On Fri, Apr 25, 2014 at 4:38 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> maxim.boguk@gmail.com writes:
> > During debug of the weird nagios database alerts about long transactions
> I
> > found that sometime pg_stat_activity contains very weird entries with
> > xact_start NOT NULL and state='idle'...
> > ...
> > What's more: i tested that problem on another databases - and found they
> > have same effect (so it doesn't look like a problem local to that single
> > exact database).
>
> So can you provide a self-contained test case for this? I tried to
> reproduce it here without much success, and I don't see a code path
> other than PREPARE TRANSACTION that could lead to this.
>
> regards, tom lane
>
I going to work on test case over weekend.
Initial quick and dirty attempts to reproduce outside of complicated
production environments provided no results yet.
What I almost sure now that reproducing of the bug requires a mix of very
active connections to the database (with a lot of requests per second) with
semi-idle connections.
And that transient problem affects only these semi-idle connections.
I will provide an update when I get more information.
Kind Regards,
Maksym