Re: [HACKERS] Re: INSERT/UPDATE waiting (another example)

Wayne Piekarski <wayne@senet.com.au> writes:
> We are experiencing exactly the same problem as above - during the day,
> all of a sudden Postgres will completely jam up, with all processing in
> one of the following states: (from ps -axwwwwww)

It seems possible that the hashtable bugs I fixed a couple months ago
are rising up to bite you.  (Basically, the shared hashtables that
contain things like locks and buffers would go nuts if there got to be
more than 256 entries ... and it sure sounds like your installation is
big enough that it could have, eg, more than 256 active locks when
under load.)  One quick thing you might try to test this is to reduce
the postmaster's -B setting to less than 256 (if you have it set that
high) and see if stability improves.

These bugs are fixed in 6.5-beta1, but it has enough other bugs that
I don't think Wayne would be wise to try moving to 6.5 just yet.
I have a patch for 6.4.2 that I believe also fixes the problems, but
it hasn't gotten quite as much testing as I would like so I haven't
committed it into the REL6_4 tree.  (There's not going to be a 6.4.3
release, according to current plans, so it's hardly worth doing anyway.)

What I will do is send the patch to Wayne in a separate message, and
also cc: it to the PATCHES list --- anyone else who needs it can get it
from there.  Please let us know if this helps, Wayne.

			regards, tom lane