Ben Chobot <bench@silentmedia.com> writes:
> On Fri, 15 Jan 2010, Tom Lane wrote:
>> OK, so it's trying to get an exclusive page-level lock on some page of
>> the index. The only thing that could block that for any long period is
>> if some other process is sitting with an open indexscan. Look around
>> for processes that have been "idle in transaction" for a long time ...
> There are some, but they've all been running less than a second. There's
> no better place to look than pg_stat_activity, is there?
Hmph. Given that, what we seem to be looking at is a leaked buffer pin
count. There isn't any easy way to fix that except a database restart
--- perhaps you can manage one over the weekend?
I don't recall having heard of any such bugs lately, but 8.1.x is kind
of an old branch. I'm still wondering if slony might've been involved
somehow.
regards, tom lane