MySQL user camp march 11th 2016

Copyright © 2015, Oracle and/or its affiliates. All rights reserved. |Tuesday, October 27, 2015 MySQL Central @ Oracle Open World 2015 - San Francisco
An Overview of the New Replication
Features in MySQL 5.7
Presentation by
Venkatesh Duggirala (venkatesh.duggiral@oracle.com)
Senior Principal Member Technical Staff
Slides prepared by
Luís Soares (luis.soares@oracle.com)
Principal Software Engineer, MySQL Replication Team Lead
1

Copyright © 2015, Oracle and/or its affiliates. All rights reserved. |
Safe Harbor Statement
The following is intended to outline our general product direction. It is intended for
information purposes only, and may not be incorporated into any contract. It is not a
commitment to deliver any material, code, or functionality, and should not be relied upon
in making purchasing decisions. The development, release, and timing of any features or
functionality described for Oracle’s products remains at the sole discretion of Oracle.
Tuesday, October 27, 2015 MySQL Central @ Oracle Open World 2015 - San Francisco 2

Program Agenda
Tuesday, October 27, 2015 MySQL Central @ Oracle Open World 2015 - San Francisco

Program Agenda
Background
Celebrating MySQL 5.7 Replication
The New Replication Features in MySQL 5.7
Summary
1
2
3
4

Celebrating MySQL 5.7 Replication
GA ready!
1

Interesting Stats Related to MySQL 5.7
• 40 replication worklogs pushed to MySQL 5.7
– 29 related to on MySQL replication Core
– 11 related to MySQL Group Replication.
• 8 contributions merged into MySQL 5.7.
• 19 major enhancements to the replication core.
• 14 refactoring and modularization related changes to the code.
• 1 entirely new out-of-core replication plugin: MySQL Group Replication.
– Still on labs, but exhibiting a rapid release rate.

Thank You!
• After two and half years baking the team is very happy to share the
outcome with you!
– We did it for you!
• Thank you for being part of it!
• Enjoy MySQL 5.7, in particular the replication features! ;)

Background2

Background: Replication Components
Tuesday, October 27, 2015
Insert...
Insert...
B
binary log
Insert...
relay log
Insert...
A
binary log
Client
Sender
thread
Receiver
thread
Applier
Threads
Receiver
Meta-data
Update
Applier
Meta-data
Update
Network
MySQL Central @ Oracle Open World 2015 - San Francisco 9

Background: Replication Components
• Binary Log
– File based logical log that records the changes on the master.
– Statement or Row based format (may be intermixed).
– Each transaction is split into groups of events.
– Control events: Rotate, Format Description, Gtid, ...
Layout of a
Binary Log File
BEGIN ...E1 E2 COMMIT BEGIN ...E1 E2 COMMIT

M S
S
S
S
M
write clients read clients
read clients
write clients
More
reads?
More
slaves!
Read scale-out
Background: What is Replication Used For?

M M?
write clients write clients
More
writes?
More
Masters?
M?
M?
MySQL Fabric helps sharding your data with MySQL...
What about write scale-out?

C
B
A
C
B
ACrash
C
B
A
B is the
new master
Uh Oh! Whew!
Redundancy: If master crashes, promote slave to master

M
S
write clients
business intelligent client applications
reporting client applications
big queries client applications
On-line Backup and Reporting

CB
BA
AC
Image from
www.ginkgomaps.com

Usability / Online
3
3.1
MySQL Central @ Oracle Open World 2015 - San Francisco

Online Reconfiguration of Global Transaction Identifiers
Insert...
Insert...
B
binary log
Insert...
relay log
Insert...
A
binary log
Client
Sender
thread
Receiver
thread
Applier
Threads
Receiver
Meta-data
Update
Applier
Meta-data
Update
Network

• The procedure is online.
– Both reads and writes are allowed while global transaction identifiers are being
turned on or off in the entire replication cluster.
• No need to synchronize servers, ever.
• No need to restart servers.
• No need to change replication topology.
– Works in arbitrary topologies.
• Should anything happen in the middle of the procedure, you can always roll
back.

