Highly efficient backups with percona xtrabackup

Highly Efficient Backups with
Percona Xtrabackup
PLMCE-2015 (MySQL 101)
Nilnandan Joshi
Support Engineer
Valerii Kravchuk
Principal Support Engineer

Introduction
❖ Percona Xtrabackup provides a fully open-source, free, high-
performance, non-blocking backup system for InnoDB or
XtraDB tables
❖ It can also backup MyISAM, Archive, Merge and other SQL-
level objects
❖ It is a reliable, widely-used alternative to Oracle's MySQL
Enterprise Backup

Features
Open Source (GPL) Encrypted backups Point-in-time recovery
support
Rsync support to minimize lock
time
Cost: Free Streaming backups Safe slave backups Improved FTWRL handling
Supported all MySQL
flavors
Parallel local backups and
compression
Compact backups Backup history table
Non-blocking InnoDB
backups*
Parallel encryption Buffer pool state backups Data & index file statistics
Incremental
backups
Partial backups(with
individual partition)
MySQL replication & Galera
Support
Individual tables or partitions
export
Full/Incremental compress
backup
Full Binary Backups Community Help/Support Throttling
For more info: http://www.percona.com/doc/percona-xtrabackup/2.2/intro.html

Compatibility
❖ Version 2.1 and 2.2 are compatible with:
Oracle MySQL 5.1 (plugin version), 5.5, 5.6,
corresponding versions of Percona Server, Percona
XtraDB Cluster and MariaDB, including MariaDB 10.0.x.
❖ Versions <= 2.0 are compatible with :
Oracle MySQL, 5.1, 5.5 and 5.6, equivalent versions of
Percona Server and MariaDB.
❖ MySQL 5.7 support is planned in upcoming 2.3, see
https://bugs.launchpad.net/percona-xtrabackup/+bug/1437415

xtrabackup vs. innobackupex
Percona XtraBackup includes two main executables:
❖ xtrabackup is the low level binary that handles the
creation and consistency checking of InnoDB backups.
❖ innobackupex is a higher-level Perl wrapper for
xtrabackup and is also responsible for the rest of the
operations, including the backup of other engines. You usually should use
innobackupex.
❖ Starting with Xtrabackup 2.3 the only official and recommended tool will
be xtrabackup, functionality of innobackupex has already been rewritten
in C and added to xtrabackup. 2.3 will be available "soon".

Installation
❖ The most straightforward way is to install from the Percona apt or
yum repositories.
❖ Then install the package with:
➢ yum install percona-xtrabackup
➢ aptitude update && aptitude install percona-xtrabackup
➢ apt-get install percona-xtrabackup
❖ Other installation options can be found at: http://www.percona.
com/downloads/XtraBackup/

Requirements and Limitations
❖ Only fully supported on Linux machines
❖ innobackupex is a Perl script , Package should take care of all
dependencies (Perl interpreter, mysql driver, other libraries)
❖ Local access to the MySQL datadir is required.
❖ READ, WRITE (?) and EXECUTE (?) filesystem privileges are needed.
❖ This makes xtrabackup incompatible with RDS or some web hosting
providers.

How Percona XtraBackup Works: Full Backup

How Percona XtraBackup Works: Full Restore

Command Line Execution
❖ Minimal options for a full backup:
# innobackupex /path/to/backup/
❖ Actual backup is stored in a timestamped subdirectory. It
reads the datadir and innodb paths from my.cnf
❖ A more complete set of options:
# innobackupex --host=localhost --user=root --password=secret /tmp

Files in Target Directory
❖ Hierarchy of files and directories mirroring the original database
structure.
❖ backup-my.cnf : Not a backup of the server configuration. It only
contains the minimal InnoDB settings at the time the backup to
execute the --apply-log phase
❖ xtrabackup_info: Contains all informations about xtrabackup binary,
server version, start and end time of backup, binlog position etc.
❖ xtrabackup_checkpoints: Metadata about the backup (type of
backup, lsn, etc.)
❖ xtrabackup_logfile: Data needed for the –apply-log phase.

