On Wed, 2002-05-29 at 13:55, Tom Lane wrote:
> Stephen Robert Norris <srn@commsecure.com.au> writes:
> > Not at all like the vacuum problem. Do you have any other tests I can
> > run? Other diagnostics that might help?
>
> This is a long shot, but ... does the behavior change if you remove the
> two lines
>
> if (signo != SIGALRM)
> act.sa_flags |= SA_RESTART;
>
> near the bottom of src/backend/libpq/pqsignal.c ? I am wondering if
> Linux gets unhappy if we try to do much in a signal handler. This
> looks to be the simplest alteration that might address such a problem.
>
> regards, tom lane
Maybe it does help. I've only seen the problem once now (just after
startup) rather than every few minutes... I'm still seeing lots of
processes in the ready queue every now and then, but they are doing more
sensible things.
I'll investigate some more and be more certain.
Stephen