On Sunday March 26 2006 9:16 am, Tom Lane wrote:
> "Ed L." <pgsql@bluepolka.net> writes:
> > I see what appear to be many single transactions holding
> > RowExclusiveLocks for sometimes 40-50 seconds while their
> > query shows "<IDLE> in transaction".
> > ...
> > I'm thinking that means the client is simply tweaking a row
> > and then failing to commit the change for 40-50 seconds. Is
> > that consistent?
>
> That's what it sounds like to me. You might consider logging
> all commands from these clients for awhile so you can check
> that theory.
This indeed appears to be locking problem from within
Apache::Session where it deletes a row from the DB but fails to
commit the change for an extended period while another
transaction waits on the same row. Not sure how/why that's
happening in the client code, but it's not a DB issue.
Thanks,
Ed