) to execute.
The algorithm is as follows: at the beginning of each table, the chunk size is initialized from the overall average
rows per second since the tool began working, or the value of --chunk-size if the tool hasn’t started working
yet. For each subsequent chunk of a table, the tool adjusts the chunk size to try to make queries run in the desired
amount of time. It keeps an exponentially decaying moving average of queries per second, so that if the server’s
performance changes due to changes in server load, the tool adapts quickly. This allows the tool to achieve
predictably timed queries for each table, and for the server overall.
If this option is set to zero, the chunk size doesn’t auto-adjust, so query checksum times will vary, but query
checksum sizes will not. Another way to do the same thing is to specify a value for --chunk-size explicitly,
instead of leaving it at the default.
-columns
short form: -c; type: array; group: Filter
Checksum only this comma-separated list of columns.
-config
type: Array; group: Config
-[no]create-replicate-table
default: yes
Create the --replicate database and table if they do not exist. The structure of the replicate table is the
same as the suggested table mentioned in --replicate.
-databases
short form: -d; type: hash; group: Filter
Only checksum this comma-separated list of databases.
-databases-regex
type: string; group: Filter
Only checksum databases whose names match this Perl regex.



-defaults-file
short form: -F; type: string; group: Connection
-[no]empty-replicate-table
default: yes
Delete previous checksums for each table before checksumming the table. This option does not truncate the
entire table, it only deletes rows (checksums) for each table just before checksumming the table. Therefore, if
checksumming stops prematurely and there was preexisting data, there will still be rows for tables that were not
checksummed before the tool was stopped.
If you’re resuming from a previous checksum run, then the checksum records for the table from which the tool
resumes won’t be emptied.
-engines
short form: -e; type: hash; group: Filter
Only checksum tables which use these storage engines.
-explain
cumulative: yes; default: 0; group: Output
Show, but do not execute, checksum queries (disables --[no]empty-replicate-table). If specified
twice, the tool actually iterates through the chunking algorithm, printing the upper and lower boundary values
for each chunk, but not executing the checksum queries.
-float-precision
type: int
Precision for FLOAT and DOUBLE number-to-string conversion. Causes FLOAT and DOUBLE values to be
rounded to the specified number of digits after the decimal point, with the ROUND() function in MySQL.
This can help avoid checksum mismatches due to different floating-point representations of the same values on
different MySQL versions and hardware. The default is no rounding; the values are converted to strings by the
CONCAT() function, and MySQL chooses the string representation. If you specify a value of 2, for example,
then the values 1.008 and 1.009 will be rounded to 1.01, and will checksum as equal.
-function
type: string
Hash function for checksums (FNV1A_64, MURMUR_HASH, SHA1, MD5, CRC32, etc).
The default is to use CRC32(), but MD5() and SHA1() also work, and you can use your own function, such as
a compiled UDF, if you wish. The function you specify is run in SQL, not in Perl, so it must be available to
MySQL.
MySQL doesn’t have good built-in hash functions that are fast. CRC32() is too prone to hash collisions, and
MD5() and SHA1() are very CPU-intensive. The FNV1A_64() UDF that is distributed with Percona Server is a
faster alternative. It is very simple to compile and install; look at the header in the source code for instructions. If
it is installed, it is preferred over MD5(). You can also use the MURMUR_HASH() function if you compile and
install that as a UDF; the source is also distributed with Percona Server, and it might be better than FNV1A_64().
-help
group: Help
Show help and exit.
-host
short form: -h; type: string; default: localhost; group: Connection
Host to connect to.



-ignore-columns
type: Hash; group: Filter
Ignore this comma-separated list of columns when calculating the checksum.
-ignore-databases
-ignore-databases-regex
Ignore databases whose names match this Perl regex.
-ignore-engines
type: Hash; default: FEDERATED,MRG_MyISAM; group: Filter
-ignore-tables
Ignore this comma-separated list of tables. Table names may be qualified with the database name. The
--replicate table is always automatically ignored.
-ignore-tables-regex
Ignore tables whose names match the Perl regex.
-lock-wait-timeout
Set the session value of innodb_lock_wait_timeout on the master host. This option helps guard against
long lock waits if the checksum queries become slow for some reason. Setting this option dynamically requires
the InnoDB plugin, so this works only on newer InnoDB and MySQL versions. If setting the value fails and the
current server value is greater than the specified value, then a warning is printed; else, if the current server value
is less than or equal to the specified value, no warning is printed.
-max-lag
type: time; default: 1s; group: Throttle
Pause checksumming until all replicas’ lag is less than this value. After each checksum query (each chunk), pt-
table-checksum looks at the replication lag of all replicas to which it connects, using Seconds_Behind_Master.
If any replica is lagging more than the value of this option, then pt-table-checksum will sleep for
--check-interval seconds, then check all replicas again. If you specify --check-slave-lag, then
the tool only examines that server for lag, not all servers. If you want to control exactly which servers the tool
monitors, use the DSN value to --recursion-method.
The tool waits forever for replicas to stop lagging. If any replica is stopped, the tool waits forever until the
replica is started. Checksumming continues once all replicas are running and not lagging too much.
The tool prints progress reports while waiting. If a replica is stopped, it prints a progress report immediately,
then again at every progress report interval.
-max-load
type: Array; default: Threads_running=25; group: Throttle
Examine SHOW GLOBAL STATUS after every chunk, and pause if any status variables are higher than the
threshold. The option accepts a comma-separated list of MySQL status variables to check for a threshold.
An optional =MAX_VALUE (or :MAX_VALUE) can follow each variable. If not given, the tool determines a
threshold by examining the current value and increasing it by 20%.



For example, if you want the tool to pause when Threads_connected gets too high, you can specify
“Threads_connected”, and the tool will check the current value when it starts working and add 20% to that
value. If the current value is 100, then the tool will pause when Threads_connected exceeds 120, and resume
working when it is below 120 again. If you want to specify an explicit threshold, such as 110, you can use either
“Threads_connected:110” or “Threads_connected=110”.
The purpose of this option is to prevent the tool from adding too much load to the server. If the checksum
queries are intrusive, or if they cause lock waits, then other queries on the server will tend to block and queue.
This will typically cause Threads_running to increase, and the tool can detect that by running SHOW GLOBAL
STATUS immediately after each checksum query finishes. If you specify a threshold for this variable, then you
can instruct the tool to wait until queries are running normally again. This will not prevent queueing, however;
it will only give the server a chance to recover from the queueing. If you notice queueing, it is best to decrease
the chunk time.
-password
short form: -p; type: string; group: Connection
-pid
type: string
-port
short form: -P; type: int; group: Connection
-progress
Print progress reports to STDERR.
The value is a comma-separated list with two parts. The first part can be percentage, time, or iterations; the
second part specifies how often an update should be printed, in percentage, seconds, or number of iterations.
The tool prints progress reports for a variety of time-consuming operations, including waiting for replicas to
catch up if they become lagged.
-quiet
short form: -q; cumulative: yes; default: 0
Print only the most important information (disables --progress). Specifying this option once causes the tool
to print only errors, warnings, and tables that have checksum differences.
Specifying this option twice causes the tool to print only errors. In this case, you can use the tool’s exit status to
determine if there were any warnings or checksum differences.
-recurse
type: int
Number of levels to recurse in the hierarchy when discovering replicas. Default is infinite. See also
--recursion-method.
-recursion-method
type: string
Preferred recursion method for discovering replicas. Possible methods are:



METHOD USES
=========== ==================
dsn=DSN DSNs from a table

The processlist method is the default, because SHOW SLAVE HOSTS is not reliable. However, the hosts
method can work better if the server uses a non-standard port (not 3306). The tool usually does the right thing
and finds all replicas, but you may give a preferred method and it will be used first.
The hosts method requires replicas to be configured with report_host, report_port, etc.
The dsn method is special: it specifies a table from which other DSN strings are read. The specified DSN must
specify a D and t, or a database-qualified t. The DSN table should have the following structure:
CREATE TABLE ‘dsns‘ (
‘id‘ int(11) NOT NULL AUTO_INCREMENT,
‘parent_id‘ int(11) DEFAULT NULL,
‘dsn‘ varchar(255) NOT NULL,
PRIMARY KEY (‘id‘)
);

To make the tool monitor only the hosts 10.10.1.16 and 10.10.1.17 for replication lag and checksum differences,
insert the values h=10.10.1.16 and h=10.10.1.17 into the table. Currently, the DSNs are ordered by id,
but id and parent_id are otherwise ignored.
-replicate
type: string; default: percona.checksums
Write checksum results to this table. The replicate table must have this structure (MAGIC_create_replicate):
CREATE TABLE checksums (
db char(64) NOT NULL,
tbl char(64) NOT NULL,
chunk int NOT NULL,
chunk_time float NULL,
chunk_index varchar(200) NULL,
lower_boundary text NULL,
upper_boundary text NULL,
this_crc char(40) NOT NULL,
this_cnt int NOT NULL,
master_crc char(40) NULL,
master_cnt int NULL,
ts timestamp NOT NULL,
PRIMARY KEY (db, tbl, chunk),
INDEX ts_db_tbl (ts, db, tbl)
) ENGINE=InnoDB;

By default, --[no]create-replicate-table is true, so the database and the table specified by this
option are created automatically if they do not exist.
Be sure to choose an appropriate storage engine for the replicate table. If you are checksumming InnoDB tables,
and you use MyISAM for this table, a deadlock will break replication, because the mixture of transactional and
non-transactional tables in the checksum statements will cause it to be written to the binlog even though it had
an error. It will then replay without a deadlock on the replicas, and break replication with “different error on
master and slave.” This is not a problem with pt-table-checksum; it’s a problem with MySQL replication, and
you can read more about it in the MySQL manual.
The replicate table is never checksummed (the tool automatically adds this table to --ignore-tables).



-[no]replicate-check
default: yes
Check replicas for data differences after finishing each table. The tool finds differences by executing a simple
SELECT statement on all detected replicas. The query compares the replica’s checksum results to the master’s
checksum results. It reports differences in the DIFFS column of the output.
-replicate-check-only
Check replicas for consistency without executing checksum queries. This option is used only with
--[no]replicate-check. If specified, pt-table-checksum doesn’t checksum any tables. It checks repli-
cas for differences found by previous checksumming, and then exits. It might be useful if you run pt-table-
checksum quietly in a cron job, for example, and later want a report on the results of the cron job, perhaps to
implement a Nagios check.
-replicate-database
type: string
USE only this database. By default, pt-table-checksum executes USE to select the database that contains
the table it’s currently working on. This is is a best effort to avoid problems with replication filters such as
binlog_ignore_db and replicate_ignore_db. However, replication filters can create a situation where there simply
is no one right way to do things. Some statements might not be replicated, and others might cause replication
to fail. In such cases, you can use this option to specify a default database that pt-table-checksum selects with
USE, and never changes. See also --[no]check-replication-filters.
-resume
Resume checksumming from the last completed chunk (disables --[no]empty-replicate-table). If
the tool stops before it checksums all tables, this option makes checksumming resume from the last chunk of
the last table that it finished.
-retries
Retry a chunk this many times when there is a nonfatal error. Nonfatal errors are problems such as a lock wait
timeout or the query being killed.
-separator
type: string; default: #
The separator character used for CONCAT_WS(). This character is used to join the values of columns when
checksumming.
-set-vars
type: string; default: wait_timeout=10000; group: Connection
executed.
-socket
short form: -S; type: string; group: Connection
-tables
short form: -t; type: hash; group: Filter
Checksum only this comma-separated list of tables. Table names may be qualified with the database name.
-tables-regex
Checksum only tables whose names match this Perl regex.



-trim
Add TRIM() to VARCHAR columns (helps when comparing 4.1 to >= 5.0). This is useful when you don’t
care about the trailing space differences between MySQL versions that vary in their handling of trailing spaces.
MySQL 5.0 and later all retain trailing spaces in VARCHAR, while previous versions would remove them.
These differences will cause false checksum differences.
-user
short form: -u; type: string; group: Connection
-version
group: Help
-where
type: string
Do only rows matching this WHERE clause. You can use this option to limit the checksum to only part of the
table. This is particularly useful if you have append-only tables and don’t want to constantly re-check all rows;
you could run a daily job to just check yesterday’s rows, for instance.
This option is much like the -w option to mysqldump. Do not specify the WHERE keyword. You might need to
quote the value. Here is an example:
:program:‘pt-table-checksum‘ --where "ts > CURRENT_DATE - INTERVAL 1 DAY"

2.29.8 DSN OPTIONS

details.
• A
• D
copy: no
DSN table database.
• F
dsn: mysql_read_default_ﬁle; copy: no
• h
Connect to host.
• p



• P
• S
dsn: mysql_socket; copy: no
• t
copy: no
DSN table table.
• u

2.29.9 ENVIRONMENT

PTDEBUG=1 pt-table-checksum ... > FILE 2>&1



Perl.

2.29.11 BUGS

For a list of known bugs, see http://www.percona.com/bugs/pt-table-checksum.
• Tool --version



2.29.12 DOWNLOADING






2.29.13 AUTHORS



Claus Jeppesen, Francois Saint-Jacques, Giuseppe Maxia, Heikki Tuuri, James Briggs, Martin Friebe, and Sergey
Zhuravlev




welcome.

2.29.17 VERSION

pt-table-checksum 2.1.1



2.30 pt-table-sync

2.30.1 NAME

pt-table-sync - Synchronize MySQL table data efficiently.

2.30.2 SYNOPSIS

Usage

pt-table-sync [OPTION...] DSN [DSN...]

pt-table-sync synchronizes data efficiently between MySQL tables.
This tool changes data, so for maximum safety, you should back up your data before you use it. When synchronizing
a server that is a replication slave with the –replicate or –sync-to-master methods, it always makes the changes on the
replication master, never the replication slave directly. This is in general the only safe way to bring a replica back
in sync with its master; changes to the replica are usually the source of the problems in the first place. However, the
changes it makes on the master should be no-op changes that set the data to their current values, and actually affect
only the replica. Please read the detailed documentation that follows to learn more about this.
Sync db.tbl on host1 to host2:
pt-table-sync --execute h=host1,D=db,t=tbl h=host2

Sync all tables on host1 to host2 and host3:
pt-table-sync --execute host1 host2 host3

Make slave1 have the same data as its replication master:
pt-table-sync --execute --sync-to-master slave1

Resolve differences that pt-table-checksum found on all slaves of master1:
pt-table-sync --execute --replicate test.checksum master1

Same as above but only resolve differences on slave1:
pt-table-sync --execute --replicate test.checksum
--sync-to-master slave1

Sync master2 in a master-master replication configuration, where master2’s copy of db.tbl is known or suspected to be
incorrect:
pt-table-sync --execute --sync-to-master h=master2,D=db,t=tbl

Note that in the master-master configuration, the following will NOT do what you want, because it will make changes
directly on master2, which will then flow through replication and change master1’s data:
# Don’t do this in a master-master setup!
pt-table-sync --execute h=master1,D=db,t=tbl master2



2.30.3 RISKS

With great power comes great responsibility! This tool changes data, so it is a good idea to back up your data. It is
also very powerful, which means it is very complex, so you should run it with the --dry-run option to see what
it will do, until you’re familiar with its operation. If you want to see which rows are different, without changing any
data, use --print instead of --execute.
Be careful when using pt-table-sync in any master-master setup. Master-master replication is inherently tricky, and
it’s easy to make mistakes. You need to be sure you’re using the tool correctly for master-master replication. See the
“SYNOPSIS” for the overview of the correct usage.
Also be careful with tables that have foreign key constraints with ON DELETE or ON UPDATE deﬁnitions because
these might cause unintended changes on the child tables.
In general, this tool is best suited when your tables have a primary key or unique index. Although it can synchronize
data in tables lacking a primary key or unique index, it might be best to synchronize that data by another means.
At the time of this release, there is a potential bug using --lock-and-rename with MySQL 5.1, a bug detecting
certain differences, a bug using ROUND() across different platforms, and a bug mixing collations.
sync.

2.30.4 DESCRIPTION

pt-table-sync does one-way and bidirectional synchronization of table data. It does not synchronize table structures,
indexes, or any other schema objects. The following describes one-way synchronization. “BIDIRECTIONAL SYNC-
ING” is described later.
This tool is complex and functions in several different ways. To use it safely and effectively, you should understand
three things: the purpose of --replicate, ﬁnding differences, and specifying hosts. These three concepts are
closely related and determine how the tool will run. The following is the abbreviated logic:
if DSN has a t part, sync only that table:
if 1 DSN:
if --sync-to-master:
The DSN is a slave. Connect to its master and sync.
if more than 1 DSN:
The first DSN is the source. Sync each DSN in turn.
else if --replicate:
if --sync-to-master:
The DSN is a slave. Connect to its master, find records
of differences, and fix.
else:
The DSN is the master. Find slaves and connect to each,
find records of differences, and fix.
else:
if only 1 DSN and --sync-to-master:
The DSN is a slave. Connect to its master, find tables and
filter with --databases etc, and sync each table to the master.
else:

2.30. pt-table-sync 229


find tables, filtering with --databases etc, and sync each
DSN to the first.

pt-table-sync can run in one of two ways: with --replicate or without. The default is to run without
--replicate which causes pt-table-sync to automatically find differences efficiently with one of several algo-
rithms (see “ALGORITHMS”). Alternatively, the value of --replicate, if specified, causes pt-table-sync to
use the differences already found by having previously ran pt-table-checksum with its own --replicate option.
Strictly speaking, you don’t need to use --replicate because pt-table-sync can find differences, but many people
use --replicate if, for example, they checksum regularly using pt-table-checksum then fix differences as needed
with pt-table-sync. If you’re unsure, read each tool’s documentation carefully and decide for yourself, or consult with
an expert.
Regardless of whether --replicate is used or not, you need to specify which hosts to sync. There are two ways:
with --sync-to-master or without. Specifying --sync-to-master makes pt-table-sync expect one and
only slave DSN on the command line. The tool will automatically discover the slave’s master and sync it so that
its data is the same as its master. This is accomplished by making changes on the master which then flow through
replication and update the slave to resolve its differences. Be careful though: although this option specifies and syncs
a single slave, if there are other slaves on the same master, they will receive via replication the changes intended for
the slave that you’re trying to sync.
Alternatively, if you do not specify --sync-to-master, the first DSN given on the command line is the source
host. There is only ever one source host. If you do not also specify --replicate, then you must specify at least
one other DSN as the destination host. There can be one or more destination hosts. Source and destination hosts must
be independent; they cannot be in the same replication topology. pt-table-sync will die with an error if it detects that
a destination host is a slave because changes are written directly to destination hosts (and it’s not safe to write directly
to slaves). Or, if you specify --replicate (but not --sync-to-master) then pt-table-sync expects one and
only one master DSN on the command line. The tool will automatically discover all the master’s slaves and sync them
to the master. This is the only way to sync several (all) slaves at once (because --sync-to-master only specifies
one slave).
Each host on the command line is specified as a DSN. The first DSN (or only DSN for cases like
--sync-to-master) provides default values for other DSNs, whether those other DSNs are specified on the com-
mand line or auto-discovered by the tool. So in this example,
pt-table-sync --execute h=host1,u=msandbox,p=msandbox h=host2

the host2 DSN inherits the u and p DSN parts from the host1 DSN. Use the --explain-hosts option to see how
pt-table-sync will interpret the DSNs given on the command line.

2.30.5 OUTPUT

If you specify the --verbose option, you’ll see information about the differences between the tables. There is one
row per table. Each server is printed separately. For example,
# Syncing h=host1,D=test,t=test1
# DELETE REPLACE INSERT UPDATE ALGORITHM START END EXIT DATABASE.TABLE
# 0 0 3 0 Chunk 13:00:00 13:00:17 2 test.test1

Table test.test1 on host1 required 3 INSERT statements to synchronize and it used the Chunk algorithm (see “ALGO-
RITHMS”). The sync operation for this table started at 13:00:00 and ended 17 seconds later (times taken from NOW()
on the source host). Because differences were found, its “EXIT STATUS” was 2.
If you specify the --print option, you’ll see the actual SQL statements that the script uses to synchronize the table
if --execute is also specified.
If you want to see the SQL statements that pt-table-sync is using to select chunks, nibbles, rows, etc., then specify
--print once and --verbose twice. Be careful though: this can print a lot of SQL statements.



There are cases where no combination of INSERT, UPDATE or DELETE statements can resolve differences without
violating some unique key. For example, suppose there’s a primary key on column a and a unique key on column b.
Then there is no way to sync these two tables with straightforward UPDATE statements:
+---+---+ +---+---+
| a | b | | a | b |
+---+---+ +---+---+
| 1 | 2 | | 1 | 1 |
| 2 | 1 | | 2 | 2 |
+---+---+ +---+---+

The tool rewrites queries to DELETE and REPLACE in this case. This is automatically handled after the first index
violation, so you don’t have to worry about it.

2.30.6 REPLICATION SAFETY

Synchronizing a replication master and slave safely is a non-trivial problem, in general. There are all sorts of issues
to think about, such as other processes changing data, trying to change data on the slave, whether the destination and
source are a master-master pair, and much more.
In general, the safe way to do it is to change the data on the master, and let the changes flow through replication to
the slave like any other changes. However, this works only if it’s possible to REPLACE into the table on the master.
REPLACE works only if there’s a unique index on the table (otherwise it just acts like an ordinary INSERT).
If your table has unique keys, you should use the --sync-to-master and/or --replicate options to sync a
slave to its master. This will generally do the right thing. When there is no unique key on the table, there is no choice
but to change the data on the slave, and pt-table-sync will detect that you’re trying to do so. It will complain and die
unless you specify --no-check-slave (see --[no]check-slave).
If you’re syncing a table without a primary or unique key on a master-master pair, you must change the data on the
destination server. Therefore, you need to specify --no-bin-log for safety (see --[no]bin-log). If you don’t,
the changes you make on the destination server will replicate back to the source server and change the data there!
The generally safe thing to do on a master-master pair is to use the --sync-to-master option so you don’t change
the data on the destination server. You will also need to specify --no-check-slave to keep pt-table-sync from
complaining that it is changing data on a slave.

