Martijn van Oosterhout <kleptog@svana.org> writes:
> My guess is that the total allocated VM is fairly close to the limit of
> your configuration and that a slightly higher than normal load and a
> background autovacuum took it over the edge. My guess is that if you up
> the swap to 4GB or perhaps 8GB, you'll get a lot more room to manouver.
The bits about
> 2008-03-10 18:43:33 EDT LOG: could not fork new process for connection:
> Cannot allocate memory
certainly suggest that the problem was a global out-of-memory condition
and not just within any one process. I concur with raising swap, but
also with the upthread suggestions to cut down the number of concurrent
connections.
The segfaults (sig11s) are a bit disturbing too --- what that probably
indicates is someplace using malloc() and failing to test for failure,
neither of which is a good thing. Did you by any chance get core dumps
from those? A stack trace would be mighty useful.
regards, tom lane