Tom Lane wrote:
> Brian Staszewski <brian(dot)staszewski(at)farheap(dot)com> 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