[pgpool-general: 4033] Re: Fwd: Nodes not detatched

[Tatsuo Ishii]

> Hi there,
> I configured a pgpool with three node in Load Balance + Connection Pooling
> + Master Salve modes, so I have node 0 as master and node 1 and 2 as slaves.
> 
> I had a problem with both slaves (no disk space left on servers) and pgpool
> stopped to accept connections with this error:
> 
> psql -p 9999 -h /tmp
> Password:
> psql: ERROR:  unable to read message kind
> DETAIL:  kind does not match between master(53) slot[1] (4e)
> 
> 
> 
> However, slaves was continuing to accept connections, but was returning
> some worning:
> 
> psql -p 5997 -h 10.0.0.3
> Password:
> WARNING:  could not create relation-cache initialization file
> "global/pg_internal.init.19889": No space left on device
> DETAIL:  Continuing anyway, but there's something wrong.
> WARNING:  could not create relation-cache initialization file
> "base/12150/pg_internal.init.19889": No space left on device
> DETAIL:  Continuing anyway, but there's something wrong.
> psql (9.4.4)
> Type "help" for help.
> 
> postgres=#
> 
> 
> 
> 
> In this situation I was expected pgpool can detach both slave nodes and can
> continue to work with only master node, instead described situation has
> occurred.

Do you enable health checking? If so, does the health checking
detaches the standbys?

> Now the question is: can pgpool detach automatically slave nodes that go in
> a wrong state? Or is use of pcp_detach command the only way to obtain my
> desiderata?

If the health checking does not help, yes.

Problem is, PostgreSQL regards this not to be an error (if error, the
message from psql would be ERROR or FATAL, rather than WARNING). So
pgpool-II cannot automatically regards the disk full state as an
error.

Best regards,
--
Tatsuo Ishii
SRA OSS, Inc. Japan
English: http://www.sraoss.co.jp/index_en.php
Japanese:http://www.sraoss.co.jp