• Over simplified procedure to turn on global transaction identifiers:
• Details: http://dev.mysql.com/doc/refman/5.7/en/replication-mode-
change-online-enable-gtids.html
1) SET @@GLOBAL.GTID_MODE = OFF_PERMISSIVE; (on every server)
2) SET @@GLOBAL.GTID_MODE = ON_PERMISSIVE; (on every server)
3) wait for a bit longer than your replication lag
4) SET @@GLOBAL.GTID_MODE = ON; (on every server)

Online Reconfiguration of Replication Filters
Insert...
Insert...
B
binary log
Insert...
relay log
Insert...
A
binary log
Client
Sender
thread
Receiver
thread
Applier
Threads
Receiver
Meta-data
Update
Applier
Meta-data
Update
Network

Online Reconfiguration of Replication Filters
• Change Slave's Replication Filters dynamically.
– No need to stop and restart the slave server for setting new replication filtering rules.
– All slave filters are supported.
– Values can be input in various character sets.
mysql> CHANGE REPLICATION FILTER REPLICATE_DO_DB= (db1, db2)

Online Reconfiguration of Replication Receiver/Applier
Insert...
Insert...
B
binary log
Insert...
relay log
Insert...
A
binary log
Client
Sender
thread
Receiver
thread
Applier
Threads
Receiver
Meta-data
Update
Applier
Meta-data
Update
Network

Change master from A to B without stopping the applier threads.
C
B
A
C
B
ACrash
C
B
A
B is the
new master
Uh Oh! Whew!

Change master from A to B without stopping the applier threads.
C
B
A
C
B
ACrash
C
B
A
Uh Oh! Whew!
DBA did not need C to stop applying its
relay log to change C's master from A to B.

• Enables more online operations during fail-over:
• Stopping, changing master and restarting the receiver thread are all done
while the applier thread is running.
• Change applier properties while the receiver thread is working:
mysql> STOP SLAVE IO_THREAD;
mysql> CHANGE MASTER TO MASTER_HOST='master2', …;
mysql> START SLAVE IO_THREAD;
mysql> STOP SLAVE SQL_THREAD;
mysql> CHANGE MASTER TO MASTER_DELAY=3600, …;
mysql> START SLAVE SQL_THREAD;

Improved Replication Monitoring
Insert...
Insert...
B
binary log
Insert...
relay log
Insert...
A
binary log
Client
Sender
thread
Receiver
thread
Applier
Threads
Receiver
Meta-data
Update
Applier
Meta-data
Update
Network
DBA

• Monitoring through SQL.
• Logically unrelated information into different places.
• Extensible, can be adapted to new features.
• More accurate and consistent identifier names.
• Master-slave, Multi-source, Group Replication support.
Performance Schema Replication Tables

Connection
Configuration
Connection
Status
Applier
Configuration
Applier
Status
(Slave) Receiver and Applier Status
Applier / Coordinator
Status
Workers
Status

mysql> select * from performance_schema.replication_applier_status_by_workerG
*************************** 1. row ***************************
CHANNEL_NAME:
WORKER_ID: 1
THREAD_ID: 35
SERVICE_STATE: ON
LAST_SEEN_TRANSACTION: 4ba0eb86-63c3-11e4-92ba-28b2bd168d07:2368
LAST_ERROR_NUMBER: 0
LAST_ERROR_MESSAGE:
LAST_ERROR_TIMESTAMP: 0000-00-00 00:00:00
*************************** 2. row ***************************
CHANNEL_NAME:
WORKER_ID: 2
THREAD_ID: 36
SERVICE_STATE: ON
LAST_SEEN_TRANSACTION: 4ba0eb86-63c3-11e4-92ba-28b2bd168d07:2367
LAST_ERROR_NUMBER: 0
LAST_ERROR_MESSAGE:
LAST_ERROR_TIMESTAMP: 0000-00-00 00:00:00
2 rows in set (0,00 sec)

Usability / Online
Performance
3
3.1
3.2

Improved Applier Throughput – Locking-based Parallelism
Insert...
Insert...
B
binary log
Insert...
relay log
Insert...
A
binary log
Client
Sender
thread
Receiver
thread
Applier
Threads
Receiver
Meta-data
Update
Applier
Meta-data
Update
Network

Single Threaded Apply 8 Threads Applier 24 Threads Applier 48 Threads Applier 96 Threads Applier
0%
50%
100%
150%
200%
250%
Fast and Scalable Multi-threaded Replication Applier
(statement-based replication, durable settings, SSD storage)
Slaveapplytime
(SysbenchRWonthemasterwith96threads=100%)
10X

• Leverage parallelization information from master execution:
– Concurrent transactions, which have not blocked each other on the master, are
marked as non-contending in the binary log.
• Meanwhile, at the slave:
– Non-contending transactions are scheduled in parallel;
– Concurrent transactions commit independently, thus no waiting involved.

• T2 is scheduled to execute
together with T1.
• T3 is scheduled to execute
as soon as T1 finishes.
T1
T2
T3
Time
Commit Finishes
Last Lock Acquired
Concurrent Execution History
on the Master
Execution
Commit
T1 and T2 execute in
parallel on the slave.
T2 and T3 execute in
parallel on the slave.

