The Hermit Hacker <scrappy@hub.org> writes:
> you still have your account on that machine if you want to take a quick
> look around ... else, anything else I should be looking at?
I poked around and couldn't learn much of anything --- the logfiles from
yesterday are already gone, apparently. I did find some interesting
entries in today's logfiles:
%grep Lru *
5432.61820:FATAL 1: ReleaseLruFile: No open files available to be closed
postmaster.5437.36290:FATAL 1: ReleaseLruFile: No open files available to be closed
postmaster.5437.62218:FATAL 1: ReleaseLruFile: No open files available to be closed
What we see here are backends choking because there are no free kernel
file descriptor slots, even after they've closed *all* of their own
discretionary FDs. So you've definitely got a serious problem with
insufficient FD slots. Time to tweak those kernel parameters.
I still don't see a linkage between too few FDs and the stuck-spinlock
crashes, but maybe there is one.
regards, tom lane