Eric Rousse <eric.rousse@telmatik.com> writes:
> ...
> 2006-11-16 04:00:39 [8763] LOG: connection received: host=10.1.1.54
> port=4894
> 2006-11-16 04:00:40 [8763] LOG: pq_recvbuf: unexpected EOF on client
> connection
> 2006-11-16 04:00:40 [8763] LOG: incomplete startup packet
> 2006-11-16 04:02:26 [2534] LOG: database system was interrupted at
> 2006-11-16 03:57:36 EST
> 2006-11-16 04:02:26 [2534] LOG: checkpoint record is at C/6733EB68
> ...
I think what you're seeing here is probably a kernel-level crash and
system reboot. It's not any normal sort of Postgres problem, because
if it were you'd see the postmaster bleating about crash of one of its
child processes. Here it appears that the postmaster and all its
children died at once leaving no messages behind --- and that just
doesn't happen without either manual intervention or a system crash.
If it seems to be triggered by running a PG backup, it could be that
you've got a disk hardware problem that only manifests when you try to
read a particular data block :-(. Have you tried running "badblocks"?
regards, tom lane