The release is 8.2.4. I haven't been able to reproduce the condition yet, but I will send along stack traces as soon as I can. I have this strange feeling that it's only going to happen when I find a reason to make a restart-worthy config change.
"Peter Koczan" <pjkoczan@gmail.com> writes: > [ lots of processes stuck in "notify interrupt" code ]
That's weird. If it's still in that state, or if you can reproduce it, could you attach to a few of those processes with gdb and get stack traces?
Looking at the async.c code, an obvious candidate is that that routine tries to take ExclusiveLock on pg_listener --- so if something had managed to exit without releasing a lock on that table, hangups could be expected. But if that were the case, you'd think the process status lines would include "waiting". My guess is they're blocked on something lower-level than a table lock, but without a stack trace it's hard to guess what.