2.30.7 ALGORITHMS

pt-table-sync has a generic data-syncing framework which uses different algorithms to find differences. The tool
automatically chooses the best algorithm for each table based on indexes, column types, and the algorithm preferences
specified by --algorithms. The following algorithms are available, listed in their default order of preference:
Chunk
Finds an index whose first column is numeric (including date and time types), and divides the column’s
range of values into chunks of approximately --chunk-size rows. Syncs a chunk at a time by check-
summing the entire chunk. If the chunk differs on the source and destination, checksums each chunk’s
rows individually to find the rows that differ.
It is efficient when the column has sufficient cardinality to make the chunks end up about the right size.
The initial per-chunk checksum is quite small and results in minimal network traffic and memory con-
sumption. If a chunk’s rows must be examined, only the primary key columns and a checksum are sent
over the network, not the entire row. If a row is found to be different, the entire row will be fetched, but
not before.
Nibble



Finds an index and ascends the index in fixed-size nibbles of --chunk-size rows, using a non-
backtracking algorithm (see pt-archiver for more on this algorithm). It is very similar to “Chunk”, but
instead of pre-calculating the boundaries of each piece of the table based on index cardinality, it uses
LIMIT to define each nibble’s upper limit, and the previous nibble’s upper limit to define the lower limit.
It works in steps: one query finds the row that will define the next nibble’s upper boundary, and the next
query checksums the entire nibble. If the nibble differs between the source and destination, it examines
the nibble row-by-row, just as “Chunk” does.
GroupBy
Selects the entire table grouped by all columns, with a COUNT(*) column added. Compares all columns,
and if they’re the same, compares the COUNT(*) column’s value to determine how many rows to insert
or delete into the destination. Works on tables with no primary key or unique index.
Stream
Selects the entire table in one big stream and compares all columns. Selects all columns. Much less
efficient than the other algorithms, but works when there is no suitable index for them to use.
Future Plans
Possibilities for future algorithms are TempTable (what I originally called bottom-up in earlier versions
of this tool), DrillDown (what I originally called top-down), and GroupByPrefix (similar to how SqlYOG
Job Agent works). Each algorithm has strengths and weaknesses. If you’d like to implement your favorite
technique for finding differences between two sources of data on possibly different servers, I’m willing to
help. The algorithms adhere to a simple interface that makes it pretty easy to write your own.

2.30.8 BIDIRECTIONAL SYNCING

Bidirectional syncing is a new, experimental feature. To make it work reliably there are a number of strict limitations:
* only works when syncing one server to other independent servers
* does not work in any way with replication
* requires that the table(s) are chunkable with the Chunk algorithm
* is not N-way, only bidirectional between two servers at a time
* does not handle DELETE changes

For example, suppose we have three servers: c1, r1, r2. c1 is the central server, a pseudo-master to the other servers
(viz. r1 and r2 are not slaves to c1). r1 and r2 are remote servers. Rows in table foo are updated and inserted on all
three servers and we want to synchronize all the changes between all the servers. Table foo has columns:
id int PRIMARY KEY
ts timestamp auto updated
name varchar

Auto-increment offsets are used so that new rows from any server do not create conflicting primary key (id) values. In
general, newer rows, as determined by the ts column, take precedence when a same but differing row is found during
the bidirectional sync. “Same but differing” means that two rows have the same primary key (id) value but different
values for some other column, like the name column in this example. Same but differing conflicts are resolved by
a “conflict”. A conflict compares some column of the competing rows to determine a “winner”. The winning row
becomes the source and its values are used to update the other row.
There are subtle differences between three columns used to achieve bidirectional syncing that you should be fa-
miliar with: chunk column (--chunk-column), comparison column(s) (--columns), and conflict column
(--conflict-column). The chunk column is only used to chunk the table; e.g. “WHERE id >= 5 AND id
< 10”. Chunks are checksummed and when chunk checksums reveal a difference, the tool selects the rows in that
chunk and checksums the --columns for each row. If a column checksum differs, the rows have one or more con-
flicting column values. In a traditional unidirectional sync, the conflict is a moot point because it can be resolved



simply by updating the entire destination row with the source row’s values. In a bidirectional sync, however, the
--conflict-column (in accordance with other --conflict-* options list below) is compared to determine
which row is “correct” or “authoritative”; this row becomes the “source”.
To sync all three servers completely, two runs of pt-table-sync are required. The first run syncs c1 and r1, then syncs
c1 and r2 including any changes from r1. At this point c1 and r2 are completely in sync, but r1 is missing any changes
from r2 because c1 didn’t have these changes when it and r1 were synced. So a second run is needed which syncs the
servers in the same order, but this time when c1 and r1 are synced r1 gets r2’s changes.
The tool does not sync N-ways, only bidirectionally between the first DSN given on the command line and each
subsequent DSN in turn. So the tool in this example would be ran twice like:
pt-table-sync --bidirectional h=c1 h=r1 h=r2

The --bidirectional option enables this feature and causes various sanity checks to be performed. You must
specify other options that tell pt-table-sync how to resolve conflicts for same but differing rows. These options are:
* --conflict-column
* --conflict-comparison
* --conflict-value
* --conflict-threshold
* --conflict-error"> (optional)

Use --print to test this option before --execute. The printed SQL statements will have comments saying on
which host the statement would be executed if you used --execute.
Technical side note: the first DSN is always the “left” server and the other DSNs are always the “right” server. Since
either server can become the source or destination it’s confusing to think of them as “src” and “dst”. Therefore, they’re
generically referred to as left and right. It’s easy to remember this because the first DSN is always to the left of the
other server DSNs on the command line.

2.30.9 EXIT STATUS

The following are the exit statuses (also called return values, or return codes) when pt-table-sync finishes and exits.
STATUS MEANING
====== =======================================================
0 Success.
1 Internal error.
2 At least one table differed on the destination.
3 Combination of 1 and 2.

2.30.10 OPTIONS

Specify at least one of --print, --execute, or --dry-run.
--where and --replicate are mutually exclusive.
-algorithms
type: string; default: Chunk,Nibble,GroupBy,Stream
Algorithm to use when comparing the tables, in order of preference.
For each table, pt-table-sync will check if the table can be synced with the given algorithms in the order that
they’re given. The first algorithm that can sync the table is used. See “ALGORITHMS”.



-ask-pass
-bidirectional
Enable bidirectional sync between ﬁrst and subsequent hosts.
See “BIDIRECTIONAL SYNCING” for more information.
-[no]bin-log
default: yes
Log to the binary log (SET SQL_LOG_BIN=1).
Specifying --no-bin-log will SET SQL_LOG_BIN=0.
-buffer-in-mysql
Instruct MySQL to buffer queries in its memory.
This option adds the SQL_BUFFER_RESULT option to the comparison queries. This causes MySQL to execute
the queries and place them in a temporary table internally before sending the results back to pt-table-sync. The
advantage of this strategy is that pt-table-sync can fetch rows as desired without using a lot of memory inside
the Perl process, while releasing locks on the MySQL table (to reduce contention with other queries). The
disadvantage is that it uses more memory on the MySQL server instead.
You probably want to leave --[no]buffer-to-client enabled too, because buffering into a temp table
and then fetching it all into Perl’s memory is probably a silly thing to do. This option is most useful for the
GroupBy and Stream algorithms, which may fetch a lot of data from the server.
-[no]buffer-to-client
default: yes
Fetch rows one-by-one from MySQL while comparing.
This option enables mysql_use_result which causes MySQL to hold the selected rows on the server until
the tool fetches them. This allows the tool to use less memory but may keep the rows locked on the server
longer.
If this option is disabled by specifying --no-buffer-to-client then mysql_store_result is used
which causes MySQL to send all selected rows to the tool at once. This may result in the results “cursor” being
held open for a shorter time on the server, but if the tables are large, it could take a long time anyway, and use
all your memory.
For most non-trivial data sizes, you want to leave this option enabled.
This option is disabled when --bidirectional is used.
-charset
-[no]check-master
default: yes
With --sync-to-master, try to verify that the detected master is the real master.
-[no]check-privileges
default: yes
Check that user has all necessary privileges on source and destination table.



-[no]check-slave
default: yes
Check whether the destination server is a slave.
If the destination server is a slave, it’s generally unsafe to make changes on it. However, sometimes you have
to; --replace won’t work unless there’s a unique index, for example, so you can’t make changes on the
master in that scenario. By default pt-table-sync will complain if you try to change data on a slave. Specify
--no-check-slave to disable this check. Use it at your own risk.
-[no]check-triggers
default: yes
Check that no triggers are defined on the destination table.
Triggers were introduced in MySQL v5.0.2, so for older versions this option has no effect because triggers will
not be checked.
-chunk-column
type: string
Chunk the table on this column.
-chunk-index
type: string
Chunk the table using this index.
-chunk-size
type: string; default: 1000
Number of rows or data size per chunk.
The size of each chunk of rows for the “Chunk” and “Nibble” algorithms. The size can be either a number of
rows, or a data size. Data sizes are specified with a suffix of k=kibibytes, M=mebibytes, G=gibibytes. Data
sizes are converted to a number of rows by dividing by the average row length.
-columns
short form: -c; type: array
Compare this comma-separated list of columns.
-config
type: Array
-conflict-column
type: string
Compare this column when rows conflict during a --bidirectional sync.
When a same but differing row is found the value of this column from each row is compared according to
--conflict-comparison, --conflict-value and --conflict-threshold to determine which
row has the correct data and becomes the source. The column can be any type for which there is an appropriate
--conflict-comparison (this is almost all types except, for example, blobs).
This option only works with --bidirectional. See “BIDIRECTIONAL SYNCING” for more information.
-conflict-comparison
type: string
Choose the --conflict-column with this property as the source.



The option affects how the --conflict-column values from the conflicting rows are compared. Possible
comparisons are one of these MAGIC_comparisons:
newest|oldest|greatest|least|equals|matches

COMPARISON CHOOSES ROW WITH
========== =========================================================
newest Newest temporal --conflict-column value
oldest Oldest temporal --conflict-column value
greatest Greatest numerical "--conflict-column value
least Least numerical --conflict-column value
equals --conflict-column value equal to --conflict-value
matches --conflict-column value matching Perl regex pattern
--conflict-value

-conflict-error
type: string; default: warn
How to report unresolvable conflicts and conflict errors
This option changes how the user is notified when a conflict cannot be resolved or causes some kind of error.
Possible values are:
* warn: Print a warning to STDERR about the unresolvable conflict
* die: Die, stop syncing, and print a warning to STDERR

-conflict-threshold
type: string
Amount by which one --conflict-column must exceed the other.
The --conflict-threshold prevents a conflict from being resolved if the absolute difference between
the two --conflict-column values is less than this amount. For example, if two --conflict-column
have timestamp values “2009-12-01 12:00:00” and “2009-12-01 12:05:00” the difference is 5 minutes. If
--conflict-threshold is set to “5m” the conflict will be resolved, but if --conflict-threshold
is set to “6m” the conflict will fail to resolve because the difference is not greater than or equal to 6 minutes. In
this latter case, --conflict-error will report the failure.
-conflict-value
type: string
Use this value for certain --conflict-comparison.
This option gives the value for equals and matches --conflict-comparison.
-databases
Sync only this comma-separated list of databases.
A common request is to sync tables from one database with tables from another database on the same or different
server. This is not yet possible. --databases will not do it, and you can’t do it with the D part of the DSN
either because in the absence of a table name it assumes the whole server should be synced and the D part
controls only the connection’s default database.



-defaults-file
-dry-run
Analyze, decide the sync algorithm to use, print and exit.
Implies --verbose so you can see the results. The results are in the same output format that you’ll see from
actually running the tool, but there will be zeros for rows affected. This is because the tool actually executes,
but stops before it compares any data and just returns zeros. The zeros do not mean there are no changes to be
made.
-engines
short form: -e; type: hash
Sync only this comma-separated list of storage engines.
-execute
Execute queries to make the tables have identical data.
This option makes pt-table-sync actually sync table data by executing all the queries that it created to resolve
table differences. Therefore, the tables will be changed! And unless you also specify --verbose, the
changes will be made silently. If this is not what you want, see --print or --dry-run.
-explain-hosts
Print connection information and exit.
Print out a list of hosts to which pt-table-sync will connect, with all the various connection options, and exit.
-float-precision
type: int
Precision for FLOAT and DOUBLE number-to-string conversion. Causes FLOAT and DOUBLE values to be
rounded to the specified number of digits after the decimal point, with the ROUND() function in MySQL.
This can help avoid checksum mismatches due to different floating-point representations of the same values on
different MySQL versions and hardware. The default is no rounding; the values are converted to strings by the
CONCAT() function, and MySQL chooses the string representation. If you specify a value of 2, for example,
then the values 1.008 and 1.009 will be rounded to 1.01, and will checksum as equal.
-[no]foreign-key-checks
default: yes
Enable foreign key checks (SET FOREIGN_KEY_CHECKS=1).
Specifying --no-foreign-key-checks will SET FOREIGN_KEY_CHECKS=0.
-function
type: string
Which hash function you’d like to use for checksums.
The default is CRC32. Other good choices include MD5 and SHA1. If you have installed the FNV_64 user-
defined function, pt-table-sync will detect it and prefer to use it, because it is much faster than the built-ins.
You can also use MURMUR_HASH if you’ve installed that user-defined function. Both of these are distributed
with Maatkit. See pt-table-checksum for more information and benchmarks.
-help
Show help and exit.
-[no]hex-blob
default: yes
HEX() BLOB, TEXT and BINARY columns.



When row data from the source is fetched to create queries to sync the data (i.e. the queries seen with --print
and executed by --execute), binary columns are wrapped in HEX() so the binary data does not produce an
invalid SQL statement. You can disable this option but you probably shouldn’t.
-host
Connect to host.
-ignore-columns
type: Hash
Ignore this comma-separated list of column names in comparisons.
This option causes columns not to be compared. However, if a row is determined to differ between tables, all
columns in that row will be synced, regardless. (It is not currently possible to exclude columns from the sync
process itself, only from the comparison.)
-ignore-databases
type: Hash
-ignore-engines
type: Hash; default: FEDERATED,MRG_MyISAM
-ignore-tables
type: Hash
Ignore this comma-separated list of tables.
-[no]index-hint
default: yes
Add FORCE/USE INDEX hints to the chunk and row queries.
By default pt-table-sync adds a FORCE/USE INDEX hint to each SQL statement to coerce MySQL into using
the index chosen by the sync algorithm or speciﬁed by --chunk-index. This is usually a good thing, but
in rare cases the index may not be the best for the query so you can suppress the index hint by specifying
--no-index-hint and let MySQL choose the index.
This does not affect the queries printed by --print; it only affects the chunk and row queries that pt-table-
sync uses to select and compare rows.
-lock
type: int
Lock tables: 0=none, 1=per sync cycle, 2=per table, or 3=globally.
This uses LOCK TABLES. This can help prevent tables being changed while you’re examining them. The
possible values are as follows:
VALUE MEANING
===== =======================================================
0 Never lock tables.
1 Lock and unlock one time per sync cycle (as implemented
by the syncing algorithm). This is the most granular
level of locking available. For example, the Chunk
algorithm will lock each chunk of C<N> rows, and then
unlock them if they are the same on the source and the



destination, before moving on to the next chunk.
2 Lock and unlock before and after each table.
3 Lock and unlock once for every server (DSN) synced, with
C<FLUSH TABLES WITH READ LOCK>.

A replication slave is never locked if --replicate or --sync-to-master is specified, since in theory
locking the table on the master should prevent any changes from taking place. (You are not changing data on
your slave, right?) If --wait is given, the master (source) is locked and then the tool waits for the slave to
catch up to the master before continuing.
If --transaction is specified, LOCK TABLES is not used. Instead, lock and unlock are implemented by
beginning and committing transactions. The exception is if --lock is 3.
If --no-transaction is specified, then LOCK TABLES is used for any value of --lock. See
--[no]transaction.
-lock-and-rename
Lock the source and destination table, sync, then swap names. This is useful as a less-blocking ALTER TABLE,
once the tables are reasonably in sync with each other (which you may choose to accomplish via any number
of means, including dump and reload or even something like pt-archiver). It requires exactly two DSNs and
assumes they are on the same server, so it does no waiting for replication or the like. Tables are locked with
LOCK TABLES.
-password
-pid
type: string
-port
-print
Print queries that will resolve differences.
If you don’t trust pt-table-sync, or just want to see what it will do, this is a good way to be safe. These queries
are valid SQL and you can run them yourself if you want to sync the tables manually.
-recursion-method
type: string
METHOD USES
=========== ================

is required if the server uses a non-standard port (not 3306). Usually pt-table-sync does the right thing and finds



-replace
Write all INSERT and UPDATE statements as REPLACE.
This is automatically switched on as needed when there are unique index violations.
-replicate
type: string
Sync tables listed as different in this table.
Specifies that pt-table-sync should examine the specified table to find data that differs. The table is exactly the
same as the argument of the same name to pt-table-checksum. That is, it contains records of which tables (and
ranges of values) differ between the master and slave.
For each table and range of values that shows differences between the master and slave,
pt-table-checksum will sync that table, with the appropriate WHERE clause, to its master.
This automatically sets --wait to 60 and causes changes to be made on the master instead of the slave.
If --sync-to-master is specified, the tool will assume the server you specified is the slave, and connect to
the master as usual to sync.
Otherwise, it will try to use SHOW PROCESSLIST to find slaves of the server you specified. If it is unable to
find any slaves via SHOW PROCESSLIST, it will inspect SHOW SLAVE HOSTS instead. You must configure
each slave’s report-host, report-port and other options for this to work right. After finding slaves, it
will inspect the specified table on each slave to find data that needs to be synced, and sync it.
The tool examines the master’s copy of the table first, assuming that the master is potentially a slave as well. Any
table that shows differences there will NOT be synced on the slave(s). For example, suppose your replication is
set up as A->B, B->C, B->D. Suppose you use this argument and specify server B. The tool will examine server
B’s copy of the table. If it looks like server B’s data in table test.tbl1 is different from server A’s copy, the
tool will not sync that table on servers C and D.
-set-vars
executed.
-socket
-sync-to-master
Treat the DSN as a slave and sync it to its master.
Treat the server you specified as a slave. Inspect SHOW SLAVE STATUS, connect to the server’s master, and
treat the master as the source and the slave as the destination. Causes changes to be made on the master. Sets
--wait to 60 by default, sets --lock to 1 by default, and disables --[no]transaction by default. See
also --replicate, which changes this option’s behavior.
-tables
Sync only this comma-separated list of tables.
-timeout-ok
Keep going if --wait fails.



If you specify --wait and the slave doesn’t catch up to the master’s position before the wait times out, the
default behavior is to abort. This option makes the tool keep going anyway. Warning: if you are trying to get a
consistent comparison between the two servers, you probably don’t want to keep going after a timeout.
-[no]transaction
Use transactions instead of LOCK TABLES.
The granularity of beginning and committing transactions is controlled by --lock. This is enabled by default,
but since --lock is disabled by default, it has no effect.
Most options that enable locking also disable transactions by default, so if you want to use transactional locking
(via LOCK IN SHARE MODE and FOR UPDATE, you must specify --transaction explicitly.
If you don’t specify --transaction explicitly pt-table-sync will decide on a per-table basis whether to use
transactions or table locks. It currently uses transactions on InnoDB tables, and table locks on all others.
If --no-transaction is speciﬁed, then pt-table-sync will not use transactions at all (not even for InnoDB
tables) and locking is controlled by --lock.
When enabled, either explicitly or implicitly, the transaction isolation level is set REPEATABLE READ and
transactions are started WITH CONSISTENT SNAPSHOT.
-trim
TRIM() VARCHAR columns in BIT_XOR and ACCUM modes. Helps when comparing MySQL 4.1 to >= 5.0.
This is useful when you don’t care about the trailing space differences between MySQL versions which vary in
their handling of trailing spaces. MySQL 5.0 and later all retain trailing spaces in VARCHAR, while previous
versions would remove them.
-[no]unique-checks
default: yes
Enable unique key checks (SET UNIQUE_CHECKS=1).
Specifying --no-unique-checks will SET UNIQUE_CHECKS=0.
-user
-verbose
short form: -v; cumulative: yes
Print results of sync operations.
See “OUTPUT” for more details about the output.
-version
-wait
short form: -w; type: time
How long to wait for slaves to catch up to their master.
Make the master wait for the slave to catch up in replication before comparing the tables. The value is
the number of seconds to wait before timing out (see also --timeout-ok). Sets --lock to 1 and
--[no]transaction to 0 by default. If you see an error such as the following,
MASTER_POS_WAIT returned -1

It means the timeout was exceeded and you need to increase it.
The default value of this option is inﬂuenced by other options. To see what value is in effect, run with --help.



To disable waiting entirely (except for locks), specify --wait 0. This helps when the slave is lagging on tables
that are not being synced.
-where
type: string
WHERE clause to restrict syncing to part of the table.
-[no]zero-chunk
default: yes
Add a chunk for rows with zero or zero-equivalent values. The only has an effect when --chunk-size is
specified. The purpose of the zero chunk is to capture a potentially large number of zero values that would
imbalance the size of the first chunk. For example, if a lot of negative numbers were inserted into an unsigned
integer column causing them to be stored as zeros, then these zero values are captured by the zero chunk instead
of the first chunk and all its non-zero values.

2.30.11 DSN OPTIONS

details.
• A
• D
Database containing the table to be synced.
• F
• h
Connect to host.
• p
• P
• S



• t
copy: yes
Table to be synced.
• u

2.30.12 ENVIRONMENT

PTDEBUG=1 pt-table-sync ... > FILE 2>&1



Perl.

