On Thu, Nov 14, 2013 at 9:09 AM, Fujii Masao <masao.fujii@gmail.com> wrote:
>> I think that pg_stat_statements should be made to do this kind of
>> thing by a third party tool that aggregates snapshots of deltas.
>> Time-series data, including (approximate) *local* minima and maxima
>> should be built from that. I think tools like KONDO-san's pg_statsinfo
>> tool have an important role to play here. I would like to see it or a
>> similar tool become a kind of defacto standard for consuming
>> pg_stat_statements' output.
> Agreed.
I would like to hear others' thoughts on how to proceed here. Has
anyone actually tried and failed with an approach that uses aggressive
aggregation of snapshots/deltas? If that's something that is
problematic, let's hear why. Maybe pg_stat_statements could do better
when snapshots are aggressively taken by tools at fixed intervals.
Most obviously, we could probably come up with a way better interface
for tools that need deltas only, where a consumer registers interest
in receiving snapshots. We could store a little bitmap structure in
shared memory, and set bits as corresponding pg_stat_statements
entries are dirtied, and only send query texts the first time.
I am still very much of the opinion that we need to expose queryid and
so on. I lost that argument several times, but it seems like there is
strong demand from it from many places - I've had people that I don't
know talk to me about it at conferences.
--
Peter Geoghegan