Thomas Munro <munro@ip9.org> writes:
> My theory is that if two connections accessed by different threads get
> shut down around the same time, there is a race scenario where each of
> them fails to write to its socket, sees errno == EPIPE and then sees a
> pending SIGPIPE with sigpending(), but only one thread returns from
> sigwait() due to signal merging.
Hm, that does sound like it could be a problem, if the platform fails
to track pending SIGPIPE on a per-thread basis.
> We never saw the problem again after we made the following change:
> ...
> Does this make any sense?
I don't think that patch would fix the problem if it's real. It would
prevent libpq from hanging up when it's trying to throw away a pending
SIGPIPE, but the fundamental issue is that that action could cause a
SIGPIPE that's meant for some other thread to get lost; and that other
thread isn't necessarily doing anything with libpq.
I don't claim to be an expert on this stuff, but I had the idea that
multithreaded environments were supposed to track signal state per-thread
not just per-process, precisely because of issues like this.
regards, tom lane