2.30.14 BUGS

For a list of known bugs, see http://www.percona.com/bugs/pt-table-sync.
• Tool --version

2.30.15 DOWNLOADING









2.30.16 AUTHORS

Baron Schwartz


My work is based in part on Giuseppe Maxia’s work on distributed databases,
http://www.sysadminmag.com/articles/2004/0408/ and code derived from that article. There is more explana-
tion, and a link to the code, at http://www.perlmonks.org/?node_id=381053.
Another programmer extended Maxia’s work even further. Fabien Coelho changed and generalized Maxia’s technique,
introducing symmetry and avoiding some problems that might have caused too-frequent checksum collisions. This
work grew into pg_comparator, http://www.coelho.net/pg_comparator/. Coelho also explained the technique further
in a paper titled “Remote Comparison of Database Tables” (http://cri.ensmp.fr/classement/doc/A-375.pdf).
This existing literature mostly addressed how to find the differences between the tables, not how to resolve them
once found. I needed a tool that would not only find them efficiently, but would then resolve them. I first began
thinking about how to improve the technique further with my article http://tinyurl.com/mysql-data-diff-algorithm,
where I discussed a number of problems with the Maxia/Coelho “bottom-up” algorithm. After writing that article,
I began to write this tool. I wanted to actually implement their algorithm with some improvements so I was sure
I understood it completely. I discovered it is not what I thought it was, and is considerably more complex than it
appeared to me at first. Fabien Coelho was kind enough to address some questions over email.
The first versions of this tool implemented a version of the Coelho/Maxia algorithm, which I called “bottom-up”, and
my own, which I called “top-down.” Those algorithms are considerably more complex than the current algorithms and
I have removed them from this tool, and may add them back later. The improvements to the bottom-up algorithm are
my original work, as is the top-down algorithm. The techniques to actually resolve the differences are also my own
work.
Another tool that can synchronize tables is the SQLyog Job Agent from webyog. Thanks to Rohit Nadhani, SJA’s
author, for the conversations about the general techniques. There is a comparison of pt-table-sync and SJA at
http://tinyurl.com/maatkit-vs-sqlyog
Thanks to the following people and organizations for helping in many ways:
The Rimm-Kaufman Group http://www.rimmkaufman.com/, MySQL AB http://www.mysql.com/, Blue Ridge Inter-
netWorks http://www.briworks.com/, Percona http://www.percona.com/, Fabien Coelho, Giuseppe Maxia and others
at MySQL AB, Kristian Koehntopp (MySQL AB), Rohit Nadhani (WebYog), The helpful monks at Perlmonks, And
others too numerous to mention.






welcome.

2.30.20 VERSION

pt-table-sync 2.1.1

2.31 pt-table-usage

2.31.1 NAME

pt-table-usage - Analyze how queries use tables.

2.31.2 SYNOPSIS

Usage

pt-table-usage [OPTIONS] [FILES]

pt-table-usage reads queries from a log and analyzes how they use tables. If no FILE is speciﬁed, it reads STDIN. It
prints a report for each query.

2.31.3 RISKS

pt-table-use is very low risk. By default, it simply reads queries from a log. It executes EXPLAIN EXTENDED if you
specify the --explain-extended option.
usage.

2.31. pt-table-usage 245


2.31.4 DESCRIPTION

pt-table-usage reads queries from a log and analyzes how they use tables. The log should be in MySQL’s slow query
log format.
Table usage is more than simply an indication of which tables the query reads or writes. It also indicates data flow:
data in and data out. The tool determines the data flow by the contexts in which tables appear. A single query can
use a table in several different contexts simultaneously. The tool’s output lists every context for every table. This
CONTEXT-TABLE list indicates how data flows between tables. The “OUTPUT” section lists the possible contexts
and describes how to read a table usage report.
The tool analyzes data flow down to the level of individual columns, so it is helpful if columns are identified un-
ambiguously in the query. If a query uses only one table, then all columns must be from that table, and there’s no
difficulty. But if a query uses multiple tables and the column names are not table-qualified, then it is necessary to use
EXPLAIN EXTENDED, followed by SHOW WARNINGS, to determine to which tables the columns belong.
If the tool does not know the query’s default database, which can occur when the database is not printed in the log,
then EXPLAIN EXTENDED can fail. In this case, you can specify a default database with --database. You can
also use the --create-table-definitions option to help resolve ambiguities.

2.31.5 OUTPUT

The tool prints a usage report for each table in every query, similar to the following:
Query_id: 0x1CD27577D202A339.1
UPDATE t1
SELECT DUAL
JOIN t1
JOIN t2
WHERE t1

UPDATE t2
SELECT DUAL
JOIN t1
JOIN t2
WHERE t1

The first line contains the query ID, which by default is the same as those shown in pt-query-digest reports. It is an
MD5 checksum of the query’s “fingerprint,” which is what remains after removing literals, collapsing white space,
and a variety of other transformations. The query ID has two parts separated by a period: the query ID and the table
number. If you wish to use a different value to identify the query, you can specify the --id-attribute option.
The previous example shows two paragraphs for a single query, not two queries. Note that the query ID is identical for
the two, but the table number differs. The table number increments by 1 for each table that the query updates. Only
multi-table UPDATE queries can update multiple tables with a single query, so the table number is 1 for all other types
of queries. (The tool does not support multi-table DELETE queries.) The example output above is from this query:
UPDATE t1 AS a JOIN t2 AS b USING (id)
SET a.foo="bar", b.foo="bat"
WHERE a.id=1;

The SET clause indicates that the query updates two tables: a aliased as t1, and b aliased as t2.
After the first line, the tool prints a variable number of CONTEXT-TABLE lines. Possible contexts are as follows:
• SELECT



SELECT means that the query retrieves data from the table for one of two reasons. The first is to be
returned to the user as part of a result set. Only SELECT queries return result sets, so the report always
shows a SELECT context for SELECT queries.
The second case is when data flows to another table as part of an INSERT or UPDATE. For example, the
UPDATE query in the example above has the usage:
SELECT DUAL

This refers to:
SET a.foo="bar", b.foo="bat"

The tool uses DUAL for any values that do not originate in a table, in this case the literal values “bar” and
“bat”. If that SET clause were SET a.foo=b.foo instead, then the complete usage would be:
UPDATE t1
SELECT t2
JOIN t1
JOIN t2
WHERE t1

The presence of a SELECT context after another context, such as UPDATE or INSERT, indicates where
the UPDATE or INSERT retrieves its data. The example immediately above reflects an UPDATE query
that updates rows in table t1 with data from table t2.
• Any other verb
Any other verb, such as INSERT, UPDATE, DELETE, etc. may be a context. These verbs indicate that the
query modifies data in some way. If a SELECT context follows one of these verbs, then the query reads
data from the SELECT table and writes it to this table. This happens, for example, with INSERT..SELECT
or UPDATE queries that use values from tables instead of constant values.
These query types are not supported: SET, LOAD, and multi-table DELETE.
• JOIN
The JOIN context lists tables that are joined, either with an explicit JOIN in the FROM clause, or implicitly
in the WHERE clause, such as t1.id = t2.id.
• WHERE
The WHERE context lists tables that are used in the WHERE clause to filter results. This does not include
tables that are implicitly joined in the WHERE clause; those are listed as JOIN contexts. For example:
WHERE t1.id > 100 AND t1.id < 200 AND t2.foo IS NOT NULL

Results in:
WHERE t1
WHERE t2

The tool lists only distinct tables; that is why table t1 is listed only once.
• TLIST
The TLIST context lists tables that the query accesses, but which do not appear in any other context.
These tables are usually an implicit cartesian join. For example, the query SELECT * FROM t1, t2
results in:



Query_id: 0xBDDEB6EDA41897A8.1
SELECT t1
SELECT t2
TLIST t1
TLIST t2

First of all, there are two SELECT contexts, because SELECT * selects rows from all tables; t1 and
t2 in this case. Secondly, the tables are implicitly joined, but without any kind of join condition, which
results in a cartesian join as indicated by the TLIST context for each.

2.31.6 EXIT STATUS

pt-table-usage exits 1 on any kind of error, or 0 if no errors.

2.31.7 OPTIONS

-ask-pass
-charset
-config
type: Array
-constant-data-value
type: string; default: DUAL
Table to print as the source for constant data (literals). This is any data not retrieved from tables (or subqueries,
because subqueries are not supported). This includes literal values such as strings (“foo”) and numbers (42),
or functions such as NOW(). For example, in the query INSERT INTO t (c) VALUES (’a’), the string
‘a’ is constant data, so the table usage report is:
INSERT t
SELECT DUAL

The first line indicates that the query inserts data into table t, and the second line indicates that the inserted data
comes from some constant value.
default: yes
Continue to work even if there is an error.
-create-table-definitions
type: array
Read CREATE TABLE definitions from this list of comma-separated files. If you cannot use
--explain-extended to fully qualify table and column names, you can save the output of mysqldump
--no-data to one or more files and specify those files with this option. The tool will parse all CREATE



TABLE definitions from the files and use this information to qualify table and column names. If a column name
appears in multiple tables, or a table name appears in multiple databases, the ambiguities cannot be resolved.
-daemonize
-database
Default database.
-defaults-file
-explain-extended
type: DSN
A server to execute EXPLAIN EXTENDED queries. This may be necessary to resolve ambiguous (unqualified)
column and table names.
-filter
type: string
This option is a string of Perl code or a file containing Perl code that is compiled into a subroutine with one
argument: $event. If the given value is a readable file, then pt-table-usage reads the entire file and uses its
contents as the code.
Filters are implemented in the same fashion as in the pt-query-digest tool, so please refer to its documentation
for more information.
-help
Show help and exit.
-host
Connect to host.
-id-attribute
type: string
Identify each event using this attribute. The default is to use a query ID, which is an MD5 checksum of the
query’s fingerprint.
-log
type: string
-password
-pid
type: string
Create the given PID file when running. The file contains the process ID of the daemonized instance. The PID
file is removed when the daemonized instance exits. The program checks for the existence of the PID file when
starting; if it exists and the process with the matching PID exists, the program exits.



-port
-progress
-query
type: string
Analyze the specified query instead of reading a log file.
-read-timeout
Wait this long for an event from the input; 0 to wait forever.
This option sets the maximum time to wait for an event from the input. If an event is not received after the
specified time, the tool stops reading the input and prints its reports.
This option requires the Perl POSIX module.
-run-time
type: time
How long to run before exiting. The default is to run forever (you can interrupt with CTRL-C).
-set-vars
executed.
-socket
-user
-version

2.31.8 DSN OPTIONS

details.
• A



• D
copy: no
Default database.
• F
dsn: mysql_read_default_ﬁle; copy: no
• h
Connect to host.
• p
• P
• S
dsn: mysql_socket; copy: no
• u

2.31.9 ENVIRONMENT

PTDEBUG=1 pt-table-usage ... > FILE 2>&1



Perl.

2.31.11 BUGS

For a list of known bugs, see http://www.percona.com/bugs/pt-table-usage.



• Tool --version

2.31.12 DOWNLOADING






2.31.13 AUTHORS

Daniel Nichter




This program is copyright 2012 Percona Inc. Feedback and improvements are welcome.



2.31.16 VERSION

pt-table-usage 2.1.1

2.32 pt-tcp-model

2.32.1 NAME

pt-tcp-model - Transform tcpdump into metrics that permit performance and scalability modeling.

2.32.2 SYNOPSIS

Usage

pt-tcp-model [OPTION...] [FILE]

pt-tcp-model parses and analyzes tcpdump files. With no FILE, or when FILE is -, it read standard input.
Dump TCP requests and responses to a file, capturing only the packet headers to avoid dropped packets, and ignoring
any packets without a payload (such as ack-only packets). Capture port 3306 (MySQL database traffic). Note that to
avoid line breaking in terminals and man pages, the TCP filtering expression that follows has a line break at the end
of the second line; you should omit this from your tcpdump command.
tcpdump -s 384 -i any -nnq -tttt
’tcp port 3306 and (((ip[2:2] - ((ip[0]&0xf)<<2))
- ((tcp[12]&0xf0)>>2)) != 0)’
> /path/to/tcp-file.txt

Extract individual response times, sorted by end time:
pt-tcp-model /path/to/tcp-file.txt > requests.txt

Sort the result by arrival time, for input to the next step:
sort -n -k1,1 requests.txt > sorted.txt

Slice the result into 10-second intervals and emit throughput, concurrency, and response time metrics for each interval:
pt-tcp-model --type=requests --run-time=10 sorted.txt > sliced.txt

Transform the result for modeling with Aspersa’s usl tool, discarding the first and last line of each file if you specify
multiple files (the first and last line are normally incomplete observation periods and are aberrant):
for f in sliced.txt; do
tail -n +2 "$f" | head -n -1 | awk ’{print $2, $3, $7/$4}’
done > usl-input.txt

2.32.3 RISKS

pt-tcp-model merely reads and transforms its input, printing it to the output. It should be very low risk.

2.32. pt-tcp-model 253


will be marked as such. You can see a list of such issues at the following URL: http://www.percona.com/bugs/pt-tcp-
model.

2.32.4 DESCRIPTION

This tool recognizes requests and responses in a TCP stream, and extracts the “conversations”. You can use it to
capture the response times of individual queries to a database, for example. It expects the TCP input to be in the
following format, which should result from the sample shown in the SYNOPSIS:
<date> <time.microseconds> IP <IP.port> > <IP.port>: <junk>

The tool watches for “incoming” packets to the port you specify with the --watch-server option. This begins a
request. If multiple inbound packets follow each other, then by default the last inbound packet seen determines the
time at which the request is assumed to begin. This is logical if one assumes that a server must receive the whole SQL
statement before beginning execution, for example.
When the first outbound packet is seen, the server is considered to have responded to the request. The tool might see
an inbound packet, but never see a response. This can happen when the kernel drops packets, for example. As a result,
the tool never prints a request unless it sees the response to it. However, the tool actually does not print any request
until it sees the “last” outbound packet. It determines this by waiting for either another inbound packet, or EOF, and
then considers the previous inbound/outbound pair to be complete. As a result, the tool prints requests in a relatively
random order. Most types of analysis require processing in either arrival or completion order. Therefore, the second
type of processing this tool can do requires that you sort the output from the first stage and supply it as input.
The second type of processing is selected with the --type option set to “requests”. In this mode, the tool reads a
group of requests and aggregates them, then emits the aggregated metrics.

2.32.5 OUTPUT

In the default mode (parsing tcpdump output), requests are printed out one per line, in the following format:
<id> <start> <end> <elapsed> <IP:port>

The ID is an incrementing number, assigned in arrival order in the original TCP traffic. The start and end timestamps,
and the elapsed time, can be customized with the --start-end option.
In --type=requests mode, the tool prints out one line per time interval as defined by --run-time, with the
following columns: ts, concurrency, throughput, arrivals, completions, busy_time, weighted_time, sum_time, vari-
ance_mean, quantile_time, obs_time. A detailed explanation follows:
ts
The timestamp that defines the beginning of the interval.
concurrency
The average number of requests resident in the server during the interval.
throughput
The number of arrivals per second during the interval.
arrivals
The number of arrivals during the interval.



completions
The number of completions during the interval.
busy_time
The total amount of time during which at least one request was resident in the server during the interval.
weighted_time
The total response time of all the requests resident in the server during the interval, including requests that
neither arrived nor completed during the interval.
sum_time
The total response time of all the requests that arrived in the interval.
variance_mean
The variance-to-mean ratio (index of dispersion) of the response times of the requests that arrived in the
interval.
quantile_time
The Nth percentile response time for all the requests that arrived in the interval. See also --quantile.
obs_time
The length of the observation time window. This will usually be the same as the interval length, except
for the first and last intervals in a file, which might have a shorter observation time.

2.32.6 OPTIONS

-config
type: Array
-help
Show help and exit.
-progress
-quantile
type: float
The percentile for the last column when --type is “requests” (default .99).
-run-time
type: float
The size of the aggregation interval in seconds when --type is “requests” (default 1). Fractional values are
permitted.
-start-end
type: Array; default: ts,end



Define how the arrival and completion timestamps of a query, and thus its response time (elapsed time) are
computed. Recall that there may be multiple inbound and outbound packets per request and response, and refer
to the following ASCII diagram. Suppose that a client sends a series of three inbound (I) packets to the server,
which computes the result and then sends two outbound (O) packets back:
I I I ..................... O O
|<---->|<---response time----->|<-->|
ts0 ts end end1

By default, the query is considered to arrive at time ts, and complete at time end. However, this might not be
what you want. Perhaps you do not want to consider the query to have completed until time end1. You can
accomplish this by setting this option to ts,end1.
-type
type: string
The type of input to parse (default tcpdump). The permitted types are
tcpdump
The parser expects the input to be formatted with the following options: -x -n -q -tttt. For
example, if you want to capture output from your local machine, you can do something like the
following (the port must come last on FreeBSD):
tcpdump -s 65535 -x -nn -q -tttt -i any -c 1000 port 3306
> mysql.tcp.txt
pt-query-digest --type tcpdump mysql.tcp.txt

The other tcpdump parameters, such as -s, -c, and -i, are up to you. Just make sure the output looks
like this (there is a line break in the first line to avoid man-page problems):
2009-04-12 09:50:16.804849 IP 127.0.0.1.42167
> 127.0.0.1.3306: tcp 37

All MySQL servers running on port 3306 are automatically detected in the tcpdump output.
Therefore, if the tcpdump out contains packets from multiple servers on port 3306 (for example,
10.0.0.1:3306, 10.0.0.2:3306, etc.), all packets/queries from all these servers will be analyzed to-
gether as if they were one server.
If you’re analyzing traffic for a protocol that is not running on port 3306, see --watch-server.
-version
-watch-server
type: string; default: 10.10.10.10:3306
This option tells pt-tcp-model which server IP address and port (such as “10.0.0.1:3306”) to watch when parsing
tcpdump for --type tcpdump. If you don’t specify it, the tool watches all servers by looking for any IP address
using port 3306. If you’re watching a server with a non-standard port, this won’t work, so you must specify the
IP address and port to watch.
Currently, IP address filtering isn’t implemented; so even though you must specify the option in IP:port form, it
ignores the IP and only looks at the port number.

2.32.7 ENVIRONMENT




PTDEBUG=1 pt-tcp-model ... > FILE 2>&1



Perl.

2.32.9 BUGS

For a list of known bugs, see http://www.percona.com/bugs/pt-tcp-model.
• Tool --version

2.32.10 DOWNLOADING






2.32.11 AUTHORS

Baron Schwartz






This program is copyright 2011 Baron Schwartz, 2011-2012 Percona Inc. Feedback and improvements are welcome.

2.32.14 VERSION

pt-tcp-model 2.1.1

2.33 pt-trend

2.33.1 NAME

pt-trend - Compute statistics over a set of time-series data points.

2.33.2 SYNOPSIS

Usage

pt-trend [OPTION...] [FILE ...]

pt-trend reads a slow query log and outputs statistics on it.

2.33.3 RISKS

pt-trend simply reads files give on the command-line. It should be very low-risk.
will be marked as such. You can see a list of such issues at the following URL: http://www.percona.com/bugs/pt-trend.

2.33.4 DESCRIPTION

You can specify multiple files on the command line. If you don’t specify any, or if you use the special filename -,
lines are read from standard input.



2.33.5 OPTIONS

-config
type: Array
-help
Show help and exit.
-pid
type: string
-progress
-quiet
short form: -q
Disables --progress.
-version

2.33.6 ENVIRONMENT

PTDEBUG=1 pt-trend ... > FILE 2>&1



Perl.

2.33.8 BUGS

For a list of known bugs, see http://www.percona.com/bugs/pt-trend.

2.33. pt-trend 259


• Tool --version

2.33.9 DOWNLOADING






2.33.10 AUTHORS

Baron Schwartz




welcome.



2.33.13 VERSION

pt-trend 2.1.1

2.34 pt-upgrade

2.34.1 NAME

pt-upgrade - Execute queries on multiple servers and check for differences.

2.34.2 SYNOPSIS

Usage

pt-upgrade [OPTION...] DSN [DSN...] [FILE]

pt-upgrade compares query execution on two hosts by executing queries in the given ﬁle (or STDIN if no ﬁle given)
and examining the results, errors, warnings, etc.produced on each.
Execute and compare all queries in slow.log on host1 to host2:
pt-upgrade slow.log h=host1 h=host2

Use pt-query-digest to get, execute and compare queries from tcpdump:
tcpdump -i eth0 port 3306 -s 65535 -x -n -q -tttt
| pt-query-digest --type tcpdump --no-report --print
| pt-upgrade h=host1 h=host2

Compare only query times on host1 to host2 and host3:
pt-upgrade slow.log h=host1 h=host2 h=host3 --compare query_times

Compare a single query, no slowlog needed:
pt-upgrade h=host1 h=host2 --query ’SELECT * FROM db.tbl’

2.34.3 RISKS

pt-upgrade is a read-only tool that is meant to be used on non-production servers. It executes the SQL that you give
it as input, which could cause undesired load on a production server.
At the time of this release, there is a bug that causes the tool to crash, and a bug that causes a deadlock.
upgrade.

2.34. pt-upgrade 261


2.34.4 DESCRIPTION

pt-upgrade executes queries from slowlogs on one or more MySQL server to find differences in query time, warn-
ings, results, and other aspects of the queries’ execution. This helps evaluate upgrades, migrations and configuration
changes. The comparisons specified by --compare determine what differences can be found. A report is printed
which outlines all the differences found; see “OUTPUT” below.
The first DSN (host) specified on the command line is authoritative; it defines the results to which the other DSNs are
compared. You can “compare” only one host, in which case there will be no differences but the output can be saved to
be diffed later against the output of another single host “comparison”.
At present, pt-upgrade only reads slowlogs. Use pt-query-digest --print to transform other log formats to
slowlog.
DSNs and slowlog files can be specified in any order. pt-upgrade will automatically determine if an argument is a
DSN or a slowlog file. If no slowlog files are given and --query is not specified then pt-upgrade will read from
STDIN.

