Tom Lane wrote:
Brian Staszewski <brian.staszewski@xxxxxxxxxxx> writes:
WARNING: oldest xmin is far in the past
HINT: Close open transactions soon to avoid wraparound problems.
Hmm ... which PG version is this? I think 8.2 will start to complain
about this much earlier than older versions (~ 100M transactions by
default, instead of 1000M). Though you've probably got a problem
here in any case.
Yep, 8.2.4.
What's been done so far:
-Cluster-wide vacuum (these are done nightly, I ran one again just for gp)
Not relevant
-Ran the following query looking for old/suspicious
connections/transactions, didn't find anything useful:
select procpid, datname, client_addr, usename, current_query,
query_start from pg_stat_activity order by query_start asc;
Most likely the problem transaction is sitting idle and hence has null
query_start. Try "ps auxww" or local equivalent and look for postgres
processes that are very old and are in "idle in transaction" state.
regards, tom lane
"ps auxww | grep trans " did in fact reveal three processes that were
2, 3, and 4 days old that the query above didn't show. The connections
were from three different application servers and restarting the pools
on those servers corrected the problem.
Thanks,
Brian
|
begin:vcard
fn:Brian Staszewski
n:Staszewski;Brian
org:Farheap Solutions, Inc.;Information Technologies
email;internet:brian.staszewski@xxxxxxxxxxx
title:Systems Administrator
tel;work:(949) 417-1500 x948278
x-mozilla-html:TRUE
version:2.1
end:vcard
---------------------------(end of broadcast)---------------------------
TIP 6: explain analyze is your friend