Preparation Phase
❖ Before the backup directory can be used, we must:
➢ Make sure that the InnoDB tablespaces have consistent, non-
corrupt data
➢ Create new transaction logs
❖ innobackupex executes xtrabackup twice:
➢ First, to do a controlled crash recovery, applying the log entries
with xtrabackup's embedded InnoDB. This redoes all written
committed transactions and undoes uncommitted ones.
➢ Second, to create empty logs

Preparation Phase (cont)
❖ Only xtrabackup/innobackupex and the backup directory are
needed
❖ It can be done on different machine than the original server (we
have the xtrabackup binary and the innodb configuration
parameters)
❖ Generally, you do it just before restore. If you only do full backups
and want to minimize the MTTR (Mean Time To Recover), you can
prepare immediately after backup

Command Line Execution
❖ The simplest form:
$ innobackupex --apply-log
/home/nilnandan/backup/2015-03-03_12-14-55
❖ You may specify a memory parameter:
$ innobackupex --apply-log --use-memory=4G
/home/nilnandan/backup/2015-03-03_12-14-55
❖ This increases the available buffer pool for the recovery phase
(100M by default), to improve performance.

Restore
❖ After --apply-log, we have a consistent full raw backup and we can
just move it back to the datadir with standard OS commands once
the server is shutdown.
❖ As a physical backup, restoration is as fast as you can copy the
directory contents to the local filesystem
❖ There are specific commands for restoration in XtraBackup

Copy Back/Move Back
❖ The integrated command to restore a backup is:
# innobackupex –copy-back /home/nilnandan/backup/2015-03-03_12-14-55
❖ It complains if the datadir, as read from the config file, is not empty [1] (it will
not overwrite existing Data) Only partial restores can be done to a running
server xtrabackup_* files are not needed.
[1] We can override this behavior with the option --force-non-empty-directories, but it will never overwrite
existing files
❖ As an alternative to --copy-back, we can also do:
# innobackupex –move-back /home/nilnandan/backup/2015-03-03_12-14-55
❖ This may be faster or more convenient if the backup is already on the same
partition or don't have enough free space.

Restoring Permissions
❖ After restore, as file permissions are preserved, you will probably
have to fix file ownership:
# chown -R mysql:mysql /var/lib/mysql
❖ If MySQL cannot write to the data directory or some of it files, it will
refuse to start

Incremental Backup
❖ Differential backup: saves only the difference in the data since the last full backup
❖ Incremental backup: saves only the difference in the data since the last backup (full
or another incremental)
❖ Xtrabackup can perform both, which can lead to an important save in space (and
potentially, time), specially if the backups are very frequent or the database is very
static.
❖ Differential and incremental backups work the same way with xtrabackup:
➢ Every page stores the Log Sequence Number (LSN) when it was last changed
➢ Only pages that have a higher LSN than the base backup (full or incremental)
are copied
❖ With PS 5.6, one can set innodb_track_changed_pages variable which keep track of
changed pages. It can be used to speed up incremental backups by removing the
need to scan whole data files to find the changed pages. For more info
http://www.percona.com/doc/percona-server/5.6/management/changed_page_tracking.html

Incremental Backup(cont..)
❖ Restoration of Incremental Backups:
➢ On prepare time, the new pages have to be applied in order to
the full backup before restore.
➢ This is the reason why you want to wait to fully apply the log
just before restore time
➢ Uncommitted transactions must not be undone, as they may
have finished on a subsequent incremental backup
➢ Intermediate backups have to apply changes with the option --
redo-only
Note : Only works for InnoDB tables, MyISAM tables still copied fully.

