Tom Lane wrote:
> If that's the right diagnosis, we probably ought to try to fix DROP
> INDEX to not do that, since it could deadlock against any number of
> things not just autovac. But there's another question here: can we
> make autovac lose the deadlock war, instead of the user process?
Hmm,
It says here that CVS HEAD has a deadlock risk:
http://www.pgbuildfarm.org/cgi-bin/show_log.pl?nm=chinchilla&dt=2008-06-09%2008:16:01
The problem manifested while attempting to drop a GIN index (though
I doubt GIN is specially at fault):
[484ceb40.47da:169] ERROR: deadlock detected
[484ceb40.47da