2.34.5 OUTPUT

Queries are group by fingerprints and any with differences are printed. The first part of a query report is a summary of
differences. In the example below, the query returns a different number of rows (row counts) on each server. The
second part is the side-by-side comparison of values obtained from the query on each server. Then a sample of the
query is printed, preceded by its ID which can be used to locate more information in the sub-report at the end. There
are sub-reports for various types of differences.
# Query 1: ID 0x3C830E3839B916D7 at byte 0 _______________________________
# Found 1 differences in 1 samples:
# column counts 0
# column types 0
# column values 0
# row counts 1
# warning counts 0
# warning levels 0
# warnings 0
# 127.1:12345 127.1:12348
# Errors 0 0
# Warnings 0 0
# Query_time
# sum 0 0
# min 0 0
# max 0 0
# avg 0 0
# pct_95 0 0
# stddev 0 0
# median 0 0
# row_count
# sum 4 3
# min 4 3
# max 4 3
# avg 4 3
# pct_95 4 3
# stddev 0 0
# median 4 3
use ‘test‘;
select i from t where i is not null



/* 3C830E3839B916D7-1 */ select i from t where i is not null

# Row count differences
# Query ID 127.1:12345 127.1:12348
# ================== =========== ===========
# 3C830E3839B916D7-1 4 3

The output will vary slightly depending on which options are speciﬁed.

2.34.6 OPTIONS

-ask-pass
-base-dir
type: string; default: /tmp
Save outﬁles for the rows comparison method in this directory.
See the rows --compare-results-method.
-charset
-[no]clear-warnings
default: yes
Clear warnings before each warnings comparison.
If comparing warnings (--compare includes warnings), this option causes pt-upgrade to execute a suc-
cessful SELECT statement which clears any warnings left over from previous queries. This requires a current
database that pt-upgrade usually detects automatically, but in some cases it might be necessary to specify
--temp-database. If pt-upgrade can’t auto-detect the current database, it will create a temporary table in
the --temp-database called mk_upgrade_clear_warnings.
-clear-warnings-table
type: string
Execute SELECT * FROM ... LIMIT 1 from this table to clear warnings.
-compare
type: Hash; default: query_times,results,warnings
What to compare for each query executed on each host.
Comparisons determine differences when the queries are executed on the hosts. More comparisons enable more
differences to be detected. The following comparisons are available:
query_times
Compare query execution times. If this comparison is disabled, the queries are still executed so that
other comparisons will work, but the query time attributes are removed from the events.
results



Compare result sets to find differences in rows, columns, etc.
What differences can be found depends on the --compare-results-method used.
warnings
Compare warnings from SHOW WARNINGS. Requires at least MySQL 4.1.
-compare-results-method
type: string; default: CHECKSUM; group: Comparisons
Method to use for --compare results. This option has no effect if --no-compare-results is given.
Available compare methods (case-insensitive):
CHECKSUM
Do CREATE TEMPORARY TABLE ‘mk_upgrade‘ AS query then CHECKSUM TABLE
‘mk_upgrade‘. This method is fast and simple but in rare cases might it be inaccurate because
the MySQL manual says:
[The] fact that two tables produce the same checksum does I<not> mean that
the tables are identical.

Requires at least MySQL 4.1.
rows
Compare rows one-by-one to find differences. This method has advantages and disadvantages. Its
disadvantages are that it may be slower and it requires writing and reading outfiles from disk. Its
advantages are that it is universal (works for all versions of MySQL), it doesn’t alter the query in any
way, and it can find column value differences.
The rows method works as follows:
1. Rows from each host are compared one-by-one.
2. If no differences are found, comparison stops, else...
3. All remain rows (after the point where they begin to differ)
are written to outfiles.
4. The outfiles are loaded into temporary tables with
C<LOAD DATA LOCAL INFILE>.
5. The temporary tables are analyzed to determine the differences.

The outfiles are written to the --base-dir.
-config
type: Array
-continue-on-error
Continue working even if there is an error.
-convert-to-select
Convert non-SELECT statements to SELECTs and compare.
By default non-SELECT statements are not allowed. This option causes non-SELECT statements (like UP-
DATE, INSERT and DELETE) to be converted to SELECT statements, executed and compared.
For example, DELETE col FROM tbl WHERE id=1 is converted to SELECT col FROM tbl
WHERE id=1.
-daemonize



-explain-hosts
Print connection information and exit.
-filter
type: string
argument: $event. This is a hashref. If the given value is a readable file, then pt-upgrade reads the entire file
and uses its contents as the code. The file should not contain a shebang (#!/usr/bin/perl) line.



If the filter code won’t compile, pt-upgrade will die with an error. If the filter code does compile, an error
may still occur at runtime if the code tries to do something wrong (like pattern match an undefined value).
pt-upgrade does not provide any safeguards so code carefully!
An example filter that discards everything but SELECT statements:

This is compiled into a subroutine like the following:
sub { $event = shift; ( $event->{arg} =~ m/^select/i ) && return $event; }

You can find an explanation of the structure of $event at http://code.google.com/p/maatkit/wiki/EventAttributes.
-fingerprints
Add query fingerprints to the standard query analysis report. This is mostly useful for debugging purposes.
-float-precision
type: int
Round float, double and decimal values to this many places.
This option helps eliminate false-positives caused by floating-point imprecision.
-help
Show help and exit.
-host



Connect to host.
-iterations
How many times to iterate through the collect-and-report cycle. If 0, iterate to infinity. See also –run-time.
-limit
type: string; default: 95%:20
Limit output to the given percentage or count.
If the argument is an integer, report only the top N worst queries. If the argument is an integer followed by the
% sign, report that percentage of the worst queries. If the percentage is followed by a colon and another integer,
report the top percentage or the number specified by that integer, whichever comes first.
-log
type: string
-max-different-rows
Stop comparing rows for --compare-results-method rows after this many differences are found.
-order-by
type: string; default: differences:sum
Sort events by this attribute and aggregate function.
-password
-pid
type: string
-port
-query
type: string
Execute and compare this single query; ignores files on command line.
This option allows you to supply a single query on the command line. Any slowlogs also specified on the
command line are ignored.
-reports
type: Hash; default: queries,differences,errors,statistics
Print these reports. Valid reports are queries, differences, errors, and statistics.
See “OUTPUT” for more information on the various parts of the report.
-run-time
type: time



How long to run before exiting. The default is to run forever (you can interrupt with CTRL-C).
-set-vars
type: string; default: wait_timeout=10000,query_cache_type=0
executed.
-shorten
Shorten long statements in reports.
Shortens long statements, replacing the omitted portion with a /*... omitted ...*/ comment. This
applies only to the output in reports. It prevents a large statement from causing difficulty in a report. The
argument is the preferred length of the shortened statement. Not all statements can be shortened, but very large
INSERT and similar statements often can; and so can IN() lists, although only the first such list in the statement
will be shortened.
If it shortens something beyond recognition, you can find the original statement in the log, at the offset shown
in the report header (see “OUTPUT”).
-socket
-temp-database
type: string
Use this database for creating temporary tables.
If given, this database is used for creating temporary tables for the results comparison (see --compare).
Otherwise, the current database (from the last event that specified its database) is used.
-temp-table
type: string; default: mk_upgrade
Use this table for checksumming results.
-user
-version
-zero-query-times
Zero the query times in the report.

2.34.7 DSN OPTIONS

sensitive, so P and p are not the same option. There cannot be whitespace before or after the =, and if the value
details.
• A



• D
Default database.
• F
• h
Connect to host.
• p
• P
• S
• u

2.34.8 ENVIRONMENT

PTDEBUG=1 pt-upgrade ... > FILE 2>&1



Perl.

2.34.10 BUGS

For a list of known bugs, see http://www.percona.com/bugs/pt-upgrade.



• Tool --version

2.34.11 DOWNLOADING






2.34.12 AUTHORS

Daniel Nichter







2.34.15 VERSION

pt-upgrade 2.1.1

2.35 pt-variable-advisor

2.35.1 NAME

pt-variable-advisor - Analyze MySQL variables and advise on possible problems.

2.35.2 SYNOPSIS

Usage

pt-variable-advisor [OPTION...] [DSN]

pt-variable-advisor analyzes variables and advises on possible problems.
Get SHOW VARIABLES from localhost:
pt-variable-advisor localhost

Get SHOW VARIABLES output saved in vars.txt:
pt-variable-advisor --source-of-variables vars.txt

2.35.3 RISKS

pt-variable-advisor reads MySQL’s conﬁguration and examines it and is thus very low risk.
variable-advisor.

2.35.4 DESCRIPTION

pt-variable-advisor examines SHOW VARIABLES for bad values and settings according to the “RULES” described
below. It reports on variables that match the rules, so you can ﬁnd bad settings in your MySQL server.
At the time of this release, pt-variable-advisor only examples SHOW VARIABLES, but other input sources are
planned like SHOW STATUS and SHOW SLAVE STATUS.



2.35.5 RULES

These are the rules that pt-variable-advisor will apply to SHOW VARIABLES. Each rule has three parts: an ID, a
severity, and a description.
The rule’s ID is a short, unique name for the rule. It usually relates to the variable that the rule examines. If a variable
is examined by several rules, then the rules’ IDs are numbered like “-1”, “-2”, “-N”.
The rule’s severity is an indication of how important it is that this rule matched a query. We use NOTE, WARN, and
CRIT to denote these levels.
The rule’s description is a textual, human-readable explanation of what it means when a variable matches this rule.
Depending on the verbosity of the report you generate, you will see more of the text in the description. By default,
you’ll see only the first sentence, which is sort of a terse synopsis of the rule’s meaning. At a higher verbosity, you’ll
see subsequent sentences.
auto_increment
severity: note
Are you trying to write to more than one server in a dual-master or ring replication configuration? This
is potentially very dangerous and in most cases is a serious mistake. Most people’s reasons for doing this
are actually not valid at all.
concurrent_insert
severity: note
Holes (spaces left by deletes) in MyISAM tables might never be reused.
connect_timeout
severity: note
A large value of this setting can create a denial of service vulnerability.
debug
severity: crit
Servers built with debugging capability should not be used in production because of the large performance
impact.
delay_key_write
severity: warn
MyISAM index blocks are never flushed until necessary. If there is a server crash, data corruption on
MyISAM tables can be much worse than usual.
flush
severity: warn
This option might decrease performance greatly.
flush_time
severity: warn
This option might decrease performance greatly.
have_bdb
severity: note
The BDB engine is deprecated. If you aren’t using it, you should disable it with the skip_bdb option.

2.35. pt-variable-advisor 271


init_connect
severity: note
The init_connect option is enabled on this server.
init_file
severity: note
The init_file option is enabled on this server.
init_slave
severity: note
The init_slave option is enabled on this server.
innodb_additional_mem_pool_size
severity: warn
This variable generally doesn’t need to be larger than 20MB.
innodb_buffer_pool_size
severity: warn
The InnoDB buffer pool size is unconfigured. In a production environment it should always be configured
explicitly, and the default 10MB size is not good.
innodb_checksums
severity: warn
InnoDB checksums are disabled. Your data is not protected from hardware corruption or other errors!
innodb_doublewrite
severity: warn
InnoDB doublewrite is disabled. Unless you use a filesystem that protects against partial page writes,
your data is not safe!
innodb_fast_shutdown
severity: warn
InnoDB’s shutdown behavior is not the default. This can lead to poor performance, or the need to perform
crash recovery upon startup.
innodb_flush_log_at_trx_commit-1
severity: warn
InnoDB is not configured in strictly ACID mode. If there is a crash, some transactions can be lost.
innodb_flush_log_at_trx_commit-2
severity: warn
Setting innodb_flush_log_at_trx_commit to 0 has no performance benefits over setting it to 2, and more
types of data loss are possible. If you are trying to change it from 1 for performance reasons, you should
set it to 2 instead of 0.
innodb_force_recovery



severity: warn
InnoDB is in forced recovery mode! This should be used only temporarily when recovering from data
corruption or other bugs, not for normal usage.
innodb_lock_wait_timeout
severity: warn
This option has an unusually long value, which can cause system overload if locks are not being released.
innodb_log_buffer_size
severity: warn
The InnoDB log buffer size generally should not be set larger than 16MB. If you are doing large BLOB
operations, InnoDB is not really a good choice of engines anyway.
innodb_log_file_size
severity: warn
The InnoDB log file size is set to its default value, which is not usable on production systems.
innodb_max_dirty_pages_pct
severity: note
The innodb_max_dirty_pages_pct is lower than the default. This can cause overly aggressive flushing and
add load to the I/O system.
flush_time
severity: warn
This setting is likely to cause very bad performance every flush_time seconds.
key_buffer_size
severity: warn
The key buffer size is unconfigured. In a production environment it should always be configured explicitly,
and the default 8MB size is not good.
large_pages
severity: note
Large pages are enabled.
locked_in_memory
severity: note
The server is locked in memory with –memlock.
log_warnings-1
severity: note
Log_warnings is disabled, so unusual events such as statements unsafe for replication and aborted con-
nections will not be logged to the error log.
log_warnings-2
severity: note
Log_warnings must be set greater than 1 to log unusual events such as aborted connections.
low_priority_updates



severity: note
The server is running with non-default lock priority for updates. This could cause update queries to wait
unexpectedly for read queries.
max_binlog_size
severity: note
The max_binlog_size is smaller than the default of 1GB.
max_connect_errors
severity: note
max_connect_errors should probably be set as large as your platform allows.
max_connections
severity: warn
If the server ever really has more than a thousand threads running, then the system is likely to spend more
time scheduling threads than really doing useful work. This variable’s value should be considered in light
of your workload.
myisam_repair_threads
severity: note
myisam_repair_threads > 1 enables multi-threaded repair, which is relatively untested and is still listed as
beta-quality code in the ofﬁcial documentation.
old_passwords
severity: warn
Old-style passwords are insecure. They are sent in plain text across the wire.
optimizer_prune_level
severity: warn
The optimizer will use an exhaustive search when planning complex queries, which can cause the planning
process to take a long time.
port
severity: note
The server is listening on a non-default port.
query_cache_size-1
severity: note
The query cache does not scale to large sizes and can cause unstable performance when larger than
128MB, especially on multi-core machines.
query_cache_size-2
severity: warn
The query cache can cause severe performance problems when it is larger than 256MB, especially on
multi-core machines.
read_buffer_size-1



severity: note
The read_buffer_size variable should generally be left at its default unless an expert determines it is
necessary to change it.
read_buffer_size-2
severity: warn
The read_buffer_size variable should not be larger than 8MB. It should generally be left at its default
unless an expert determines it is necessary to change it. Making it larger than 2MB can hurt performance
significantly, and can make the server crash, swap to death, or just become extremely unstable.
read_rnd_buffer_size-1
severity: note
The read_rnd_buffer_size variable should generally be left at its default unless an expert determines it is
necessary to change it.
read_rnd_buffer_size-2
severity: warn
The read_rnd_buffer_size variable should not be larger than 4M. It should generally be left at its default
unless an expert determines it is necessary to change it.
relay_log_space_limit
severity: warn
Setting relay_log_space_limit can cause replicas to stop fetching binary logs from their master immedi-
ately. This could increase the risk that your data will be lost if the master crashes. If the replicas have
encountered a limit on relay log space, then it is possible that the latest transactions exist only on the
master and no replica has retrieved them.
slave_net_timeout
severity: warn
This variable is set too high. This is too long to wait before noticing that the connection to the master
has failed and retrying. This should probably be set to 60 seconds or less. It is also a good idea to use
pt-heartbeat to ensure that the connection does not appear to time out when the master is simply idle.
slave_skip_errors
severity: crit
You should not set this option. If replication is having errors, you need to find and resolve the cause of that;
it is likely that your slave’s data is different from the master. You can find out with pt-table-checksum.
sort_buffer_size-1
severity: note
The sort_buffer_size variable should generally be left at its default unless an expert determines it is nec-
essary to change it.
sort_buffer_size-2
severity: note
The sort_buffer_size variable should generally be left at its default unless an expert determines it is nec-
essary to change it. Making it larger than a few MB can hurt performance significantly, and can make the
server crash, swap to death, or just become extremely unstable.
sql_notes



severity: note
This server is configured not to log Note level warnings to the error log.
sync_frm
severity: warn
It is best to set sync_frm so that .frm files are flushed safely to disk in case of a server crash.
tx_isolation-1
severity: note
This server’s transaction isolation level is non-default.
tx_isolation-2
severity: warn
Most applications should use the default REPEATABLE-READ transaction isolation level, or in a few
cases READ-COMMITTED.
expire_log_days
severity: warn
Binary logs are enabled, but automatic purging is not enabled. If you do not purge binary logs, your disk
will fill up. If you delete binary logs externally to MySQL, you will cause unwanted behaviors. Always
ask MySQL to purge obsolete logs, never delete them externally.
innodb_file_io_threads
severity: note
This option is useless except on Windows.
innodb_data_file_path
severity: note
Auto-extending InnoDB files can consume a lot of disk space that is very difficult to reclaim later. Some
people prefer to set innodb_file_per_table and allocate a fixed-size file for ibdata1.
innodb_flush_method
severity: note
Most production database servers that use InnoDB should set innodb_flush_method to O_DIRECT to
avoid double-buffering, unless the I/O system is very low performance.
innodb_locks_unsafe_for_binlog
severity: warn
This option makes point-in-time recovery from binary logs, and replication, untrustworthy if statement-
based logging is used.
innodb_support_xa
severity: warn
MySQL’s internal XA transaction support between InnoDB and the binary log is disabled. The binary
log might not match InnoDB’s state after crash recovery, and replication might drift out of sync due to
out-of-order statements in the binary log.
log_bin



severity: warn
Binary logging is disabled, so point-in-time recovery and replication are not possible.
log_output
severity: warn
Directing log output to tables has a high performance impact.
max_relay_log_size
severity: note
A custom max_relay_log_size is defined.
myisam_recover_options
severity: warn
myisam_recover_options should be set to some value such as BACKUP,FORCE to ensure that table cor-
ruption is noticed.
storage_engine
severity: note
The server is using a non-standard storage engine as default.
sync_binlog
severity: warn
Binary logging is enabled, but sync_binlog isn’t configured so that every transaction is flushed to the
binary log for durability.
tmp_table_size
severity: note
The effective minimum size of in-memory implicit temporary tables used internally during query execu-
tion is min(tmp_table_size, max_heap_table_size), so max_heap_table_size should be at least as large as
tmp_table_size.
old mysql version
severity: warn
These are the recommended minimum version for each major release: 3.23, 4.1.20, 5.0.37, 5.1.30.
end-of-life mysql version
severity: note
Every release older than 5.1 is now officially end-of-life.

2.35.6 OPTIONS

-ask-pass
-charset



-config
type: Array
-daemonize
-defaults-file
-help
Show help and exit.
-host
Connect to host.
-ignore-rules
type: hash
Ignore these rule IDs.
Specify a comma-separated list of rule IDs (e.g. LIT.001,RES.002,etc.) to ignore.
-password
-pid
type: string
-port
-set-vars
executed.
-socket
-source-of-variables
type: string; default: mysql
Read SHOW VARIABLES from this source. Possible values are “mysql”, “none” or a ﬁle name. If “mysql” is
speciﬁed then you must also specify a DSN on the command line.



-user
-verbose
Increase verbosity of output. At the default level of verbosity, the program prints only the ﬁrst sentence of each
rule’s description. At higher levels, the program prints more of the description.
-version

2.35.7 DSN OPTIONS

details.
• A
• D
Default database.
• F
• h
Connect to host.
• p
• P
• S
• u




2.35.8 ENVIRONMENT

PTDEBUG=1 pt-variable-advisor ... > FILE 2>&1



Perl.

2.35.10 BUGS

For a list of known bugs, see http://www.percona.com/bugs/pt-variable-advisor.
• Tool --version

2.35.11 DOWNLOADING






2.35.12 AUTHORS








2.35.15 VERSION

pt-variable-advisor 2.1.1

2.36 pt-visual-explain

2.36.1 NAME

pt-visual-explain - Format EXPLAIN output as a tree.

2.36.2 SYNOPSIS

Usage

pt-visual-explain [OPTION...] [FILE...]

pt-visual-explain transforms EXPLAIN output into a tree representation of the query plan. If FILE is given, input is
read from the ﬁle(s). With no FILE, or when FILE is -, read standard input.

Examples

pt-visual-explain <file_containing_explain_output>

pt-visual-explain -c <file_containing_query>

mysql -e "explain select * from mysql.user" | pt-visual-explain

2.36. pt-visual-explain 281


2.36.3 RISKS

pt-visual-explain is read-only and very low-risk.
visual-explain.

