[Pgpool-general] Online recovery on streaming replication + hot standby

[Tatsuo Ishii]

> When doing a failover, shoudn't pgpool-II keep the new master as the
> master from that point on?
> 
>  
> 
>>From my experience, after the master database fails and the slave takes
> over as master, if pcp_recovery_node is used to recover the old master,
> pgpool-II is assumes the master is back (because pgpool seems to mainain
> its master at backend0). I think pgpool should attach any other database
> as an slave, therefore not assuming it is the master again if it's
> backend0.
> 
>  
> 
> I'm tempted to stop pgpool-II, change its configuration file to swap
> backend values (backend0 with backend1 values and viceversa) and restart
> pgpool. This way, pgpool would work as I expect, but of course it's a
> hack. Is there a better idea around this?
> 
>  
> 
> I'm using: pgpool-II 3.0.2 (CVS) and PostgreSQL 9.0.1, in a two-backend
> configuration.

Next version of pgpool-II will be able to deal with the
problem. pgpool-II will have info regarding which node is the
primary. For example it will be able to deal with following situation:

backend0: master: primary
backend1: standby
|
primary goes down and standby promotes to primary
|
V
backend0: master: down
backend1: primary
|
backend0 recovered and becomes a standby
|
V
backend0: master: standby
backend1: primary
|
backend1 goes down and backend0 promotes to primary 
|
V
backend0: master: primary
backend1: down
|
backend1 recovered and becomes a standby
|
V
backend0: master: primary
backend1: standby

I have written a tutorial which can handle this kind of situation.
Please take a look at:
http://pgpool.projects.postgresql.org/contrib_docs/simple_sr_setting/index.html
--
Tatsuo Ishii
SRA OSS, Inc. Japan
English: http://www.sraoss.co.jp/index_en.php
Japanese: http://www.sraoss.co.jp