On Sat, Mar 16, 2019 at 5:20 PM Tom Lane <tgl@sss.pgh.pa.us> wrote:
>
> Robert Haas <robertmhaas@gmail.com> writes:
> > On Fri, Mar 15, 2019 at 9:50 PM Tom Lane <tgl@sss.pgh.pa.us> wrote:
> >> pg_stat_statements has a notion of query ID, but that notion might be
> >> quite inappropriate for other usages, which is why it's an extension
> >> and not core.
>
> > Having written an extension that also wanted a query ID, I disagree
> > with this position.
>
> [ shrug... ] The fact remains that pg_stat_statements's definition is
> pretty lame. There's a lot of judgment calls in which query fields
> it chooses to examine or ignore, and there's been no attempt at all
> to make the ID PG-version-independent, and I rather doubt that it's
> platform-independent either. Nor will the IDs survive a dump/reload
> even on the same server, since object OIDs will likely change.
>
> These things are OK, or at least mostly tolerable, for pg_stat_statements'
> usage ... but I don't think it's a good idea to have the core code
> dictating that definition to all extensions. Right now, if you have
> an extension that needs some other query-ID definition, you can do it,
> you just can't run that extension alongside pg_stat_statements.
> But you'll be out of luck if the core code starts filling that field.
>
> I'd be happier about having the core code compute a query ID if we
> had a definition that was not so obviously slapped together.
But the queryId itself is stored in core. Exposing it in
pg_stat_activity or log_line_prefix would still allow users to choose
the implementation of their choice, or none. That seems like a
different complaint from asking pgss integration in core to have all
its metrics available by default (or at least without a restart).
Maybe we could add a GUC for pg_stat_statements to choose whether it
should set the queryid itself and not, if anyone wants to have its
metrics but with different queryid semantics?