• Supports statement-based or row-based formats.
• Scheduling policy controlled through:
– logical_clock - means schedule based on the locking interval timestamps.
– database - the scheduling policy from 5.6 (concurrency control done per database).
• Work to improve slave scalability continues, does not stop here.
mysql> SET slave_parallel_type= [logical_clock|database]

Improved User Threads-Sender Threads Synchronization
Insert...
Insert...
B
binary log
Insert...
relay log
Insert...
A
binary log
Client
Sender
thread
Receiver
thread
Applier
Threads
Receiver
Meta-data
Update
Applier
Meta-data
Update
Network

Improved User Threads-Sender Threads Synchronization
• Concurrent reads by the sender thread with ongoing writes from user
threads.
– Sender thread does not block user sessions more than necessary.
– Higher throughput for both sender threads and user sessions.
Sender Thread Reads Binary Log User Thread Writes to Binary Log
Sender Thread Reads Binary Log
User Thread Writes to Binary Log
pre 5.7.2
5.7.2+

Faster Sender Thread
Insert...
Insert...
B
binary log
Insert...
relay log
Insert...
A
binary log
Client
Sender
thread
Receiver
thread
Applier
Threads
Receiver
Meta-data
Update
Applier
Meta-data
Update
Network

• Send buffer is not allocated and freed every time an event is sent.
• When sender threads require larger send buffer - buffer grows as needed.
• When buffer is larger than needed - buffer shrinks dynamically.
• Together with the sender thread enhancements released on MySQL 5.7.2:
– increases master scalability;
– reduces resource consumption (CPU);
– Master, with dump threads attached, copes better with peak loads.

• Ad hoc microbenchmark using
mysqlslap
• 1M queries, concurrency=200,
commit=1
• N slaves attached (fake slaves using
remote mysqlbinlogs)
• 48 cores HT / 512 GB RAM / HDD
0
2000
4000
6000
8000
10000
12000
14000
16000
0 1 5 10 20 30 40 50 100
QpS
Number of Connected Sender Threads
Master’s Throughput
MySQL 5.6.16 MySQL 5.7.4

Faster Semi-sync Replication – ACK Receiver Thread
Insert...
Insert...
B
binary log
Insert...
relay log
Insert...
A
binary log
Client
ACK
Receiver
thread
Receiver
thread
Applier
Threads
Receiver
Meta-data
Update
Applier
Meta-data
Update
Network
Sender
thread

Faster Semi-sync Replication – ACK Receiver Thread
• Consecutive transactions do not block each other while waiting for ACKs
– Transaction t1 and t2 are sent immediately to the slave by the sender thread
– ACKs are received only by a special thread
– Transaction t2 does not include t1 round trip in its semi-sync overall latency
• Thread starts when semi-sync is activated
• Thread stops when semi-sync is deactivated
mysql> SET GLOBAL rpl_semi_sync_master_enabled= ON
mysql> SET GLOBAL rpl_semi_sync_master_enabled= OFF

Faster Semi-sync: Durability over the Network
Orange bar percentage shows the throughput gain (orange) when compared to the corresponding
blue bar. Blue and stacked bars show throughput ratio computed against the best blue bar.

Usability / Online
Performance
Dependability
3
3.1
3.2
3.3

Loss-less Semi-sync Replication
Insert...
Insert...
B
binary log
Insert...
relay log
Insert...
A
binary log
Client
Sender
thread
Receiver
thread
Applier
Threads
Receiver
Meta-data
Update
Applier
Meta-data
Update
Network

Master
Slave
T1: INSERT INTO t1 VALUES (1000)
ACK
Time
T2: SELECT * FROM t1; empty set
execute prepare binlog
relay log
commitexecute

• Master waits for slave's ACK before committing (as opposed to: master
waits for slave's ACK after committing).
– Therefore, concurrent transactions do not see changes while this transaction waits for
ack.
• Should a master fail, then any transaction that it may have externalized is
also persisted on a slave.
• User can choose between the original semi-sync behavior and the new
one.
mysql> SET rpl_semi_sync_master_wait_point= [AFTER_SYNC|AFTER_COMMIT]

Semi-sync Replication – Wait for Multiple ACKs
Insert...
Insert...
B
binary log
Insert...
relay log
Insert...
A
binary log
Client
Sender
thread
Receiver
thread
Applier
Threads
Receiver
Meta-data
Update
Applier
Meta-data
Update
Network

• Master does not commit transaction until it receives N ACKs from N slaves.
• Dynamically settable:
mysql> SET GLOBAL rpl_semi_sync_master_wait_for_slave_count= N