Partial Backups
❖ When using xtrabackup (or innobackupex), we can filter out some of the tables,
effectively copying only some databases or tables.
❖ Two main restrictions apply for InnoDB tables:
➢ Original database must use innodb_file_per_table, as xtrabackup works at
filesystem level
➢ Recovering individual tables is only possible for MySQL 5.6. previous versions
require Percona XtraDB extensions with tablespace import enabled
❖ Command Line Options for Partial Backups:
➢ --include : Matches the fully qualified name (database.table) against a regular
expression
➢ --databases : Matches a list of databases (or tables) separated by spaces.
➢ --tables-file : Matches a list of tables, one per line,stored on the specified file

Streaming
❖ Xtrabackup allows the streaming of the backup files to the UNIX standard output.
This allows to post-process them or send them to a remote location while the
backup is ongoing.
❖ The stream option accepts two parameters:
➢ --stream=tar : It sends all files to the standard output in the tar archive format
(single file without compression
➢ --stream=xbstream : The standard output will using the new included
xbstream format, which allows simultaneous compression, encryption and
parallelization. A temporary file path is still required

Streaming(Cont.)
❖ Using tar:
# innobackupex --stream=tar /tmp > backup.tar
# innobackupex --stream=tar /tmp | bzip2 - > backup.tar.bz2
To extract, the tar -i option (ignore zeros) must be used:
# tar -xivf backup.tar
# tar -jxif backup.tar.bz2
Note: only GNU tar can extract archives created by xtrabackup.
❖ Using xbstream:
# innobackupex --stream=xbstream /tmp > backup.xbstream
To extract, use the xbstream binary provided with Percona Xtrabackup:
# xbstream -x < backup.xbstream -C /root

Compression and xbstream
❖ We can use --compress to produce quick compressed files:
# innobackupex --compress /media/backups
❖ To decompress, we can use --decompress or qpress* to decompress the
individual .qp files manually:
# innobackupex --decompress /media/backups/2017-07-08_13-22-12
# find . -name "*.qp" -execdir qpress -d {} . ;
❖ The --compress flag can be combined with xbstream (not with tar):
# innobackupex --compress --stream=xbstream /tmp
> /media/backups/backup.xbstream
❖ Before preparation, we must use first xbstream, then decompress:
# xbstream -x -C /media/backups < backup.xbstream
# innobackupex --decompress /media/backups

Parallel Copy and Compression
❖ If using xbstream, we can use --parallel to archive several files at the
same time:
# innobackupex --parallel=4 --stream=xbstream /
tmp > /media/backups/backup.tar
Note: parallel option works with streaming as well as local backups but it doesn't work for TAR
streaming.
❖ If we are using compression, we can parallelize the process for each
file with –compress-threads
# innobackupex --compress --compress-threads=8
--stream=xbstream /tmp > /media/backups/backup.tar

Remote Backups
❖ Xtrabackup can send automatically the backup files to a remote host by redirecting
the output to applications like ssh or netcat:
# innobackupex --stream=tar /tmp | ssh user@host "cat - > /media/backups/backup.tar"
# ssh user@host "( nc -l 9999 > /media/backups/backup.tar & )"
&& innobackupex --stream=tar /tmp | nc host 9999
❖ Throttling Remote Backups : pv (pipe viewer) utility can be useful for both
monitoring and limiting the bandwidth of the copy process. The following limits the
transmission to 10MB/s:
# innobackupex --stream=tar ./ | pv -q -L10m | ssh user@desthost
"cat - > /media/backups/backup.tar"

Compact Backups
❖ In order to reduce backup size, we can skip the copy of secondary
index pages:
# innobackupex --compact /media/backups
❖ The space saved will depend on the size of the secondary indexes.
For example:
#backup size without --compact : 2.0G 2015-02-01_10-18-38
#backup size with --compact : 1.4G 2015-02-01_10-29-48