2.36.4 DESCRIPTION

pt-visual-explain reverse-engineers MySQL’s EXPLAIN output into a query execution plan, which it then formats
as a left-deep tree – the same way the plan is represented inside MySQL. It is possible to do this by hand, or to
read EXPLAIN’s output directly, but it requires patience and expertise. Many people find a tree representation more
understandable.
You can pipe input into pt-visual-explain or specify a filename at the command line, including the magical ‘-‘ file-
name, which will read from standard input. It can do two things with the input: parse it for something that looks like
EXPLAIN output, or connect to a MySQL instance and run EXPLAIN on the input.
When parsing its input, pt-visual-explain understands three formats: tabular like that shown in the mysql command-
line client, vertical like that created by using the G line terminator in the mysql command-line client, and tab separated.
It ignores any lines it doesn’t know how to parse.
When executing the input, pt-visual-explain replaces everything in the input up to the first SELECT keyword with
‘EXPLAIN SELECT,’ and then executes the result. You must specify --connect to execute the input as a query.
Either way, it builds a tree from the result set and prints it to standard output. For the following query,
select * from sakila.film_actor join sakila.film using(film_id);

pt-visual-explain generates this query plan:
JOIN
+- Bookmark lookup
| +- Table
| | table film_actor
| | possible_keys idx_fk_film_id
| +- Index lookup
| key film_actor->idx_fk_film_id
| possible_keys idx_fk_film_id
| key_len 2
| ref sakila.film.film_id
| rows 2
+- Table scan
rows 952
+- Table
table film
possible_keys PRIMARY

The query plan is left-deep, depth-first search, and the tree’s root is the output node – the last step in the execution
plan. In other words, read it like this:



1
Table scan the ‘film’ table, which accesses an estimated 952 rows.
2
For each row, find matching rows by doing an index lookup into the film_actor->idx_fk_film_id index
with the value from sakila.film.film_id, then a bookmark lookup into the film_actor table.
For more information on how to read EXPLAIN output, please see http://dev.mysql.com/doc/en/explain.html, and this
talk titled “Query Optimizer Internals and What’s New in the MySQL 5.2 Optimizer,” from Timour Katchaounov, one
of the MySQL developers: http://maatkit.org/presentations/katchaounov_timour.pdf.

2.36.5 MODULES

This program is actually a runnable module, not just an ordinary Perl script. In fact, there are two modules embedded
in it. This makes unit testing easy, but it also makes it easy for you to use the parsing and tree-building functionality if
you want.
The ExplainParser package accepts a string and parses whatever it thinks looks like EXPLAIN output from it. The
synopsis is as follows:
require "pt-visual-explain";
my $p = ExplainParser->new();
my $rows = $p->parse("some text");
# $rows is an arrayref of hashrefs.

The ExplainTree package accepts a set of rows and turns it into a tree. For convenience, you can also have it delegate
to ExplainParser and parse text for you. Here’s the synopsis:
require "pt-visual-explain";
my $e = ExplainTree->new();
my $tree = $e->parse("some text", %options);
my $output = $e->pretty_print($tree);
print $tree;

2.36.6 ALGORITHM

This section explains the algorithm that converts EXPLAIN into a tree. You may be interested in reading this if
you want to understand EXPLAIN more fully, or trying to figure out how this works, but otherwise this section will
probably not make your life richer.
The tree can be built by examining the id, select_type, and table columns of each row. Here’s what I know about them:
The id column is the sequential number of the select. This does not indicate nesting; it just comes from counting
SELECT from the left of the SQL statement. It’s like capturing parentheses in a regular expression. A UNION
RESULT row doesn’t have an id, because it isn’t a SELECT. The source code actually refers to UNIONs as a fake_lex,
as I recall.
If two adjacent rows have the same id value, they are joined with the standard single-sweep multi-join method.
The select_type column tells a) that a new sub-scope has opened b) what kind of relationship the row has to the
previous row c) what kind of operation the row represents.
• SIMPLE means there are no subqueries or unions in the whole query.
• PRIMARY means there are, but this is the outermost SELECT.
• [DEPENDENT] UNION means this result is UNIONed with the previous result (not row; a result might encom-
pass more than one row).



• UNION RESULT terminates a set of UNIONed results.
• [DEPENDENT|UNCACHEABLE] SUBQUERY means a new sub-scope is opening. This is the kind of sub-
query that happens in a WHERE clause, SELECT list or whatnot; it does not return a so-called “derived table.”
• DERIVED is a subquery in the FROM clause.
Tables that are JOINed all have the same select_type. For example, if you JOIN three tables inside a dependent
subquery, they’ll all say the same thing: DEPENDENT SUBQUERY.
The table column usually speciﬁes the table name or alias, but may also say <derivedN> or <unionN,N...N>. If it says
<derivedN>, the row represents an access to the temporary table that holds the result of the subquery whose id is N. If
it says <unionN,..N> it’s the same thing, but it refers to the results it UNIONs together.
Finally, order matters. If a row’s id is less than the one before it, I think that means it is dependent on something other
than the one before it. For example,
explain select
(select 1 from sakila.film),
(select 2 from sakila.film_actor),
(select 3 from sakila.actor);

| id | select_type | table |
+----+-------------+------------+
| 1 | PRIMARY | NULL |
| 4 | SUBQUERY | actor |
| 3 | SUBQUERY | film_actor |
| 2 | SUBQUERY | film |

If the results were in order 2-3-4, I think that would mean 3 is a subquery of 2, 4 is a subquery of 3. As it is, this means
4 is a subquery of the nearest previous recent row with a smaller id, which is 1. Likewise for 3 and 2.
This structure is hard to programmatically build into a tree for the same reason it’s hard to understand by inspection:
there are both forward and backward references. <derivedN> is a forward reference to selectN, while <unionM,N> is a
backward reference to selectM and selectN. That makes recursion and other tree-building algorithms hard to get right
(NOTE: after implementation, I now see how it would be possible to deal with both forward and backward references,
but I have no motivation to change something that works). Consider the following:
select * from (
select 1 from sakila.actor as actor_1
union
) as der_1
union
select * from (
union all
) as der_2;

+------+--------------+------------+
| 1 | PRIMARY | <derived2> |
| 2 | DERIVED | actor_1 |
| 3 | UNION | actor_2 |
| NULL | UNION RESULT | <union2,3> |
| 4 | UNION | <derived5> |




This would be a lot easier to work with if it looked like this (I’ve bracketed the id on rows I moved):
+------+--------------+------------+
| [1] | UNION RESULT | <union1,4> |
| 4 | UNION | <derived5> |

In fact, why not re-number all the ids, so the PRIMARY row becomes 2, and so on? That would make it even easier
to read. Unfortunately that would also have the effect of destroying the meaning of the id column, which I think is
important to preserve in the final tree. Also, though it makes it easier to read, it doesn’t make it easier to manipulate
programmatically; so it’s fine to leave them numbered as they are.
The goal of re-ordering is to make it easier to figure out which rows are children of which rows in the execution plan.
Given the reordered list and some row whose table is <union...> or <derived>, it is easy to find the beginning of the
slice of rows that should be child nodes in the tree: you just look for the first row whose ID is the same as the first
number in the table.
The next question is how to find the last row that should be a child node of a UNION or DERIVED. I’ll start with
DERIVED, because the solution makes UNION easy.
Consider how MySQL numbers the SELECTs sequentially according to their position in the SQL, left-to-right. Since
a DERIVED table encloses everything within it in a scope, which becomes a temporary table, there are only two things
to think about: its child subqueries and unions (if any), and its next siblings in the scope that encloses it. Its children
will all have an id greater than it does, by definition, so any later rows with a smaller id terminate the scope.
Here’s an example. The middle derived table here has a subquery and a UNION to make it a little more complex for
the example.
explain select 1
from (
select film_id from sakila.film limit 1
) as der_1
join (
select film_id, actor_id, (select count(*) from sakila.rental) as r
from sakila.film_actor limit 1
union all
select 1, 1, 1 from sakila.film_actor as dummy
) as der_2 using (film_id)
join (
select actor_id from sakila.actor limit 1
) as der_3 using (actor_id);

Here’s the output of EXPLAIN:
| 6 | DERIVED | actor |
| 3 | DERIVED | film_actor |



| 4 | SUBQUERY | rental |
| 5 | UNION | dummy |
| 2 | DERIVED | film |

The siblings all have id 1, and the middle one I care about is derived3. (Notice MySQL doesn’t execute them in the
order I defined them, which is fine). Now notice that MySQL prints out the rows in the opposite order I defined the
subqueries: 6, 3, 2. It always seems to do this, and there might be other methods of finding the scope boundaries
including looking for the lower boundary of the next largest sibling, but this is a good enough heuristic. I am forced
to rely on it for non-DERIVED subqueries, so I rely on it here too. Therefore, I decide that everything greater than or
equal to 3 belongs to the DERIVED scope.
The rule for UNION is simple: they consume the entire enclosing scope, and to find the component parts of each one,
you find each part’s beginning as referred to in the <unionN,...> definition, and its end is either just before the next
one, or if it’s the last part, the end is the end of the scope.
This is only simple because UNION consumes the entire scope, which is either the entire statement, or the scope of a
DERIVED table. This is because a UNION cannot be a sibling of another UNION or a table, DERIVED or not. (Try
writing such a statement if you don’t see it intuitively). Therefore, you can just find the enclosing scope’s boundaries,
and the rest is easy. Notice in the example above, the UNION is over <union3,5>, which includes the row with id 4 –
it includes every row between 3 and 5.
Finally, there are non-derived subqueries to deal with as well. In this case I can’t look at siblings to find the end of the
scope as I did for DERIVED. I have to trust that MySQL executes depth-first. Here’s an example:
explain
select actor_id,
(
select count(film_id)
+ (select count(*) from sakila.film)
from sakila.film join sakila.film_actor using(film_id)
where exists(
select * from sakila.actor
where sakila.actor.actor_id = sakila.film_actor.actor_id
)
)
from sakila.actor;

| 1 | PRIMARY | actor |
| 4 | DEPENDENT SUBQUERY | actor |

In order, the tree should be built like this:
• See row 1.
• See row 2. It’s a higher id than 1, so it’s a subquery, along with every other row whose id is greater than 2.
• Inside this scope, see 2 and 2 and JOIN them. See 4. It’s a higher id than 2, so it’s again a subquery; recurse.
After that, see 3, which is also higher; recurse.
But the only reason the nested subquery didn’t include select 3 is because select 4 came first. In other words, if
EXPLAIN looked like this,
| 1 | PRIMARY | actor |



| 4 | DEPENDENT SUBQUERY | actor |

I would be forced to assume upon seeing select 3 that select 4 is a subquery of it, rather than just being the next sibling
in the enclosing scope. If this is ever wrong, then the algorithm is wrong, and I don’t see what could be done about it.
UNION is a little more complicated than just “the entire scope is a UNION,” because the UNION might itself be inside
an enclosing scope that’s only indicated by the first item inside the UNION. There are only three kinds of enclosing
scopes: UNION, DERIVED, and SUBQUERY. A UNION can’t enclose a UNION, and a DERIVED has its own
“scope markers,” but a SUBQUERY can wholly enclose a UNION, like this strange example on the empty table t1:
explain select * from t1 where not exists(
(select t11.i from t1 t11) union (select t12.i from t1 t12));

| id | select_type | table | Extra |
+------+--------------+------------+--------------------------------+
| 1 | PRIMARY | t1 | const row not found |
| 2 | SUBQUERY | NULL | No tables used |
| 3 | SUBQUERY | NULL | no matching row in const table |
| 4 | UNION | t12 | const row not found |
| NULL | UNION RESULT | <union2,4> | |

The UNION’s backward references might make it look like the UNION encloses the subquery, but studying the query
makes it clear this isn’t the case. So when a UNION’s first row says SUBQUERY, it is this special case.
By the way, I don’t fully understand this query plan; there are 4 numbered SELECT in the plan, but only 3 in the
query. The parens around the UNIONs are meaningful. Removing them will make the EXPLAIN different. Please
tell me how and why this works if you know.
Armed with this knowledge, it’s possible to use recursion to turn the parent-child relationship between all the rows
into a tree representing the execution plan.
MySQL prints the rows in execution order, even the forward and backward references. At any given scope, the rows
are processed as a left-deep tree. MySQL does not do “bushy” execution plans. It begins with a table, finds a matching
row in the next table, and continues till the last table, when it emits a row. When it runs out, it backtracks till it can
find the next row and repeats. There are subtleties of course, but this is the basic plan. This is why MySQL transforms
all RIGHT OUTER JOINs into LEFT OUTER JOINs and cannot do FULL OUTER JOIN.
This means in any given scope, say
| 1 | SIMPLE | tbl1 |

The execution plan looks like a depth-first traversal of this tree:
JOIN
/
JOIN tbl3
/
tbl1 tbl2

The JOIN might not be a JOIN. It might be a subquery, for example. This comes from the type column of EXPLAIN.
The documentation says this is a “join type,” but I think “access type” is more accurate, because it’s “how MySQL
accesses rows.”
pt-visual-explain decorates the tree significantly more than just turning rows into nodes. Each node may get a series
of transformations that turn it into a subtree of more than one node. For example, an index scan not marked with



‘Using index’ must do a bookmark lookup into the table rows; that is a three-node subtree. However, after the above
node-ordering and scoping stuff, the rest of the process is pretty simple.

2.36.7 OPTIONS

-ask-pass
-charset
-clustered-pk
Assume that PRIMARY KEY index accesses don’t need to do a bookmark lookup to retrieve rows. This is the
case for InnoDB.
-config
type: Array
-connect
Treat input as a query, and obtain EXPLAIN output by connecting to a MySQL instance and running EXPLAIN
on the query. When this option is given, pt-visual-explain uses the other connection-specific options such as
--user to connect to the MySQL instance. If you have a .my.cnf file, it will read it, so you may not need to
specify any connection-specific options.
-database
Connect to this database.
-defaults-file
-format
type: string; default: tree
Set output format.
The default is a terse pretty-printed tree. The valid values are:
Value Meaning
===== ================================================
tree Pretty-printed terse tree.
dump Data::Dumper output (see Data::Dumper for more).

-help
Show help and exit.
-host
Connect to host.



-password
-pid
type: string
-port
-set-vars
executed.
-socket
-user
-version

2.36.8 DSN OPTIONS

details.
• A
• D
Default database.
• F
• h



Connect to host.
• p
• P
• S
• u

2.36.9 ENVIRONMENT

PTDEBUG=1 pt-visual-explain ... > FILE 2>&1



Perl.

2.36.11 BUGS

For a list of known bugs, see http://www.percona.com/bugs/pt-visual-explain.
• Tool --version



2.36.12 DOWNLOADING






2.36.13 AUTHORS

Baron Schwartz




welcome.

2.36.16 VERSION

pt-visual-explain 2.1.1


CHAPTER

THREE

CONFIGURATION

3.1 CONFIGURATION FILES

Percona Toolkit tools can read options from configuration files. The configuration file syntax is simple and direct,
and bears some resemblances to the MySQL command-line client tools. The configuration files all follow the same
conventions.
Internally, what actually happens is that the lines are read from the file and then added as command-line options and
arguments to the tool, so just think of the configuration files as a way to write your command lines.

3.1.1 SYNTAX

The syntax of the configuration files is as follows:
*
Whitespace followed by a hash sign (#) signifies that the rest of the line is a comment. This is deleted.
For example:
*
Whitespace is stripped from the beginning and end of all lines.
*
Empty lines are ignored.
*
Each line is permitted to be in either of the following formats:
option
option=value

Do not prefix the option with --. Do not quote the values, even if it has spaces; value are literal. Whites-
pace around the equals sign is deleted during processing.
*
Only long options are recognized.
*
A line containing only two hyphens signals the end of option parsing. Any further lines are interpreted as
additional arguments (not options) to the program.

293


3.1.2 EXAMPLE

This config file for pt-stalk,
# Config for pt-stalk
variable=Threads_connected
cycles=2 # trigger if problem seen twice in a row
--
--user daniel

is equivalent to this command line:
pt-stalk --variable Threads_connected --cycles 2 -- --user daniel

Options after -- are passed literally to mysql and mysqladmin.

3.1.3 READ ORDER

The tools read several configuration files in order:
1. The global Percona Toolkit configuration file, /etc/percona-toolkit/percona-toolkit.conf. All tools read this file,
so you should only add options to it that you want to apply to all tools.
2. The global tool-specific configuration file, /etc/percona-toolkit/TOOL.conf, where TOOL is a tool name like
pt-query-digest. This file is named after the specific tool you’re using, so you can add options that apply
only to that tool.
3. The user’s own Percona Toolkit configuration file, $HOME/.percona-toolkit.conf. All tools read this file, so you
should only add options to it that you want to apply to all tools.
4. The user’s tool-specific configuration file, $HOME/.TOOL.conf, where TOOL is a tool name like
pt-query-digest. This file is named after the specific tool you’re using, so you can add options that
apply only to that tool.

3.1.4 SPECIFYING

There is a special --config option, which lets you specify which configuration files Percona Toolkit should read.
You specify a comma-separated list of files. However, its behavior is not like other command-line options. It must
be given first on the command line, before any other options. If you try to specify it anywhere else, it will cause an
error. Also, you cannot specify --config=/path/to/file; you must specify the option and the path to the file
separated by whitespace without an equal sign between them, like:
--config /path/to/file

If you don’t want any configuration files at all, specify --config ” to provide an empty list of files.

3.2 DSN (DATA SOURCE NAME) SPECIFICATIONS

Percona Toolkit tools use DSNs to specify how to create a DBD connection to a MySQL server. A DSN is a comma-
separated string of key=value parts, like:
h=host1,P=3306,u=bob

294 Chapter 3. Configuration


The standard key parts are shown below, but some tools add additional key parts. See each tool’s documentation for
details.
Some tools do not use DSNs but still connect to MySQL using options like --host, --user, and --password.
Such tools uses these options to create a DSN automatically, behind the scenes.
Other tools uses both DSNs and options like the ones above. The options provide defaults for all DSNs that do
not specify the option’s corresponding key part. For example, if DSN h=host1 and option --port=12345 are
specified, then the tool automatically adds P=12345 to DSN.

3.2.1 KEY PARTS

Many of the tools add more parts to DSNs for special purposes, and sometimes override parts to make them do
something slightly different. However, all the tools support at least the following:
A
Specifies the default character set for the connection.
Enables character set settings in Perl and MySQL. If the value is utf8, sets Perl’s binmode on STDOUT
to utf8, passes the mysql_enable_utf8 option to DBD::mysql, and runs SET NAMES UTF8 after
connecting to MySQL. Any other value sets binmode on STDOUT without the utf8 layer, and runs SET
NAMES after connecting to MySQL.
Unfortunately, there is no way from within Perl itself to specify the client library’s character set. SET
NAMES only affects the server; if the client library’s settings don’t match, there could be problems. You
can use the defaults file to specify the client library’s character set, however. See the description of the F
part below.
D
Specifies the connection’s default database.
F
Specifies a defaults file the mysql client library (the C client library used by DBD::mysql, not Percona
Toolkit itself ) should read. The tools all read the [client] section within the defaults file. If you omit
this, the standard defaults files will be read in the usual order. “Standard” varies from system to system,
because the filenames to read are compiled into the client library. On Debian systems, for example, it’s
usually /etc/mysql/my.cnf then ~/.my.cnf. If you place the following into ~/.my.cnf, tools will Do The
Right Thing:
[client]
user=your_user_name
pass=secret

Omitting the F part is usually the right thing to do. As long as you have configured your ~/.my.cnf
correctly, that will result in tools connecting automatically without needing a username or password.
You can also specify a default character set in the defaults file. Unlike the “A” part described above, this
will actually instruct the client library (DBD::mysql) to change the character set it uses internally, which
cannot be accomplished any other way as far as I know, except for utf8.
h
Hostname or IP address for the connection.
p
P

3.2. DSN (DATA SOURCE NAME) SPECIFICATIONS 295


Port number to use for the connection. Note that the usual special-case behaviors apply: if you specify
localhost as your hostname on Unix systems, the connection actually uses a socket file, not a TCP/IP
connection, and thus ignores the port.
S
Socket file to use for the connection (on Unix systems).
u

3.2.2 BAREWORD

Many of the tools will let you specify a DSN as a single word, without any key=value syntax. This is called a
‘bareword’. How this is handled is tool-specific, but it is usually interpreted as the “h” part. The tool’s --help output
will tell you the behavior for that tool.

3.2.3 PROPAGATION

Many tools will let you propagate values from one DSN to the next, so you don’t have to specify all the parts for each
DSN. For example, if you want to specify a username and password for each DSN, you can connect to three hosts as
follows:
h=host1,u=fred,p=wilma host2 host3

This is tool-specific.

3.3 ENVIRONMENT

PTDEBUG=1 pt-table-checksum ... > FILE 2>&1


3.4 SYSTEM REQUIREMENTS

Most tools require:
* Perl v5.8 or newer
* Bash v3 or newer
* Core Perl modules like Time::HiRes
Tools that connect to MySQL require:
* Perl modules DBI and DBD::mysql
* MySQL 5.0 or newer
Percona Toolkit is only tested on UNIX systems, primarily Debian and Red Hat derivatives; other operating systems
are not supported.



Tools that connect to MySQL may work with MySQL v4.1, but this is not test or supported.

3.4. SYSTEM REQUIREMENTS 297

CHAPTER

FOUR

MISCELLANEOUS

4.1 BUGS

* Complete command-line used to run the tool
* Tool --version
* MySQL version of all servers involved
* Output from the tool including STDERR
* Input files (log/dump/config files, etc.)

4.2 AUTHORS

Percona Toolkit is primarily developed by Baron Schwartz and Daniel Nichter, both of whom are employed by Percona
Inc. See each program’s documentation for details.

4.3 COPYRIGHT, LICENSE, AND WARRANTY

Percona Toolkit is copyright 2011-2012 Percona Inc. and others. See each program’s documentation for complete
copyright notices.

299


4.4 VERSION

Percona Toolkit v2.1.1 released 2012-04-03

4.5 Release Notes

4.5.1 v2.1.1 released 2012-04-03

Percona Toolkit 2.1.1 has been released. This is the first release in the new 2.1 series which supersedes the 2.0 series.
We will continue to fix bugs in 2.0, but 2.1 is now the focus of development.
2.1 introduces a lot of new code for:
• pt-online-schema-change (completely redesigned)
• pt-mysql-summary (completely redesigned)
• pt-summary (completely redesigned)
• pt-fingerprint (new tool)
• pt-table-usage (new tool)
There were also several bug fixes.
The redesigned tools are meant to replace their 2.0 counterparts because the 2.1 versions have the same or more
functionality and they are simpler and more reliable. pt-online-schema-change was particularly enhanced to be as safe
as possible given that the tool is inherently risky.
Percona Toolkit packages can be downloaded from http://www.percona.com/downloads/percona-toolkit/ or the Per-
cona Software Repositories (http://www.percona.com/software/repositories/).

