Re: IDEA: pg_stat_statements tracking utility statements by tag?

On Wed, Jul 29, 2020 at 2:42 PM Fujii Masao <masao(dot)fujii(at)oss(dot)nttdata(dot)com> wrote:
>
> On 2020/07/29 18:24, Martijn van Oosterhout wrote:
> > Hoi hackers,
> >
> > We've been using the pg_stat_statements extension to get an idea of the queries used in the database, but the table is being filled with entries like:
> >
> > SAVEPOINT sa_savepoint_NNN;
> > RELEASE SAVEPOINT sa_savepoint_NNN;
> > DECLARE "c_7f9451c4dcc0_5" CURSOR WITHOUT HOLD ...
> > FETCH FORWARD 250 FROM "c_7f9451b03908_5"
> >
> > Since the unique id is different for each query, the aggregation does nothing and there are quite a lot of these drowning out the normal queries (yes, I'm aware this is an issue of itself). The only way to deal with this is "pg_stat_statements.track_utility=off". However, it occurs to me that if you just tracked the tags rather than the full query text you could at least track the number of such queries and how much time they take. So the above queries would be tracked under SAVEPOINT, RELEASE, DECLARE CURSOR and (I guess) FETCH respectively. But it would also catch DDL.
> >
> > Does this sound like something for which a patch would be accepted?
>
> Or, we should extend the existing query normalization to handle also DDL?

+1, introducing DDL normalization seems like a better way to go in the
long run. Defining what should and shouldn't be normalized can be
tricky though.