• Master does not commit transaction until it receives N ACKs from N slaves.
• Dynamically settable:
mysql> SET GLOBAL rpl_semi_sync_master_wait_for_slave_count= N
Master
Slave 2
T1: COMMIT
ACK
Slave 1
ACK
T1: COMMIT
succeeds
mysql> SET GLOBAL rpl_semi_sync_master_wait_for_slave_count= 2
Slave 3
Time
relay log
relay log
relay log
ACK

Usability / Online
Performance
Dependability
Flexibility
3
3.1
3.2
3.3
3.4

Storing Global Transaction Identifiers in a Table
Insert...
B
Insert...
relay log
Insert...
A
binary log
Client
Sender
thread
Receiver
thread
Applier
Threads
Receiver
Meta-data
Update
Applier
Meta-data
Update
Network

• Slaves can use GTIDs with binary logs disabled.
– Slaves that are never candidates to become a master can still use GTIDs for auto
positioning.
Use Cases

• GTIDs are saved into a mysql system table (range-compressed).
• GTIDs are inserted into the table as transactions commit.
• A compression thread runs periodically and compresses the table into
ranges.
– New global variable controls the period: gtid_executed_compression_period.
Mechanics
CREATE TABLE gtid_executed(
source_uuid CHAR(36) NOT NULL,
interval_start BIGINT NOT NULL,
interval_end BIGINT NOT NULL,
PRIMARY KEY(source_uuid, interval_start)
);
mysql> SET GLOBAL gtid_executed_compression_period= N;(N – number of transactions)

• Binary Log Enabled
– Store GTID in binary log (transactionally).
– Copy GTID set to table on binary log rotation.
• Binary Log Disabled
– Store GTID in table (transactionally).
– New transactions are do not get an automatic GTID assigned.
• Always store GTIDs transactionally.
Mechanics

Multi-Source Replication
S
M M Slave can have more than one master.M M
The need for gathering data in a central server:
• Integrated backup;
• Complex queries for analytics purposes;
• Data HUB for inter-cluster replication.

• A server (slave) can replicate from multiple sources (masters).
• Multiple channels (channel: receiver thread, relay logs, applier threads) that
can be stopped started individually.
• Integrated with Multi-threaded Slave (each channel has its own multi-
threaded applier set of threads).
• Integrated with the new P_S tables.
– replication_applier_status_by_coordinator shows multiple entries, one per
channel/source coordinator.
– replication_applier_status_by_worker shows multiple entries, multiple entries per
channel
– replication_connection_status shows multiple entries, one per connection to different
sources.

• Integrated with GTIDs.
• Integrated with crash-safe tables.
– Progress state is stored in these tables for recoverability purposes.
• Virtually no limit on the number of sources (capped to 256, but can be
changed if server binary is rebuilt).
• Able to manage each source separately.

Usability / Online
Performance
Dependability
Flexibility
Misc
3
3.1
3.2
3.3
3.4
3.5

“Smaller”, yet interesting, enhancements!
• Multi-threaded applier is able to retry failed transactions.
• Option to make the Multi-threaded applier preserve commit order.
• SSL options for mysqlbinlog tool.
• Rewrite DB rules for the mysqlbinlog tool.
• Function to wait for transactions to be applied, regardless of the replication stream
they come from.
• Options to track global transaction identifiers in the return packet of the mysql
protocol. Useful for tracking GTID session state and enabler for session consistency.
• Support for XA transactions when the binary log is enabled.
• Change in the defaults. E.g., binlog_format=ROW and sync_binlog=1.
• Options to fine tune the binary log group commit procedure.

Summary4

Summary
• Replication feature set in MySQL 5.7 shows many improvements:
– Semi-sync is more flexible and faster.
– Less contention on the binary log.
– Slave is faster and provides an inter-transaction parallelization scheme.
– More online reconfiguration: Global Transaction Identifiers, filters, configuration.
– Improved monitoring.
– Even more flexible replication with multi-source enabling new deployment scenarios.
– Enhancements all over: transaction retries, sequential commits, XA support, new
replication functions, security.

Where to go from here?
• Packages
– http://dev.mysql.com
– http://labs.mysql.com
• Reference Documentation
– http://dev.mysql.com/doc/#manual
• Blogs from the Engineers (news, quirks, technical information and much
more)
– http://mysqlhighavailability.com

MySQL user camp march 11th 2016

MySQL user camp march 11th 2016

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to MySQL user camp march 11th 2016 (20)

Recently uploaded (20)

MySQL user camp march 11th 2016