Changelog

• Completely redesigned pt-online-schema-change
• Completely redesigned pt-mysql-summary
• Completely redesigned pt-summary
• Added new tool: pt-table-usage
• Added new tool: pt-fingerprint
• Fixed bug 955860: pt-stalk doesn’t run vmstat, iostat, and mpstat for –run-time
• Fixed bug 960513: SHOW TABLE STATUS is used needlessly
• Fixed bug 969726: pt-online-schema-change loses foreign keys
• Fixed bug 846028: pt-online-schema-change does not show progress until completed
• Fixed bug 898695: pt-online-schema-change add useless ORDER BY
• Fixed bug 952727: pt-diskstats shows incorrect wr_mb_s
• Fixed bug 963225: pt-query-digest fails to set history columns for disk tmp tables and disk filesort
• Fixed bug 967451: Char chunking doesn’t quote column name
• Fixed bug 972399: pt-table-checksum docs are not rendered right

300 Chapter 4. Miscellaneous


• Fixed bug 896553: Various documentation spelling fixes
• Fixed bug 949154: pt-variable-advisor advice for relay-log-space-limit
• Fixed bug 953461: pt-upgrade manual broken ‘output’ section
• Fixed bug 949653: pt-table-checksum docs don’t mention risks posed by inconsistent schemas

4.5.2 v2.0.4 released 2012-03-07

Percona Toolkit 2.0.4 has been released. 23 bugs were fixed in this release, and three new features were implemented.
First, –filter was added to pt-kill which allows for arbitrary –group-by. Second, pt-online-schema-change now requires
that its new –execute option be given, else the tool will just check the tables and exit. This is a safeguard to encourage
users to read the documentation, particularly when replication is involved. Third, pt-stalk also received a new option:
–[no]stalk. To collect immediately without stalking, specify –no-stalk and the tool will collect once and exit.
This release is completely backwards compatible with previous 2.0 releases. Given the number of bug fixes, it’s worth
upgrading to 2.0.4.

Changelog

• Added –filter to pt-kill to allow arbitrary –group-by
• Added –[no]stalk to pt-stalk (bug 932331)
• Added –execute to pt-online-schema-change (bug 933232)
• Fixed bug 873598: pt-online-schema-change doesn’t like reserved words in column names
• Fixed bug 928966: pt-pmp still uses insecure /tmp
• Fixed bug 933232: pt-online-schema-change can break replication
• Fixed bug 941225: Use of qw(...) as parentheses is deprecated at pt-kill line 3511
• Fixed bug 821694: pt-query-digest doesn’t recognize hex InnoDB txn IDs
• Fixed bug 894255: pt-kill shouldn’t check if STDIN is a tty when –daemonize is given
• Fixed bug 916999: pt-table-checksum error: DBD::mysql::st execute failed: called with 2 bind variables when
6 are needed
• Fixed bug 926598: DBD::mysql bug causes pt-upgrade to use wrong precision (M) and scale (D)
• Fixed bug 928226: pt-diskstats illegal division by zero
• Fixed bug 928415: Typo in pt-stalk doc: –trigger should be –function
• Fixed bug 930317: pt-archiver doc refers to nonexistent pt-query-profiler
• Fixed bug 930533: pt-sift looking for *-processlist1; broken compatibility with pt-stalk
• Fixed bug 932331: pt-stalk cannot collect without stalking
• Fixed bug 932442: pt-table-checksum error when column name has two spaces
• Fixed bug 932883: File Debian bug after each release
• Fixed bug 940503: pt-stalk disk space checks wrong on 32bit platforms
• Fixed bug 944420: –daemonize doesn’t always close STDIN
• Fixed bug 945834: pt-sift invokes pt-diskstats with deprecated argument
• Fixed bug 945836: pt-sift prints awk error if there are no stack traces to aggregate

4.5. Release Notes 301


• Fixed bug 945842: pt-sift generates wrong state sum during processlist analysis
• Fixed bug 946438: pt-query-digest should print a better message when an unsupported log format is specified
• Fixed bug 946776: pt-table-checksum ignores –lock-wait-timeout
• Fixed bug 940440: Bad grammar in pt-kill docs

4.5.3 v2.0.3 released 2012-02-03

Percona Toolkit 2.0.3 has been released. The development team was very busy last month making this release signifi-
cant: two completely redesigned and improved tools, pt-diskstats and pt-stalk, and 20 bug fixes.
Both pt-diskstats and pt-stalk were redesigned and rewritten from the ground up. This allowed us to greatly improve
these tools’ functionality and increase testing for them. The accuracy and output of pt-diskstats was enhanced, and the
tool was rewritten in Perl. pt-collect was removed and its functionality was put into a new, enhanced pt-stalk. pt-stalk
is now designed to be a stable, long-running daemon on a variety of common platforms. It is worth re-reading the
documentation for each of these tools.
The 20 bug fixes cover a wide range of problems. The most important are fixes to pt-table-checksum, pt-iostats,
and pt-kill. Apart from pt-diskstats, pt-stalk, and pt-collect (which was removed), no other tools were changed in
backwards-incompatible ways, so it is worth reviewing the full changelog for this release and upgrading if you use
any tools which had bug fixes.
Thank you to the many people who reported bugs and submitted patches.
Download the latest release of Percona Toolkit 2.0 from http://www.percona.com/software/percona-toolkit/ or the
Percona Software Repositories (http://www.percona.com/docs/wiki/repositories:start).

Changelog

• Completely redesigned pt-diskstats
• Completely redesigned pt-stalk
• Removed pt-collect and put its functionality in pt-stalk
• Fixed bug 871438: Bash tools are insecure
• Fixed bug 897758: Failed to prepare TableSyncChunk plugin: Use of uninitialized value $args{“chunk_range”}
in lc at pt-table-sync line 3055
• Fixed bug 919819: pt-kill –execute-command creates zombies
• Fixed bug 925778: pt-ioprofile doesn’t run without a file
• Fixed bug 925477: pt-ioprofile docs refer to pt-iostats
• Fixed bug 857091: pt-sift downloads http://percona.com/get/pt-pmp, which does not work
• Fixed bug 857104: pt-sift tries to invoke mext, should be pt-mext
• Fixed bug 872699: pt-diskstats: rd_avkb & wr_avkb derived incorrectly
• Fixed bug 897029: pt-diskstats computes wrong values for md0
• Fixed bug 882918: pt-stalk spams warning if oprofile isn’t installed
• Fixed bug 884504: pt-stalk doesn’t check pt-collect
• Fixed bug 897483: pt-online-schema-change “uninitialized value” due to update-foreign-keys-method



• Fixed bug 925007: pt-online-schema-change Use of uninitialized value $tables{“old_table”} in concatenation
(.) or string at line 4330
• Fixed bug 915598: pt-config-diff ignores –ask-pass option
• Fixed bug 919352: pt-table-checksum changes binlog_format even if already set to statement
• Fixed bug 921700: pt-table-checksum doesn’t add –where to chunk size test on replicas
• Fixed bug 921802: pt-table-checksum does not recognize –recursion-method=processlist
• Fixed bug 925855: pt-table-checksum index check is case-sensitive
• Fixed bug 821709: pt-show-grants –revoke and –separate don’t work together
• Fixed bug 918247: Some tools use VALUE instead of VALUES

4.5.4 v2.0.2 released 2012-01-05

Percona Toolkit 2.0.2 fixes one critical bug: pt-table-sync –replicate did not work with character values, causing an
“Unknown column” error. If using Percona Toolkit 2.0.1, you should upgrade to 2.0.2.

Changelog

• Fixed bug 911996: pt-table-sync –replicate causes “Unknown column” error

4.5.5 v2.0.1 released 2011-12-30

The Percona Toolkit development team is proud to announce a new major version: 2.0. Beginning with Percona
Toolkit 2.0, we are overhauling, redesigning, and improving the major tools. 2.0 tools are therefore not backwards
compatible with 1.0 tools, which we still support but will not continue to develop.
New in Percona Toolkit 2.0.1 is a completely redesigned pt-table-checksum. The original pt-table-checksum 1.0 was
rather complex, but it worked well for many years. By contrast, the new pt-table-checksum 2.0 is much simpler but
also much more efficient and reliable. We spent months rethinking, redesigning, and testing every aspect of the tool.
The three most significant changes: pt-table-checksum 2.0 does only –replicate, it has only one chunking algorithm,
and its memory usage is stable even with hundreds of thousands of tables and trillions of rows. The tool is now
dedicated to verifying MySQL replication integrity, nothing else, which it does extremely well.
In Percona Toolkit 2.0.1 we also fixed various small bugs and forked ioprofile and align (as pt-ioprofile and pt-align)
from Aspersa.
If you still need functionalities in the original pt-table-checksum, the latest Percona Toolkit 1.0 release remains avail-
able for download. Otherwise, all new development in Percona Toolkit will happen in 2.0.

Changelog

• Completely redesigned pt-table-checksum
• Fixed bug 856065: pt-trend does not work
• Fixed bug 887688: Prepared statements crash pt-query-digest

4.5. Release Notes 303


• Fixed bug 888286: align not part of percona-toolkit
• Fixed bug 897961: ptc 2.0 replicate-check error does not include hostname
• Fixed bug 898318: ptc 2.0 –resume with –tables does not always work
• Fixed bug 903513: MKDEBUG should be PTDEBUG
• Fixed bug 908256: Percona Toolkit should include pt-ioprofile
• Fixed bug 821717: pt-tcp-model –type=requests crashes
• Fixed bug 844038: pt-online-schema-change documentation example w/drop-tmp-table does not work
• Fixed bug 864205: Remove the query to reset @crc from pt-table-checksum
• Fixed bug 898663: Typo in pt-log-player documentation

4.5.6 v1.0.1 released 2011-09-01

Percona Toolkit 1.0.1 has been released. In July, Baron announced planned changes to Maatkit and Aspersa devel-
opment;[1] Percona Toolkit is the result. In brief, Percona Toolkit is the combined fork of Maatkit and Aspersa, so
although the toolkit is new, the programs are not. That means Percona Toolkit 1.0.1 is mature, stable, and production-
ready. In fact, it’s even a little more stable because we fixed a few bugs in this release.
Percona Toolkit packages can be downloaded from http://www.percona.com/downloads/percona-toolkit/ or the Per-
cona Software Repositories (http://www.percona.com/docs/wiki/repositories:start).
Although Maatkit and Aspersa development use Google Code, Percona Toolkit uses Launchpad:
https://launchpad.net/percona-toolkit
[1] http://www.xaprb.com/blog/2011/07/06/planned-change-in-maatkit-aspersa-development/

Changelog

• Fixed bug 819421: MasterSlave::is_replication_thread() doesn’t match all
• Fixed bug 821673: pt-table-checksum doesn’t include –where in min max queries
• Fixed bug 821688: pt-table-checksum SELECT MIN MAX for char chunking is wrong
• Fixed bug 838211: pt-collect: line 24: [: : integer expression expected
• Fixed bug 838248: pt-collect creates a “5.1” file

4.5.7 v0.9.5 released 2011-08-04

Percona Toolkit 0.9.5 represents the completed transition from Maatkit and Aspersa. There are no bug fixes or new
features, but some features have been removed (like –save-results from pt-query-digest). This release is the starting
point for the 1.0 series where new development will happen, and no more changes will be made to the 0.9 series.

Changelog

• Forked, combined, and rebranded Maatkit and Aspersa as Percona Toolkit.


INDEX

Symbols pt-table-usage command line option, 248
–aggregate pt-upgrade command line option, 263
pt-ioprofile command line option, 92 pt-variable-advisor command line option, 277
–algorithms pt-visual-explain command line option, 288
pt-table-sync command line option, 233 –attribute-aliases
–all-structs pt-query-digest command line option, 153
pt-duplicate-key-checker command line option, 47 –attribute-value-limit
–alter pt-query-digest command line option, 153
pt-online-schema-change command line option, 127 –autoinc
–alter-foreign-keys-method pt-find command line option, 58
pt-online-schema-change command line option, 127 –aux-dsn
–always pt-query-digest command line option, 153
pt-slave-restart command line option, 195 –avgrowlen
–analyze pt-find command line option, 58
pt-archiver command line option, 9 –base-dir
–any-busy-time pt-log-player command line option, 108
pt-kill command line option, 103 pt-upgrade command line option, 263
–apdex-threshold –base-file-name
pt-query-digest command line option, 153 pt-log-player command line option, 108
–ascend-first –bidirectional
pt-archiver command line option, 9 pt-table-sync command line option, 234
–ask-pass –buffer
pt-archiver command line option, 10 pt-archiver command line option, 10
pt-config-diff command line option, 26 –buffer-in-mysql
pt-deadlock-logger command line option, 32 pt-table-sync command line option, 234
pt-duplicate-key-checker command line option, 47 –bulk-delete
pt-find command line option, 56 pt-archiver command line option, 10
pt-fk-error-logger command line option, 69 –bulk-insert
pt-heartbeat command line option, 75 pt-archiver command line option, 10
pt-index-usage command line option, 84 –busy-time
pt-kill command line option, 97 pt-kill command line option, 100
pt-log-player command line option, 108 –case-insensitive
pt-online-schema-change command line option, 128 pt-find command line option, 56
pt-query-advisor command line option, 142 –cell
pt-query-digest command line option, 153 pt-ioprofile command line option, 92
pt-show-grants command line option, 175 –charset
pt-slave-delay command line option, 184 pt-archiver command line option, 10
pt-slave-find command line option, 189 pt-config-diff command line option, 26
pt-slave-restart command line option, 195 pt-deadlock-logger command line option, 32
pt-table-checksum command line option, 218 pt-duplicate-key-checker command line option, 47
pt-table-sync command line option, 233 pt-find command line option, 56
pt-fk-error-logger command line option, 69

305


pt-heartbeat command line option, 75 –collation
pt-index-usage command line option, 84 pt-find command line option, 58
pt-kill command line option, 97 –collect
pt-log-player command line option, 108 pt-stalk command line option, 203
pt-online-schema-change command line option, 129 –collect-gdb
pt-query-advisor command line option, 142 pt-stalk command line option, 203
pt-query-digest command line option, 154 –collect-oprofile
pt-show-grants command line option, 176 pt-stalk command line option, 204
pt-slave-delay command line option, 184 –collect-strace
pt-slave-find command line option, 189 pt-stalk command line option, 204
pt-slave-restart command line option, 195 –collect-tcpdump
pt-table-sync command line option, 234 pt-stalk command line option, 204
pt-table-usage command line option, 248 –column-name
pt-upgrade command line option, 263 pt-find command line option, 58
pt-variable-advisor command line option, 277 –column-type
pt-visual-explain command line option, 288 pt-find command line option, 58
–check –columns
–check-attributes-limit pt-deadlock-logger command line option, 33
pt-query-digest command line option, 154 pt-table-checksum command line option, 219
–check-interval pt-table-sync command line option, 235
pt-archiver command line option, 11 –columns-regex
pt-online-schema-change command line option, 129 pt-diskstats command line option, 43
pt-table-checksum command line option, 218 –comment
–check-slave-lag pt-find command line option, 58
pt-archiver command line option, 11 –commit-each
pt-online-schema-change command line option, 129 pt-archiver command line option, 11
pt-table-checksum command line option, 218 –compare
–checksum pt-upgrade command line option, 263
pt-find command line option, 58 –compare-results-method
–chunk-column pt-upgrade command line option, 264
pt-table-sync command line option, 235 –config
–chunk-index pt-archiver command line option, 11
pt-online-schema-change command line option, 129 pt-config-diff command line option, 26
pt-table-checksum command line option, 218 pt-deadlock-logger command line option, 33
pt-table-sync command line option, 235 pt-diskstats command line option, 43
–chunk-size pt-duplicate-key-checker command line option, 47
pt-online-schema-change command line option, 129 pt-fifo-split command line option, 52
pt-table-checksum command line option, 218 pt-find command line option, 56
pt-table-sync command line option, 235 pt-fingerprint command line option, 66
–chunk-size-limit pt-fk-error-logger command line option, 69
pt-online-schema-change command line option, 129 pt-heartbeat command line option, 75
pt-table-checksum command line option, 219 pt-index-usage command line option, 84
–chunk-time pt-kill command line option, 97
pt-online-schema-change command line option, 130 pt-log-player command line option, 108
pt-table-checksum command line option, 219 pt-mysql-summary command line option, 124
–clear-deadlocks pt-online-schema-change command line option, 130
pt-deadlock-logger command line option, 32 pt-query-advisor command line option, 142
–clear-warnings-table pt-query-digest command line option, 154
pt-upgrade command line option, 263 pt-show-grants command line option, 176
–clustered-pk pt-slave-delay command line option, 184
pt-visual-explain command line option, 288 pt-slave-find command line option, 189
–cmin pt-slave-restart command line option, 196
pt-find command line option, 58 pt-stalk command line option, 204

306 Index


pt-summary command line option, 212 pt-kill command line option, 97
pt-table-checksum command line option, 219 pt-query-advisor command line option, 142
pt-table-sync command line option, 235 pt-query-digest command line option, 154
pt-table-usage command line option, 248 pt-slave-delay command line option, 184
pt-tcp-model command line option, 255 pt-slave-restart command line option, 196
pt-trend command line option, 259 pt-stalk command line option, 204
pt-upgrade command line option, 264 pt-table-usage command line option, 249
pt-variable-advisor command line option, 278 pt-upgrade command line option, 264
pt-visual-explain command line option, 288 pt-variable-advisor command line option, 278
–conflict-column –database
pt-table-sync command line option, 235 pt-heartbeat command line option, 76
–conflict-comparison pt-index-usage command line option, 84
pt-table-sync command line option, 235 pt-query-advisor command line option, 142
–conflict-error pt-show-grants command line option, 176
pt-table-sync command line option, 236 pt-slave-find command line option, 190
–conflict-threshold pt-slave-restart command line option, 196
pt-table-sync command line option, 236 pt-table-usage command line option, 249
–conflict-value pt-visual-explain command line option, 288
pt-table-sync command line option, 236 –databases
–connect pt-duplicate-key-checker command line option, 48
pt-visual-explain command line option, 288 pt-index-usage command line option, 84
–connection-id pt-mysql-summary command line option, 124
pt-find command line option, 58 pt-table-checksum command line option, 219
–constant-data-value pt-table-sync command line option, 236
pt-table-usage command line option, 248 –databases-regex
–continue-on-error pt-index-usage command line option, 84
pt-upgrade command line option, 264 pt-table-checksum command line option, 219
–convert-to-select –datafree
pt-upgrade command line option, 264 pt-find command line option, 59
–create-dest-table –datasize
pt-deadlock-logger command line option, 33 pt-find command line option, 59
–create-review-history-table –day-start
pt-query-digest command line option, 154 pt-find command line option, 56
–create-review-table –dbi-driver
pt-query-digest command line option, 154 pt-heartbeat command line option, 76
–create-save-results-database –dblike
pt-index-usage command line option, 84 pt-find command line option, 59
–create-table –dbregex
pt-heartbeat command line option, 75 pt-find command line option, 59
–create-table-definitions –defaults-file
pt-table-usage command line option, 248 pt-config-diff command line option, 26
–createopts pt-deadlock-logger command line option, 33
pt-find command line option, 59 pt-duplicate-key-checker command line option, 48
–critical-load pt-find command line option, 56
pt-online-schema-change command line option, 130 pt-fk-error-logger command line option, 69
–ctime pt-heartbeat command line option, 76
pt-find command line option, 59 pt-index-usage command line option, 84
–cycles pt-kill command line option, 97
pt-stalk command line option, 204 pt-log-player command line option, 108
–daemonize pt-online-schema-change command line option, 130
pt-config-diff command line option, 26 pt-query-advisor command line option, 142
pt-deadlock-logger command line option, 33 pt-query-digest command line option, 154
pt-fk-error-logger command line option, 69 pt-show-grants command line option, 176
pt-heartbeat command line option, 76 pt-slave-delay command line option, 184

Index 307


pt-slave-find command line option, 190 –exec-plus
pt-slave-restart command line option, 196 pt-find command line option, 61
pt-table-checksum command line option, 219 –execute
pt-table-sync command line option, 236 pt-online-schema-change command line option, 130
pt-table-usage command line option, 249 pt-query-digest command line option, 155
pt-variable-advisor command line option, 278 pt-table-sync command line option, 237
pt-visual-explain command line option, 288 –execute-command
–delay pt-kill command line option, 103
pt-slave-delay command line option, 184 –execute-throttle
–delayed-insert pt-query-digest command line option, 155
pt-archiver command line option, 12 –expected-range
–dest pt-query-digest command line option, 155
pt-archiver command line option, 12 –explain
pt-fk-error-logger command line option, 69 pt-table-checksum command line option, 220
pt-stalk command line option, 204 –explain-extended
–devices-regex pt-table-usage command line option, 249
pt-diskstats command line option, 43 –explain-hosts
–disk-bytes-free pt-table-sync command line option, 237
pt-stalk command line option, 204 pt-upgrade command line option, 264
–disk-pct-free –fifo
pt-stalk command line option, 204 pt-fifo-split command line option, 52
–drop –file
pt-index-usage command line option, 84 pt-archiver command line option, 12
pt-show-grants command line option, 176 pt-heartbeat command line option, 76
–dry-run –filter
pt-archiver command line option, 12 pt-kill command line option, 97
pt-log-player command line option, 108 pt-log-player command line option, 108
pt-online-schema-change command line option, 130 pt-query-digest command line option, 156
pt-table-sync command line option, 237 pt-table-usage command line option, 249
–each-busy-time pt-upgrade command line option, 265
pt-kill command line option, 103 –fingerprints
–embedded-attributes pt-query-digest command line option, 157
pt-query-digest command line option, 154 pt-upgrade command line option, 265
–empty –float-precision
–empty-save-results-tables pt-table-sync command line option, 237
pt-index-usage command line option, 85 pt-upgrade command line option, 265
–engine –flush
pt-find command line option, 59 pt-show-grants command line option, 176
–engines –for-update
pt-duplicate-key-checker command line option, 48 pt-archiver command line option, 12
pt-table-checksum command line option, 220 –force
pt-table-sync command line option, 237 pt-fifo-split command line option, 53
–error-length –format
pt-slave-restart command line option, 196 pt-visual-explain command line option, 288
–error-numbers –frames
pt-slave-restart command line option, 196 pt-heartbeat command line option, 76
–error-text –function
pt-slave-restart command line option, 196 pt-find command line option, 60
–exec pt-stalk command line option, 204
–exec-dsn pt-table-sync command line option, 237
pt-find command line option, 61 –group-by

