Re: Very long execution time of "select nextval('..');"

mljv(at)planwerk6(dot)de writes:
> we run postgresql-8.1 on a dedicated debian box 64bit, dual-core CPU, 8GB RAM,
> RAID-1.

8.1.what?

> LOG: duration: 12636.746 ms statement: EXECUTE <unnamed> [PREPARE: select
> nextval ('member_id_seq')]

That's just bizarre, especially if your system isn't showing any other
signs of stress.

> Unfortunatly i can not tell at which time this happens as the log doesn't
> show the time of day.

See log_line_prefix. I think what you need to do is gather some
evidence about what else is happening at the same time --- can you
afford to enable log_statement = all? Also, you should try to correlate
this with spikes in I/O demand (try running "vmstat 1" or similar).

It could be that this is related to checkpointing, which you won't see
in a log_statement trace. In 8.1 you'd have to crank up
log_min_messages to DEBUG2 to get log entries for checkpoint start and
end, which is going to result in a mighty verbose log, but you may have
to do that to confirm or disprove the idea.

regards, tom lane