Re: pg_stat_activity.waiting_start

Tom,

* Tom Lane (tgl(at)sss(dot)pgh(dot)pa(dot)us) wrote:
> In practice, there should never be waits on LWLocks (much less spinlocks)
> that exceed order-of-milliseconds; if there are, either we chose the wrong
> lock type or the system is pretty broken in general. So maybe it's
> sufficient if we provide a wait start time for heavyweight locks ...
> though that still seems kind of ugly.

While I agree that it's a bit ugly, if the alternative is "don't have
anything", then I'd argue that it's worth it. The use-case for this
feature, as I see it, is for admins to be able to go look at how long
something has been waiting and monitoring scripts to which fire only
every minute or more, and order-of-milliseconds differences aren't
significant there.

It's terribly ugly, but from a practical standpoint, we could probably
make it "waiting after deadlock timeout" and just set the time when the
deadlock timeout fires and the use-case for this would be satisfied.

Thanks!

Stephen