308 Index


pt-diskstats command line option, 43 pt-log-player command line option, 110
pt-ioprofile command line option, 92 pt-online-schema-change command line option, 131
pt-kill command line option, 98 pt-query-advisor command line option, 143
pt-query-advisor command line option, 143 pt-query-digest command line option, 158
pt-query-digest command line option, 158 pt-show-grants command line option, 176
–header pt-slave-delay command line option, 184
pt-archiver command line option, 13 pt-slave-find command line option, 190
–headers pt-slave-restart command line option, 196
pt-diskstats command line option, 43 pt-table-checksum command line option, 220
–help pt-table-sync command line option, 238
pt-archiver command line option, 13 pt-table-usage command line option, 249
pt-config-diff command line option, 26 pt-upgrade command line option, 265
pt-deadlock-logger command line option, 34 pt-variable-advisor command line option, 278
pt-diskstats command line option, 44 pt-visual-explain command line option, 288
pt-duplicate-key-checker command line option, 48 –id-attribute
pt-fifo-split command line option, 53 pt-table-usage command line option, 249
pt-find command line option, 56 –idle-time
pt-fingerprint command line option, 66 pt-kill command line option, 100
pt-fk-error-logger command line option, 69 –ignore
pt-index-usage command line option, 85 pt-show-grants command line option, 176
pt-ioprofile command line option, 93 –ignore-attributes
pt-kill command line option, 98 pt-query-digest command line option, 158
pt-log-player command line option, 110 –ignore-columns
pt-mysql-summary command line option, 124 pt-table-checksum command line option, 220
pt-online-schema-change command line option, 131 pt-table-sync command line option, 238
pt-query-advisor command line option, 143 –ignore-command
pt-query-digest command line option, 158 pt-kill command line option, 100
pt-show-grants command line option, 176 –ignore-databases
pt-slave-delay command line option, 184 pt-duplicate-key-checker command line option, 48
pt-slave-find command line option, 190 pt-index-usage command line option, 85
pt-slave-restart command line option, 196 pt-table-checksum command line option, 221
pt-stalk command line option, 205 pt-table-sync command line option, 238
pt-summary command line option, 212 –ignore-databases-regex
pt-table-checksum command line option, 220 pt-index-usage command line option, 85
pt-table-sync command line option, 237 pt-table-checksum command line option, 221
pt-table-usage command line option, 249 –ignore-db
pt-tcp-model command line option, 255 pt-kill command line option, 100
pt-trend command line option, 259 –ignore-engines
pt-upgrade command line option, 265 pt-duplicate-key-checker command line option, 48
pt-variable-advisor command line option, 278 pt-table-checksum command line option, 221
pt-visual-explain command line option, 288 pt-table-sync command line option, 238
–high-priority-select –ignore-host
–host –ignore-info
pt-config-diff command line option, 26 –ignore-order
pt-deadlock-logger command line option, 34 pt-duplicate-key-checker command line option, 48
pt-duplicate-key-checker command line option, 48 –ignore-rules
pt-find command line option, 57 pt-query-advisor command line option, 143
pt-fk-error-logger command line option, 70 pt-variable-advisor command line option, 278
pt-heartbeat command line option, 77 –ignore-state
pt-index-usage command line option, 85 pt-kill command line option, 101
pt-kill command line option, 98 –ignore-tables

Index 309


pt-duplicate-key-checker command line option, 48 pt-table-checksum command line option, 221
pt-index-usage command line option, 85 –log
pt-table-checksum command line option, 221 pt-deadlock-logger command line option, 34
pt-table-sync command line option, 238 pt-fk-error-logger command line option, 70
–ignore-tables-regex pt-heartbeat command line option, 77
pt-index-usage command line option, 85 pt-kill command line option, 98
pt-table-checksum command line option, 221 pt-query-digest command line option, 159
–ignore-user pt-slave-delay command line option, 185
pt-kill command line option, 101 pt-slave-restart command line option, 197
–ignore-variables pt-stalk command line option, 205
pt-config-diff command line option, 26 pt-table-usage command line option, 249
–indexsize pt-upgrade command line option, 266
pt-find command line option, 60 –low-priority-delete
–inherit-attributes pt-archiver command line option, 13
pt-query-digest command line option, 158 –low-priority-insert
–interval pt-archiver command line option, 13
pt-deadlock-logger command line option, 34 –master-server-id
pt-diskstats command line option, 44 pt-heartbeat command line option, 77
pt-fk-error-logger command line option, 70 –match
pt-heartbeat command line option, 77 pt-stalk command line option, 206
pt-kill command line option, 98 –match-all
pt-query-digest command line option, 159 pt-kill command line option, 101
pt-slave-delay command line option, 185 –match-command
pt-stalk command line option, 205 pt-kill command line option, 101
–iterations –match-db
pt-diskstats command line option, 44 pt-kill command line option, 101
pt-log-player command line option, 110 –match-embedded-numbers
pt-query-digest command line option, 159 pt-fingerprint command line option, 66
pt-stalk command line option, 205 –match-host
pt-upgrade command line option, 266 pt-kill command line option, 102
–key-types –match-info
pt-duplicate-key-checker command line option, 48 pt-kill command line option, 102
–kill –match-md5-checksums
pt-kill command line option, 103 pt-fingerprint command line option, 66
–kill-query –match-state
pt-kill command line option, 104 pt-kill command line option, 102
–kmin –match-user
pt-find command line option, 60 pt-kill command line option, 102
–ktime –max-different-rows
pt-find command line option, 60 pt-upgrade command line option, 266
–limit –max-lag
pt-query-digest command line option, 159 pt-online-schema-change command line option, 131
pt-upgrade command line option, 266 pt-table-checksum command line option, 221
–lines –max-load
pt-fifo-split command line option, 53 pt-online-schema-change command line option, 131
–local pt-table-checksum command line option, 221
pt-archiver command line option, 13 –max-sessions
–lock pt-log-player command line option, 110
pt-table-sync command line option, 238 –max-sleep
–lock-and-rename pt-slave-restart command line option, 197
pt-table-sync command line option, 239 –min-sleep
–lock-wait-timeout pt-slave-restart command line option, 197
pt-online-schema-change command line option, 131 –mirror

310 Index


pt-query-digest command line option, 159 pt-visual-explain command line option, 288
–mmin –pid
pt-find command line option, 60 pt-archiver command line option, 14
–monitor pt-config-diff command line option, 26
pt-heartbeat command line option, 77 pt-deadlock-logger command line option, 34
pt-slave-restart command line option, 197 pt-duplicate-key-checker command line option, 48
–mtime pt-fifo-split command line option, 53
pt-find command line option, 60 pt-find command line option, 57
–no-ascend pt-fk-error-logger command line option, 70
pt-archiver command line option, 14 pt-heartbeat command line option, 78
–no-delete pt-kill command line option, 98
pt-archiver command line option, 14 pt-log-player command line option, 110
–notify-by-email pt-online-schema-change command line option, 131
pt-stalk command line option, 206 pt-query-advisor command line option, 143
–numeric-ip pt-query-digest command line option, 160
pt-deadlock-logger command line option, 34 pt-show-grants command line option, 177
–offset pt-slave-delay command line option, 185
pt-fifo-split command line option, 53 pt-slave-find command line option, 190
–only pt-slave-restart command line option, 197
pt-show-grants command line option, 176 pt-stalk command line option, 206
–only-select pt-table-checksum command line option, 222
pt-log-player command line option, 110 pt-table-sync command line option, 239
–optimize pt-table-usage command line option, 249
pt-archiver command line option, 14 pt-trend command line option, 259
–or pt-upgrade command line option, 266
pt-find command line option, 57 pt-variable-advisor command line option, 278
–order-by pt-visual-explain command line option, 289
pt-query-digest command line option, 159 –pipeline-profile
pt-upgrade command line option, 266 pt-query-digest command line option, 160
–outliers –play
pt-query-digest command line option, 160 pt-log-player command line option, 110
–password –plugin
pt-config-diff command line option, 26 –port
pt-deadlock-logger command line option, 34 pt-archiver command line option, 15
pt-duplicate-key-checker command line option, 48 pt-config-diff command line option, 26
pt-find command line option, 57 pt-deadlock-logger command line option, 34
pt-fk-error-logger command line option, 70 pt-duplicate-key-checker command line option, 48
pt-heartbeat command line option, 78 pt-find command line option, 57
pt-index-usage command line option, 85 pt-fk-error-logger command line option, 70
pt-kill command line option, 98 pt-heartbeat command line option, 78
pt-log-player command line option, 110 pt-index-usage command line option, 85
pt-online-schema-change command line option, 131 pt-kill command line option, 99
pt-query-advisor command line option, 143 pt-log-player command line option, 110
pt-query-digest command line option, 160 pt-online-schema-change command line option, 132
pt-show-grants command line option, 176 pt-query-advisor command line option, 143
pt-slave-delay command line option, 185 pt-query-digest command line option, 160
pt-slave-find command line option, 190 pt-show-grants command line option, 177
pt-slave-restart command line option, 197 pt-slave-delay command line option, 185
pt-table-checksum command line option, 222 pt-slave-find command line option, 190
pt-table-sync command line option, 239 pt-slave-restart command line option, 197
pt-table-usage command line option, 249 pt-table-checksum command line option, 222
pt-upgrade command line option, 266 pt-table-sync command line option, 239
pt-variable-advisor command line option, 278 pt-table-usage command line option, 249

Index 311


pt-upgrade command line option, 266 –quiet
pt-variable-advisor command line option, 278 pt-archiver command line option, 15
pt-visual-explain command line option, 289 pt-index-usage command line option, 85
–prefix pt-log-player command line option, 110
pt-stalk command line option, 206 pt-online-schema-change command line option, 132
–primary-key-only pt-slave-delay command line option, 185
pt-archiver command line option, 15 pt-slave-restart command line option, 197
–print pt-table-checksum command line option, 222
pt-deadlock-logger command line option, 34 pt-trend command line option, 259
pt-find command line option, 62 –read-samples
pt-fk-error-logger command line option, 70 pt-mysql-summary command line option, 124
pt-kill command line option, 104 pt-summary command line option, 212
pt-log-player command line option, 110 –read-timeout
pt-online-schema-change command line option, 132 pt-query-digest command line option, 161
pt-query-digest command line option, 160 pt-table-usage command line option, 250
pt-table-sync command line option, 239 –recurse
–print-all pt-heartbeat command line option, 78
pt-query-advisor command line option, 143 pt-online-schema-change command line option, 132
–print-iterations pt-slave-find command line option, 190
pt-query-digest command line option, 160 pt-slave-restart command line option, 197
–print-master-server-id pt-table-checksum command line option, 222
pt-heartbeat command line option, 78 –recursion-method
–printf pt-heartbeat command line option, 78
pt-find command line option, 62 pt-online-schema-change command line option, 132
–procedure pt-slave-find command line option, 190
pt-find command line option, 60 pt-slave-restart command line option, 197
–processlist pt-table-checksum command line option, 222
pt-query-digest command line option, 161 pt-table-sync command line option, 239
–profile-pid –replace
pt-ioprofile command line option, 93 pt-archiver command line option, 15
–profile-process pt-heartbeat command line option, 78
pt-ioprofile command line option, 93 pt-table-sync command line option, 240
–progress –replicate
pt-archiver command line option, 15 pt-table-checksum command line option, 223
pt-index-usage command line option, 85 pt-table-sync command line option, 240
pt-online-schema-change command line option, 132 –replicate-check-only
pt-table-checksum command line option, 222 –replicate-database
pt-table-usage command line option, 250 pt-table-checksum command line option, 224
pt-tcp-model command line option, 255 –replication-threads
pt-trend command line option, 259 pt-kill command line option, 102
–purge –report-all
pt-archiver command line option, 15 pt-query-digest command line option, 161
–quantile –report-format
pt-tcp-model command line option, 255 pt-index-usage command line option, 86
–query pt-query-advisor command line option, 143
pt-fingerprint command line option, 66 pt-query-digest command line option, 161
pt-query-advisor command line option, 143 pt-slave-find command line option, 190
pt-table-usage command line option, 250 –report-histogram
pt-upgrade command line option, 266 pt-query-digest command line option, 162
–query-count –report-width
pt-kill command line option, 103 pt-config-diff command line option, 27
–quick-delete –reports
pt-archiver command line option, 15 pt-upgrade command line option, 266

312 Index


–resume –separate
pt-table-checksum command line option, 224 pt-show-grants command line option, 177
–retention-time –separator
pt-stalk command line option, 206 pt-table-checksum command line option, 224
–retries –server-id
pt-archiver command line option, 15 pt-find command line option, 60
pt-online-schema-change command line option, 133 –session-files
pt-table-checksum command line option, 224 pt-log-player command line option, 111
–review –set-vars
pt-query-advisor command line option, 144 pt-archiver command line option, 16
pt-query-digest command line option, 162 pt-config-diff command line option, 27
–review-history pt-deadlock-logger command line option, 34
pt-query-digest command line option, 163 pt-duplicate-key-checker command line option, 49
–revoke pt-find command line option, 57
pt-show-grants command line option, 177 pt-fk-error-logger command line option, 70
–rowformat pt-heartbeat command line option, 79
pt-find command line option, 60 pt-index-usage command line option, 88
–rows pt-kill command line option, 99
pt-find command line option, 60 pt-log-player command line option, 111
–run-time pt-online-schema-change command line option, 133
pt-archiver command line option, 16 pt-query-advisor command line option, 144
pt-fk-error-logger command line option, 70 pt-show-grants command line option, 177
pt-heartbeat command line option, 78 pt-slave-delay command line option, 185
pt-ioprofile command line option, 93 pt-slave-find command line option, 191
pt-kill command line option, 99 pt-slave-restart command line option, 198
pt-slave-delay command line option, 185 pt-table-sync command line option, 240
pt-slave-restart command line option, 198 pt-table-usage command line option, 250
pt-stalk command line option, 206 pt-upgrade command line option, 267
pt-table-usage command line option, 250 pt-variable-advisor command line option, 278
pt-tcp-model command line option, 255 pt-visual-explain command line option, 289
pt-upgrade command line option, 266 –share-lock
–run-time-mode pt-archiver command line option, 16
pt-query-digest command line option, 165 –shorten
–sample pt-query-digest command line option, 167
pt-query-advisor command line option, 144 pt-upgrade command line option, 267
pt-query-digest command line option, 166 –show-all
–sample-time pt-query-digest command line option, 167
pt-diskstats command line option, 44 –show-inactive
–save-results-database pt-diskstats command line option, 44
pt-index-usage command line option, 86 –show-timestamps
–save-samples pt-diskstats command line option, 44
pt-diskstats command line option, 44 –since
pt-ioprofile command line option, 93 pt-query-digest command line option, 167
pt-mysql-summary command line option, 124 –skew
pt-summary command line option, 212 pt-heartbeat command line option, 79
–select –skip-count
pt-query-digest command line option, 166 pt-slave-restart command line option, 198
–sentinel –skip-foreign-key-checks
pt-heartbeat command line option, 79 –sleep
pt-kill command line option, 99 pt-archiver command line option, 16
pt-slave-restart command line option, 198 pt-mysql-summary command line option, 124

Index 313


pt-slave-restart command line option, 198 –summarize-processes
pt-stalk command line option, 206 pt-summary command line option, 212
pt-summary command line option, 212 –sync-to-master
–sleep-coef pt-table-sync command line option, 240
pt-archiver command line option, 16 –tab
–socket pt-deadlock-logger command line option, 35
pt-archiver command line option, 17 –table
pt-config-diff command line option, 27 pt-heartbeat command line option, 79
pt-deadlock-logger command line option, 34 –table-access
pt-duplicate-key-checker command line option, 49 pt-query-digest command line option, 168
pt-find command line option, 57 –tables
pt-fk-error-logger command line option, 70 pt-duplicate-key-checker command line option, 49
pt-heartbeat command line option, 79 pt-index-usage command line option, 89
pt-index-usage command line option, 88 pt-table-checksum command line option, 224
pt-kill command line option, 99 pt-table-sync command line option, 240
pt-log-player command line option, 111 –tables-regex
pt-online-schema-change command line option, 133 pt-index-usage command line option, 89
pt-query-advisor command line option, 144 pt-table-checksum command line option, 224
pt-query-digest command line option, 167 –tablesize
pt-show-grants command line option, 177 pt-find command line option, 60
pt-slave-delay command line option, 185 –tbllike
pt-slave-find command line option, 191 pt-find command line option, 61
pt-slave-restart command line option, 198 –tblregex
pt-table-sync command line option, 240 –tblversion
pt-table-usage command line option, 250 pt-find command line option, 61
pt-upgrade command line option, 267 –tcpdump-errors
pt-variable-advisor command line option, 278 pt-query-digest command line option, 168
pt-visual-explain command line option, 289 –temp-database
–source pt-upgrade command line option, 267
pt-archiver command line option, 17 –temp-table
–source-of-variables pt-upgrade command line option, 267
pt-variable-advisor command line option, 278 –test-matching
–split pt-kill command line option, 102
pt-log-player command line option, 111 –threads
–split-random pt-log-player command line option, 111
pt-log-player command line option, 111 –threshold
–stalk pt-stalk command line option, 206
pt-stalk command line option, 206 –timeline
–start-end pt-query-digest command line option, 168
pt-tcp-model command line option, 255 –timeout-ok
–statistics pt-table-sync command line option, 240
pt-archiver command line option, 17 –trigger
pt-fifo-split command line option, 53 pt-find command line option, 61
pt-query-digest command line option, 168 –trigger-table
–stop pt-find command line option, 61
pt-archiver command line option, 18 –trim
pt-heartbeat command line option, 79 pt-table-checksum command line option, 224
pt-kill command line option, 99 pt-table-sync command line option, 241
pt-slave-restart command line option, 198 –txn-size
–summarize-mounts pt-archiver command line option, 18
pt-summary command line option, 212 –type
–summarize-network pt-log-player command line option, 111
pt-summary command line option, 212 pt-query-advisor command line option, 144

314 Index


pt-query-digest command line option, 169 pt-fifo-split command line option, 53
pt-tcp-model command line option, 256 pt-find command line option, 57
–until pt-fingerprint command line option, 66
pt-query-digest command line option, 171 pt-fk-error-logger command line option, 70
–until-master pt-heartbeat command line option, 80
pt-slave-restart command line option, 199 pt-index-usage command line option, 89
–until-relay pt-ioprofile command line option, 93
pt-slave-restart command line option, 199 pt-kill command line option, 99
–update pt-log-player command line option, 112
pt-heartbeat command line option, 79 pt-mysql-summary command line option, 124
–use-master pt-online-schema-change command line option, 133
pt-slave-delay command line option, 185 pt-query-advisor command line option, 144
–user pt-query-digest command line option, 171
pt-archiver command line option, 18 pt-show-grants command line option, 177
pt-config-diff command line option, 27 pt-slave-delay command line option, 186
pt-deadlock-logger command line option, 35 pt-slave-find command line option, 191
pt-duplicate-key-checker command line option, 49 pt-slave-restart command line option, 199
pt-find command line option, 57 pt-stalk command line option, 206
pt-fk-error-logger command line option, 70 pt-summary command line option, 212
pt-heartbeat command line option, 79 pt-table-checksum command line option, 225
pt-kill command line option, 99 pt-table-usage command line option, 250
pt-log-player command line option, 112 pt-tcp-model command line option, 256
pt-online-schema-change command line option, 133 pt-trend command line option, 259
pt-query-advisor command line option, 144 pt-upgrade command line option, 267
pt-query-digest command line option, 171 pt-variable-advisor command line option, 279
pt-show-grants command line option, 177 pt-visual-explain command line option, 289
pt-slave-delay command line option, 185 –victims
pt-slave-find command line option, 191 pt-kill command line option, 99
pt-slave-restart command line option, 199 –view
pt-table-sync command line option, 241 –wait
pt-table-usage command line option, 250 pt-table-sync command line option, 241
pt-upgrade command line option, 267 –wait-after-kill
pt-variable-advisor command line option, 278 pt-kill command line option, 100
pt-visual-explain command line option, 289 –wait-before-kill
–variable pt-kill command line option, 100
pt-stalk command line option, 206 –watch-server
–variations pt-query-digest command line option, 171
pt-query-digest command line option, 171 pt-tcp-model command line option, 256
–verbose –where
pt-duplicate-key-checker command line option, 49 pt-archiver command line option, 18
pt-kill command line option, 103 pt-query-advisor command line option, 144
pt-log-player command line option, 112 pt-table-checksum command line option, 225
pt-query-advisor command line option, 144 pt-table-sync command line option, 242
pt-slave-restart command line option, 199 –why-quit
pt-table-sync command line option, 241 pt-archiver command line option, 19
pt-variable-advisor command line option, 279 –zero-query-times
–version pt-upgrade command line option, 267
pt-archiver command line option, 18 –[no]bin-log
pt-config-diff command line option, 27 pt-table-sync command line option, 234
pt-deadlock-logger command line option, 35 –[no]buffer-to-client
pt-diskstats command line option, 44 pt-table-sync command line option, 234
pt-duplicate-key-checker command line option, 49 –[no]bulk-delete-limit

