Ciprian Grigoras wrote:
> Has anyone seen a similar issue (maybe a bug in the 9.0 version ?) where
> simple statements don't finish and get locked there and can't be killed by
> the pg_terminate_backed ? What is the cause of this ?
Interesting. I wonder if you just wrapped around pg_multixact and the
backend is stuck in the loop there. I have never heard of anyone with
this problem, and wraparound is supposed to work fine in 9.0. (As I
recall that code is there since 8.1 so it'd be very surprising that it
would have bugs there and not notice all this time).
Since you've been evidently running with this for years and never seen
this problem, I would be surprised if you ever see it again. Still, if
it does happen, please do grab a stack trace with GDB on the blocked
processes. Make sure to have debug symbols.
--
Álvaro Herrera http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services