Andres Freund <andres@anarazel.de> writes:
> On 2017-07-18 09:42:31 -0400, Tom Lane wrote:
>> I wonder if a better answer wouldn't be to reduce PGSTAT_STAT_INTERVAL.
> Not sure if that really does that much to solve the concern.
Well, it reduces the amount of data churn that a statement shorter than
PGSTAT_STAT_INTERVAL could cause.
> Another,
> pretty half-baked, approach would be to add a procsignal triggering idle
> backends to send stats, and send that to all idle backends when querying
> stats. We could even publish the number of outstanding stats updates in
> PGXACT or such, without any locking, and send it only to those that have
> outstanding ones.
If somebody wanted to do the work, that'd be a viable answer IMO. You'd
really want to not wake backends that have nothing more to send, but
I agree that it'd be possible to advertise that in shared memory.
regards, tom lane
