[pgpool-general: 2056] Re: trouble with recovery of a downed node
ishii at postgresql.org
Wed Aug 21 11:44:07 JST 2013
> Tatsuo, thanks very much for your response. I've tried online
> recovery with the database clients disconnected, but it didn't have
> any effect. The database recovers correctly, works for a while and
> then when a certain update takes place one of the nodes blows up. It's
> not always the same statement but one of them is:
> 2013-08-20 17:04:02 ERROR: pid 30688: pgpool detected difference of
> the number of inserted, updated or deleted tuples. Possible last query
> was: "UPDATE ws_cached_searches SET search_data = $1, cache_time =
> current_timestamp WHERE cached_search_id = $2"
> 2013-08-20 17:04:02 LOG: pid 30688: CommandComplete: Number of
> affected tuples are: 1 0 1
> I've posted my configuration to
> https://github.com/clixtec/redundant-pgpool-config - pgpool and
> PostgreSQL config files in the root, and the four script files in a
> subdirectory. Again it is modelled on the approach described at
> http://zetetic.net/blog/2012/3/9/point-in-time-recovery-from-backup-using-postgresql-continuo.html. Am
> I doing anything fundamentally wrong?
I think there's no instance way to find out the cause of the
problem. I recommend following steps:
1) right after recovery, make sure that the contents of the DBs are
identical. For this you can use pg_dump or some existing tool to
compare database (e.g. export as text file and compare).
2) enable log_per_node_statement and wait until the error happens.
3) examine the log of SQL command issued to each DB node to look for
cause of the problem.
SRA OSS, Inc. Japan
More information about the pgpool-general