Compact Backups(cont.)
❖ The downside of compact backups is that the prepare phase will take longer, as the
secondary keys have to be regenerated ( --rebuild-indexes ):
# cat xtrabackup_checkpoints
backup_type = full-backuped
from_lsn = 0
to_lsn = 9541454813
last_lsn = 9541454813
compact = 1
# innobackupex --apply-log --rebuild-indexes /media/backups/2017-07-08_13-22-12
❖ Rebuilding Indexes: The --rebuild-threads option can be used to speed up the
recreation of the keys, by doing it in parallel:
#innobackupex --apply-log --rebuild-indexes --rebuild-threads=8
/media/backups/2017-07-08_13-22-12
❖ Rebuilding indexes has the side effect of defragmenting them, as they are done in
order. We can rebuild regular full-backups, too.

Troubleshooting and Performance
❖ Locking Issues: By default, innobackupex will lock the tables (FTWRL) during the last
phase of the backup. You can use –no-lock
➢ if you are only backing up InnoDB tables and no DDL are executing on that last
Phase*
➢ if you backup other engines if no DDLs are executed, and non- innodb tables
are not modified (e.g. users are created on the mysql database)
➢ You can use --rsync to minimize the copy time when locked
* Please note that --no-lock also makes impossible to obtain a reliable binary log coordinates.
❖ File Not Found : Set a correct datadir option on /etc/my.cnf
2017-07-08 12:56:56 7f0dfe4737e0 InnoDB: Operating system error number 2 in a file operation.
InnoDB: The error means the system cannot find the path specified.
2017-07-08 12:56:56 7f0dfe4737e0 InnoDB: File name ./ibdata1
2017-07-08 12:56:56 7f0dfe4737e0 InnoDB: File operation call: 'open' returned OS error 71.
2017-07-08 12:56:56 7f0dfe4737e0 InnoDB: Cannot continue operation.
innobackupex: Error: ibbackup child process has died at /usr/bin/ innobackupex line 389.

❖ Permission Denied : Use the Linux root user/sudo or any other account
with read permissions on datadir and on target-dir.
InnoDB: read-write can't be opened in ./ibdata1 mode
xtrabackup: Could not open or create data files. [...]
xtrabackup: error: xb_data_files_init() failed with error code 1000
innobackupex: Error: ibbackup child process has died at /usr/bin/ innobackupex line 389.
innobackupex: Error: Failed to create backup directory /backups/
2017-07-08_12-19-01: Permission denied at /usr/bin/innobackupex line 389.
❖ MySQL Does Not Start After Recovery :
Starting MySQL server... FAILED
The server quit without updating PID file
❖ Did you change file permissions of the files to the mysql user?
❖ Did you remember to apply the log before restart? Failing to do so can also corrupt the
backup.
❖ Did you restore an appropriate my.cnf file? Previous versions of MySQL (< 5.6) do not start if
transaction log files have a different size than specified/default

❖ The Backup Takes Too Long : As MySQL performs physical backups, the
speed should be close to a filesystem copy. If IO impact is not a concern, we
can speed it up by:
➢ Using parallelization : ( --parallel , --compression-threads , --encrypt-
threads , --rebuild-threads )
➢ Using compression : ( --compress ) if the extra CPU compensates the
lower bandwidth required
➢ Using incremental backups : (specially with Percona extensions) to
perform rolling full backups

Known Bugs
● To check of Report bug, you can visit:
https://launchpad.net/percona-xtrabackup/+bugs

Valerii Kravchuk
Principal Support Engineer
http://mysqlentomologist.blogspot.com/
https://www.facebook.com/valerii.kravchuk
valerii.kravchuk@percona.com
Nilnandan Joshi
Support Engineer
http://nilinfobin.com/
nilnandan.joshi@percona.com

Highly efficient backups with percona xtrabackup

More Related Content

What's hot (20)

Viewers also liked (11)

Similar to Highly efficient backups with percona xtrabackup (20)

Recently uploaded (20)

Highly efficient backups with percona xtrabackup