PostgreSQL9.3 Switchover/Switchback

© 2013 EnterpriseDB Corporation - All rights reserved. 1

Streaming ReplicationSwitchover/SwitchbackVibhor Kumar

Principal System Engineer.


• PostgreSQL Streaming Replication

• Limitations in PG Version < 9.3

• Understanding Timelines

• New in PostgreSQL >=9.3

• Possible switchover/switchback

• Things to know about switchover/switchback

• recovery.conf

Agenda


PostgreSQL Streaming Replication (SR)

• Streaming Replication (SR) provides capability to continuously ship and apply WAL/XLOG records to any number of standbys.

• Introduced in PostgreSQL 9.0

• Currently two types supported:− Synchronous SR− Asynchronous SR


History: Streaming Replication

• 9.0: Asynchronous SR

• 9.1: − Synchronous SR− pg_basebackup

• 9.2:− Cascaded Replication− Backup from Standby− New Sync replication mode. (remote_write, local write)

• 9.3− Streaming only replication.− Follow timeline changes.


Limitations before 9.3

• Timeline streaming issue

• Intentional shutdown on master can leave Standby behind the master before 9.1.

• If Standby is behind master:− Promote lead to new timeline.− Difference in Master and Standby.− Old Master can’t be standby of new master.− Need to take fresh backup or manual copy timeline file to old master.

• Painful to take fresh backup if database size is large.


Understanding Timelines

• Introduced with Point-in-Time-Recovery in version 8.0

• Every time you do PITR, a new timeline is formed

• Timelines helps to differentiate WAL generated from different PITR.

• When new WAL generated after PITR, PostgreSQL shouldn’t overwrite old WAL.


More about Timelines


More about Timelines


Timeline history file


New in 9.3 SR

• Important Patches:− commit 985bd7d49726c9f178558491d31a570d47340459

− Author: Fujii Masao− Committed on Wed Jun 26 02:14:37 2013 +0900− Make sures when we shutdown the master, walsender should tries to send all outstandng WAL records to Standby, and then to exit.

− Back patched till 9.1− commit abfd192b1b5ba5216ac4b1f31dcd553106304b19

− Author: Heikki Linnakangas− Committed on Thu Dec 13 19:00:00 2012 +0200− Allow a streaming replication standby to follow a timeline switch. (Only in 9.3)


9.3: Now switchover/Switchback is possible

• Difference between Switchover and Failover. (add points)

• New commit in PostgreSQL helps:

− Clean shutdown for maintenance operation of master server

− Old master server can be new standby for New Master.

− No more fresh backup in intentional promotion.


Things to know: switchover/switchback

• Use –m fast or –m smart shutdown on Master for clean shutdown.

• Before promoting standby:− Make sure all WAL sent by Master applied on elected node for New Master.

− Use following function for when to promote:− pg_last_xlog_receive_location− pg_last_xlog_replay_location

− Have proper archive location accessible to old Master.

− Set proper archive_command


recovery.conf

recovery_target_timeline = 'latest'standby_mode = 'on’primary_conninfo = 'user=repuser password=repuser host=172.17.0.3 port=5432restore_command = ‘cp /location/%f %p’


EFM: Promote Steps

• When promote happens, EFM does following:− Send prepare information to Master, Witness and Slave

− Agent on Master get promote signal and does following:− Release VIP if any.− create dummy recovery.conf, so that next startup master shouldn’t come up.

− Message about Master failed.

− Standby signaled to be promoted by trigger.

• Explore EFM for possible switchover/switchback


Demo


Thanks

Presentations & Public Speaking

PostgreSQL9.3 Switchover/Switchback