Michael Paquier <michael(dot)paquier(at)gmail(dot)com> writes:
> Doing a join on pg_stat_activity and pg_locks is not going to help
> much as you could only get the moment when query has started or its
> state has changed. Have you thought about the addition of a new column
> in pg_locks containing the timestamp of the moment a lock has been
> taken? I am sure that we are concerned about the performance impact
> that extra calls to gettimeofday could have though...
In theory this could be driven off the same gettimeofday needed to
start the deadlock_timeout timer. Not sure how messy that'd be.
regards, tom lane