Index 315


pt-archiver command line option, 10 –[no]replicate-check
–[no]check-charset pt-table-checksum command line option, 223
pt-archiver command line option, 11 –[no]report
–[no]check-columns pt-config-diff command line option, 26
pt-archiver command line option, 11 pt-index-usage command line option, 85
–[no]check-master pt-query-digest command line option, 161
pt-table-sync command line option, 234 –[no]results
–[no]check-privileges pt-log-player command line option, 111
pt-table-sync command line option, 234 –[no]safe-auto-increment
–[no]check-relay-log pt-archiver command line option, 16
pt-slave-restart command line option, 196 –[no]show-create-table
–[no]check-replication-filters pt-query-advisor command line option, 144
pt-online-schema-change command line option, 129 –[no]sql
pt-table-checksum command line option, 218 pt-duplicate-key-checker command line option, 49
–[no]check-slave –[no]strip-comments
pt-table-sync command line option, 234 pt-kill command line option, 99
–[no]check-triggers –[no]summary
pt-table-sync command line option, 235 pt-duplicate-key-checker command line option, 49
–[no]clear-warnings –[no]swap-tables
pt-upgrade command line option, 263 pt-online-schema-change command line option, 133
–[no]clustered –[no]timestamp
pt-duplicate-key-checker command line option, 47 pt-show-grants command line option, 177
–[no]collapse –[no]transaction
pt-deadlock-logger command line option, 32 pt-table-sync command line option, 241
–[no]continue –[no]unique-checks
pt-slave-delay command line option, 184 pt-table-sync command line option, 241
–[no]continue-on-error –[no]warnings
pt-query-advisor command line option, 142 pt-log-player command line option, 112
pt-query-digest command line option, 154 –[no]zero-admin
pt-table-usage command line option, 248 pt-query-digest command line option, 172
–[no]create-replicate-table –[no]zero-bool
pt-table-checksum command line option, 219 pt-query-digest command line option, 172
–[no]create-views –[no]zero-chunk
–[no]drop-old-table
pt-online-schema-change command line option, 130 P
–[no]empty-replicate-table pt-archiver command line option
pt-table-checksum command line option, 220 –analyze, 9
–[no]for-explain –ascend-first, 9
pt-query-digest command line option, 157 –ask-pass, 10
–[no]foreign-key-checks –buffer, 10
pt-table-sync command line option, 237 –bulk-delete, 10
–[no]header –bulk-insert, 10
pt-show-grants command line option, 176 –charset, 10
–[no]hex-blob –check-interval, 11
pt-table-sync command line option, 237 –check-slave-lag, 11
–[no]ignore-self –columns, 11
pt-kill command line option, 101 –commit-each, 11
–[no]index-hint –config, 11
pt-table-sync command line option, 238 –delayed-insert, 12
–[no]insert-heartbeat-row –dest, 12
pt-heartbeat command line option, 77 –dry-run, 12
–[no]quote –file, 12
pt-find command line option, 57 –for-update, 12

316 Index


–header, 13 –pid, 26
–help, 13 –port, 26
–high-priority-select, 13 –report-width, 27
–host, 13 –set-vars, 27
–ignore, 13 –socket, 27
–limit, 13 –user, 27
–local, 13 –version, 27
–low-priority-delete, 13 –[no]report, 26
–low-priority-insert, 13 pt-deadlock-logger command line option
–max-lag, 13 –ask-pass, 32
–no-ascend, 14 –charset, 32
–no-delete, 14 –clear-deadlocks, 32
–optimize, 14 –columns, 33
–password, 14 –config, 33
–pid, 14 –create-dest-table, 33
–plugin, 14 –daemonize, 33
–port, 15 –defaults-file, 33
–primary-key-only, 15 –dest, 33
–progress, 15 –help, 34
–purge, 15 –host, 34
–quick-delete, 15 –interval, 34
–quiet, 15 –log, 34
–replace, 15 –numeric-ip, 34
–retries, 15 –password, 34
–run-time, 16 –pid, 34
–sentinel, 16 –port, 34
–set-vars, 16 –print, 34
–share-lock, 16 –run-time, 34
–skip-foreign-key-checks, 16 –set-vars, 34
–sleep, 16 –socket, 34
–sleep-coef, 16 –tab, 35
–socket, 17 –user, 35
–source, 17 –version, 35
–statistics, 17 –[no]collapse, 32
–stop, 18 pt-diskstats command line option
–txn-size, 18 –columns-regex, 43
–user, 18 –config, 43
–version, 18 –devices-regex, 43
–where, 18 –group-by, 43
–why-quit, 19 –headers, 43
–[no]bulk-delete-limit, 10 –help, 44
–[no]check-charset, 11 –interval, 44
–[no]check-columns, 11 –iterations, 44
–[no]safe-auto-increment, 16 –sample-time, 44
pt-config-diff command line option –save-samples, 44
–ask-pass, 26 –show-inactive, 44
–charset, 26 –show-timestamps, 44
–config, 26 –version, 44
–daemonize, 26 pt-duplicate-key-checker command line option
–defaults-file, 26 –all-structs, 47
–help, 26 –ask-pass, 47
–host, 26 –charset, 47
–ignore-variables, 26 –config, 47
–password, 26 –databases, 48

Index 317


–defaults-file, 48 –engine, 59
–engines, 48 –exec, 61
–help, 48 –exec-dsn, 61
–host, 48 –exec-plus, 61
–ignore-databases, 48 –function, 60
–ignore-engines, 48 –help, 56
–ignore-order, 48 –host, 57
–ignore-tables, 48 –indexsize, 60
–key-types, 48 –kmin, 60
–password, 48 –ktime, 60
–pid, 48 –mmin, 60
–port, 48 –mtime, 60
–set-vars, 49 –or, 57
–socket, 49 –password, 57
–tables, 49 –pid, 57
–user, 49 –port, 57
–verbose, 49 –print, 62
–version, 49 –printf, 62
–[no]clustered, 47 –procedure, 60
–[no]sql, 49 –rowformat, 60
–[no]summary, 49 –rows, 60
pt-fifo-split command line option –server-id, 60
–config, 52 –set-vars, 57
–fifo, 52 –socket, 57
–force, 53 –tablesize, 60
–help, 53 –tbllike, 61
–lines, 53 –tblregex, 61
–offset, 53 –tblversion, 61
–pid, 53 –trigger, 61
–statistics, 53 –trigger-table, 61
–version, 53 –user, 57
pt-find command line option –version, 57
–ask-pass, 56 –view, 61
–autoinc, 58 –[no]quote, 57
–avgrowlen, 58 pt-fingerprint command line option
–case-insensitive, 56 –config, 66
–charset, 56 –help, 66
–checksum, 58 –match-embedded-numbers, 66
–cmin, 58 –match-md5-checksums, 66
–collation, 58 –query, 66
–column-name, 58 –version, 66
–column-type, 58 pt-fk-error-logger command line option
–comment, 58 –ask-pass, 69
–config, 56 –charset, 69
–connection-id, 58 –config, 69
–createopts, 59 –daemonize, 69
–ctime, 59 –defaults-file, 69
–datafree, 59 –dest, 69
–datasize, 59 –help, 69
–day-start, 56 –host, 70
–dblike, 59 –interval, 70
–dbregex, 59 –log, 70
–defaults-file, 56 –password, 70
–empty, 59 –pid, 70

318 Index


–port, 70 –help, 85
–print, 70 –host, 85
–run-time, 70 –ignore-databases, 85
–set-vars, 70 –ignore-databases-regex, 85
–socket, 70 –ignore-tables, 85
–user, 70 –ignore-tables-regex, 85
–version, 70 –password, 85
pt-heartbeat command line option –port, 85
–ask-pass, 75 –progress, 85
–charset, 75 –quiet, 85
–check, 75 –report-format, 86
–config, 75 –save-results-database, 86
–create-table, 75 –set-vars, 88
–daemonize, 76 –socket, 88
–database, 76 –tables, 89
–dbi-driver, 76 –tables-regex, 89
–defaults-file, 76 –user, 89
–file, 76 –version, 89
–frames, 76 –[no]create-views, 84
–help, 77 –[no]report, 85
–host, 77 pt-ioprofile command line option
–interval, 77 –aggregate, 92
–log, 77 –cell, 92
–master-server-id, 77 –group-by, 92
–monitor, 77 –help, 93
–password, 78 –profile-pid, 93
–pid, 78 –profile-process, 93
–port, 78 –run-time, 93
–print-master-server-id, 78 –save-samples, 93
–recurse, 78 –version, 93
–recursion-method, 78 pt-kill command line option
–replace, 78 –any-busy-time, 103
–run-time, 78 –ask-pass, 97
–sentinel, 79 –busy-time, 100
–set-vars, 79 –charset, 97
–skew, 79 –config, 97
–socket, 79 –daemonize, 97
–stop, 79 –defaults-file, 97
–table, 79 –each-busy-time, 103
–update, 79 –execute-command, 103
–user, 79 –filter, 97
–version, 80 –group-by, 98
–[no]insert-heartbeat-row, 77 –help, 98
pt-index-usage command line option –host, 98
–ask-pass, 84 –idle-time, 100
–charset, 84 –ignore-command, 100
–config, 84 –ignore-db, 100
–create-save-results-database, 84 –ignore-host, 100
–database, 84 –ignore-info, 101
–databases, 84 –ignore-state, 101
–databases-regex, 84 –ignore-user, 101
–defaults-file, 84 –interval, 98
–drop, 84 –kill, 103
–empty-save-results-tables, 85 –kill-query, 104

Index 319


–log, 98 –type, 111
–match-all, 101 –user, 112
–match-command, 101 –verbose, 112
–match-db, 101 –version, 112
–match-host, 102 –[no]results, 111
–match-info, 102 –[no]warnings, 112
–match-state, 102 pt-mysql-summary command line option
–match-user, 102 –config, 124
–password, 98 –databases, 124
–pid, 98 –help, 124
–port, 99 –read-samples, 124
–print, 104 –save-samples, 124
–query-count, 103 –sleep, 124
–replication-threads, 102 –version, 124
–run-time, 99 pt-online-schema-change command line option
–sentinel, 99 –alter, 127
–set-vars, 99 –alter-foreign-keys-method, 127
–socket, 99 –ask-pass, 128
–stop, 99 –charset, 129
–test-matching, 102 –check-interval, 129
–user, 99 –check-slave-lag, 129
–verbose, 103 –chunk-index, 129
–version, 99 –chunk-size, 129
–victims, 99 –chunk-size-limit, 129
–wait-after-kill, 100 –chunk-time, 130
–wait-before-kill, 100 –config, 130
–[no]ignore-self, 101 –critical-load, 130
–[no]strip-comments, 99 –defaults-file, 130
pt-log-player command line option –dry-run, 130
–ask-pass, 108 –execute, 130
–base-dir, 108 –help, 131
–base-file-name, 108 –host, 131
–charset, 108 –lock-wait-timeout, 131
–config, 108 –max-lag, 131
–defaults-file, 108 –max-load, 131
–dry-run, 108 –password, 131
–filter, 108 –pid, 131
–help, 110 –port, 132
–host, 110 –print, 132
–iterations, 110 –progress, 132
–max-sessions, 110 –quiet, 132
–only-select, 110 –recurse, 132
–password, 110 –recursion-method, 132
–pid, 110 –retries, 133
–play, 110 –set-vars, 133
–port, 110 –socket, 133
–print, 110 –user, 133
–quiet, 110 –version, 133
–session-files, 111 –[no]check-replication-filters, 129
–set-vars, 111 –[no]drop-old-table, 130
–socket, 111 –[no]swap-tables, 133
–split, 111 pt-query-advisor command line option
–split-random, 111 –ask-pass, 142
–threads, 111 –charset, 142

320 Index


–config, 142 –mirror, 159
–daemonize, 142 –order-by, 159
–database, 142 –outliers, 160
–defaults-file, 142 –password, 160
–group-by, 143 –pid, 160
–help, 143 –pipeline-profile, 160
–host, 143 –port, 160
–ignore-rules, 143 –print, 160
–password, 143 –print-iterations, 160
–pid, 143 –processlist, 161
–port, 143 –progress, 161
–print-all, 143 –read-timeout, 161
–query, 143 –report-all, 161
–report-format, 143 –report-format, 161
–review, 144 –report-histogram, 162
–sample, 144 –review, 162
–set-vars, 144 –review-history, 163
–socket, 144 –run-time, 165
–type, 144 –run-time-mode, 165
–user, 144 –sample, 166
–verbose, 144 –select, 166
–version, 144 –set-vars, 166
–where, 144 –shorten, 167
–[no]continue-on-error, 142 –show-all, 167
–[no]show-create-table, 144 –since, 167
pt-query-digest command line option –socket, 167
–apdex-threshold, 153 –statistics, 168
–ask-pass, 153 –table-access, 168
–attribute-aliases, 153 –tcpdump-errors, 168
–attribute-value-limit, 153 –timeline, 168
–aux-dsn, 153 –type, 169
–charset, 154 –until, 171
–check-attributes-limit, 154 –user, 171
–config, 154 –variations, 171
–create-review-history-table, 154 –version, 171
–create-review-table, 154 –watch-server, 171
–daemonize, 154 –[no]continue-on-error, 154
–defaults-file, 154 –[no]for-explain, 157
–embedded-attributes, 154 –[no]report, 161
–execute, 155 –[no]zero-admin, 172
–execute-throttle, 155 –[no]zero-bool, 172
–expected-range, 155 pt-show-grants command line option
–explain, 156 –ask-pass, 175
–filter, 156 –charset, 176
–fingerprints, 157 –config, 176
–group-by, 158 –database, 176
–help, 158 –defaults-file, 176
–host, 158 –drop, 176
–ignore-attributes, 158 –flush, 176
–inherit-attributes, 158 –help, 176
–interval, 159 –host, 176
–iterations, 159 –ignore, 176
–limit, 159 –only, 176
–log, 159 –password, 176

Index 321


–pid, 177 –config, 196
–port, 177 –daemonize, 196
–revoke, 177 –database, 196
–separate, 177 –defaults-file, 196
–set-vars, 177 –error-length, 196
–socket, 177 –error-numbers, 196
–user, 177 –error-text, 196
–version, 177 –help, 196
–[no]header, 176 –host, 196
–[no]timestamp, 177 –log, 197
pt-slave-delay command line option –max-sleep, 197
–ask-pass, 184 –min-sleep, 197
–charset, 184 –monitor, 197
–config, 184 –password, 197
–daemonize, 184 –pid, 197
–defaults-file, 184 –port, 197
–delay, 184 –quiet, 197
–help, 184 –recurse, 197
–host, 184 –recursion-method, 197
–interval, 185 –run-time, 198
–log, 185 –sentinel, 198
–password, 185 –set-vars, 198
–pid, 185 –skip-count, 198
–port, 185 –sleep, 198
–quiet, 185 –socket, 198
–run-time, 185 –stop, 198
–set-vars, 185 –until-master, 199
–socket, 185 –until-relay, 199
–use-master, 185 –user, 199
–user, 185 –verbose, 199
–version, 186 –version, 199
–[no]continue, 184 –[no]check-relay-log, 196
pt-slave-find command line option pt-stalk command line option
–ask-pass, 189 –collect, 203
–charset, 189 –collect-gdb, 203
–config, 189 –collect-oprofile, 204
–database, 190 –collect-strace, 204
–defaults-file, 190 –collect-tcpdump, 204
–help, 190 –config, 204
–host, 190 –cycles, 204
–password, 190 –daemonize, 204
–pid, 190 –dest, 204
–port, 190 –disk-bytes-free, 204
–recurse, 190 –disk-pct-free, 204
–recursion-method, 190 –function, 204
–report-format, 190 –help, 205
–set-vars, 191 –interval, 205
–socket, 191 –iterations, 205
–user, 191 –log, 205
–version, 191 –match, 206
pt-slave-restart command line option –notify-by-email, 206
–always, 195 –pid, 206
–ask-pass, 195 –prefix, 206
–charset, 195 –retention-time, 206

322 Index


–run-time, 206 –resume, 224
–sleep, 206 –retries, 224
–stalk, 206 –separator, 224
–threshold, 206 –set-vars, 224
–variable, 206 –socket, 224
–version, 206 –tables, 224
pt-summary command line option –tables-regex, 224
–config, 212 –trim, 224
–help, 212 –user, 225
–read-samples, 212 –version, 225
–save-samples, 212 –where, 225
–sleep, 212 –[no]check-replication-filters, 218
–summarize-mounts, 212 –[no]create-replicate-table, 219
–summarize-network, 212 –[no]empty-replicate-table, 220
–summarize-processes, 212 –[no]replicate-check, 223
–version, 212 pt-table-sync command line option
pt-table-checksum command line option –algorithms, 233
–ask-pass, 218 –ask-pass, 233
–check-interval, 218 –bidirectional, 234
–check-slave-lag, 218 –buffer-in-mysql, 234
–chunk-index, 218 –charset, 234
–chunk-size, 218 –chunk-column, 235
–chunk-size-limit, 219 –chunk-index, 235
–chunk-time, 219 –chunk-size, 235
–columns, 219 –columns, 235
–config, 219 –config, 235
–databases, 219 –conflict-column, 235
–databases-regex, 219 –conflict-comparison, 235
–defaults-file, 219 –conflict-error, 236
–engines, 220 –conflict-threshold, 236
–explain, 220 –conflict-value, 236
–float-precision, 220 –databases, 236
–function, 220 –defaults-file, 236
–help, 220 –dry-run, 237
–host, 220 –engines, 237
–ignore-columns, 220 –execute, 237
–ignore-databases, 221 –explain-hosts, 237
–ignore-databases-regex, 221 –float-precision, 237
–ignore-engines, 221 –function, 237
–ignore-tables, 221 –help, 237
–ignore-tables-regex, 221 –host, 238
–lock-wait-timeout, 221 –ignore-columns, 238
–max-lag, 221 –ignore-databases, 238
–max-load, 221 –ignore-engines, 238
–password, 222 –ignore-tables, 238
–pid, 222 –lock, 238
–port, 222 –lock-and-rename, 239
–progress, 222 –password, 239
–quiet, 222 –pid, 239
–recurse, 222 –port, 239
–recursion-method, 222 –print, 239
–replicate, 223 –recursion-method, 239
–replicate-check-only, 224 –replace, 240
–replicate-database, 224 –replicate, 240

Index 323


–set-vars, 240 –quantile, 255
–socket, 240 –run-time, 255
–sync-to-master, 240 –start-end, 255
–tables, 240 –type, 256
–timeout-ok, 240 –version, 256
–trim, 241 –watch-server, 256
–user, 241 pt-trend command line option
–verbose, 241 –config, 259
–version, 241 –help, 259
–wait, 241 –pid, 259
–where, 242 –progress, 259
–[no]bin-log, 234 –quiet, 259
–[no]buffer-to-client, 234 –version, 259
–[no]check-master, 234 pt-upgrade command line option
–[no]check-privileges, 234 –ask-pass, 263
–[no]check-slave, 234 –base-dir, 263
–[no]check-triggers, 235 –charset, 263
–[no]foreign-key-checks, 237 –clear-warnings-table, 263
–[no]hex-blob, 237 –compare, 263
–[no]index-hint, 238 –compare-results-method, 264
–[no]transaction, 241 –config, 264
–[no]unique-checks, 241 –continue-on-error, 264
–[no]zero-chunk, 242 –convert-to-select, 264
pt-table-usage command line option –daemonize, 264
–ask-pass, 248 –explain-hosts, 264
–charset, 248 –filter, 265
–config, 248 –fingerprints, 265
–constant-data-value, 248 –float-precision, 265
–create-table-definitions, 248 –help, 265
–daemonize, 249 –host, 265
–database, 249 –iterations, 266
–defaults-file, 249 –limit, 266
–explain-extended, 249 –log, 266
–filter, 249 –max-different-rows, 266
–help, 249 –order-by, 266
–host, 249 –password, 266
–id-attribute, 249 –pid, 266
–log, 249 –port, 266
–password, 249 –query, 266
–pid, 249 –reports, 266
–port, 249 –run-time, 266
–progress, 250 –set-vars, 267
–query, 250 –shorten, 267
–read-timeout, 250 –socket, 267
–run-time, 250 –temp-database, 267
–set-vars, 250 –temp-table, 267
–socket, 250 –user, 267
–user, 250 –version, 267
–version, 250 –zero-query-times, 267
–[no]continue-on-error, 248 –[no]clear-warnings, 263
pt-tcp-model command line option pt-variable-advisor command line option
–config, 255 –ask-pass, 277
–help, 255 –charset, 277
–progress, 255 –config, 278

324 Index


–daemonize, 278
–defaults-file, 278
–help, 278
–host, 278
–ignore-rules, 278
–password, 278
–pid, 278
–port, 278
–set-vars, 278
–socket, 278
–source-of-variables, 278
–user, 278
–verbose, 279
–version, 279
pt-visual-explain command line option
–ask-pass, 288
–charset, 288
–clustered-pk, 288
–config, 288
–connect, 288
–database, 288
–defaults-file, 288
–format, 288
–help, 288
–host, 288
–password, 288
–pid, 289
–port, 289
–set-vars, 289
–socket, 289
–user, 289
–version, 289

Index 325

Percona toolkit 2_1_operations_manual

More Related Content

What's hot (18)

Similar to Percona toolkit 2_1_operations_manual (20)

Recently uploaded (20)

Percona toolkit 2_1_operations_manual