[Pgpool-general] pgpool-II and online recovery process
Tatsuo Ishii
ishii at sraoss.co.jp
Thu Sep 11 01:34:22 UTC 2008
> Hi,
>
> I have recently setup a VM environment to test out an online recovery
> process which works great. Basically I have the following.
>
> pgpool-II 2.1 server:
> 1 - pgpool
>
> 3 postgresql 8.3.3 servers:
> 1 - db1 (backend_node0)
> 2 - db2 (backend_node1)
> 3 - db3 (backend_node2)
>
> The 3rd PG server is kept in detached status so that I can use it for
> online recovery tests.
> I have created the scripts for 1st/2nd stage and also the
> pg_remote_start script.
> I can use pcp_recovery_node to bring a new node online (3rd PG server)
> or recover one of the existing ones without any issues.
>
> Now, until today all my previous tests on performing online recovery
> involved calling pcp_recovery_node and during that time no clients
> were using the database servers through pgpool of course. There was no
> activity going on on the database servers at all. So everything
> worked great, pgpool went through 1st and 2nd stage and then called
> the remote start and finished the recovery process. So server got
> online and sync'd.
>
> Then, today I tried to do the same but right before calling
> pcp_recovery_node I started a pgbench process pointing to the pgpool
> server to create some activity on the database. I was under the
> impression that during 2nd stage pgpool would perhaps start to queue
> some of the transactions along with not allowing new clients to
> connect to it. Allowing the 2nd stage to occur and then bring the new
> node online. Once the online recovery was finished, pgpool would go
> through its queue and send those transactions to all nodes.
> Is that not the case ? Cause basically it went through 1st stage and
> then pcp_recovery_node timed out.
Perhaps you need -C option to pgbench?
--
Tatsuo Ishii
SRA OSS, Inc. Japan
More information about the Pgpool-general
mailing list