[Pgpool-general] Questions about online recovery

[Tatsuo Ishii]

> Hi,
> 
> I've a pgpool-II replication + load balance environment with seven
> backend servers running PostgreSQL 8 and CentOS 5.
> 
> The backend 0 (master node) runs both pgpool-II and PostgreSQL. I'm
> setting up online recovery with rsync using the same script for both 1st
> and 2nd stage commands.
> 
> Considering a scenario where the backend 0 (which runs pgpool-II and
> PostgreSQL) PostgreSQL is disconnected from pgpool-II, I have the
> following questions:
> 
> * How online recovery is supposed to work in this case?

1) backend 1 is chosen as the new master node.

2) pgpool-II connects to PostgreSQL on backend 1. This connection is
   different from ordinary backend connection.

3) pgpool-II requests backend 1 to execute CHECKPOINT.

4) pgpool-II requests backend 1 to execute "SELECT pgpool_recovery"
   which will execute online recovery stage 1 command.

5) pgppool-II waits until all client connection disconnected.

6) pgpool-II requests backend 1 to execute CHECKPOINT.

7) pgpool-II requests backend 1 to execute "SELECT pgpool_recovery"
   which will execute online recovery stage 2 command.

8) pgpool-II requests backend 1 to execute "SELECT pgpool_remote_start"
   which will execute "pgpool_remote_start" script, which is supposed
   to remote start postmaster on backend 0.

9) backend 0 connected again and recovery done.

> * Do I need to have pgpool-II installed on all backend servers? If so, I
> will have dummy pgpool-II processes running on backend serves just to
> have the pgpool communication manager running and be able to run
> pcp_recovery_node from another backend?

No. What you need on all backend servers are:

1) C functions under pgpool-II/sql/pgpool-recovery installed
   (pgpool_recovery and pgpool_remote_start).

2) pgpool_remote_start script installed under PostgreSQL database
   cluster directory (/usr/local/pgsql/data).

3) recovery scripts installed under PostgreSQL database cluster
   directory (/usr/local/pgsql/data).

> * How 2nd stage of online recovery is supporsed to work if I am
> recovering from a different backend than the one running pgpool-II and
> PostgreSQL will continue accepting connections? Is cascading pgpool-II
> connections to the backend servers a correct solution or workaround
> here?

See above.
--
Tatsuo Ishii
SRA OSS, Inc. Japan