Re: Forcibly vacating locks

On Fri, 2008-06-20 at 18:38 -0400, Bruce Momjian wrote:
> Laurent Birtz wrote:
> > > No. The closest thing we have is log_lock_waits in 8.3. I wonder if
> > > you could hack up something to monitor the server logs for such messages
> > > and cancel the queries.
> >
> > Assuming I can monitor the logs in this way, how would I cancel the
> > queries (or lack thereof, in the case of a client that sits doing nothing
> > with a held lock)?
>
> Use log_line_prefix to get the process id in the log line, then use
> pg_cancel_backend() on the process id.
>
> > >> 2) Is there any hostility about the notion of implementing this feature
> > >> into Postgres?
> > >
> > > Probabably --- it seems like a narrow use case.
> >
> > I'll consider this to be the definite answer unless I hear a dissenting
> > opinion in the next few days.
>
> Yea, I might be wrong.

I think you're very right. Don't ever want to see locks disappearing
because of a timeout, sorry Laurent.

--
Simon Riggs www.2ndQuadrant.com
PostgreSQL Training, Services and Support