David Jericho <david.jericho@bytecomm.com.au> writes:
> I have a Solaris 8 machine with all recommended patches which I'm attempting
> to run 7.3.2 on without much success.
> The source compiled fine using gcc 3.2.3, and it seems to run up okay.
> I can create a number of databases on the machine totalling around a
> million rows, at 300MB.
> It will then run fine for anywhere between 5 minutes, and about an
> hour, and then logged on the console is a message telling me the
> database is doing a smart shutdown.
The only way that can happen is if something sends the postmaster a
SIGTERM signal. Sleuthing down why the signal is happening is not
really a Postgres issue, but here are two theories:
1. You are launching the postmaster manually from a shell, and not
disengaging it from the shell (via nohup and/or redirecting all three
of its stdin,stdout,stderr away from /dev/tty). When you log out of
that shell, it tries to SIGTERM all its remaining descendants including
the postmaster.
2. You are running with "ulimit"-type resource consumption limits,
and after the postmaster has used up X CPU seconds or X file write
operations or something like that, it gets SIGTERM'd. I'm not real
strong on this one, because on most Unixen that have ulimit, there
are other signals that get delivered for resource limits. But I
dunno how Solaris does it exactly.
Anyway, look for reasons why something in your environment would try to
SIGTERM the postmaster.
regards, tom lane