On Thu, Jul 18, 2024 at 01:37:40PM +0200, Anthonin Bonnefoy wrote:
> For utility statements defined within a function, the queryTree is
> copied to a plannedStmt as utility commands don't require planning.
> However, the queryId is not propagated to the plannedStmt. This leads
> to plugins relying on queryId like pg_stat_statements to not be able
> to track utility statements within function calls.
You are right, good catch. This leads to only partial information
being reported depending on the setting of pg_stat_statements.track.
It is a point of detail, but I'd rather expand a bit more the tests on
top of what you are proposing:
- Upper and down-casing for non-top utility commands, to check that
they are counted consistently.
- Check with pg_stat_statements.track = 'top'
- Not cross-checking pg_stat_statements.track_utility = false is OK.
While this qualifies as something that could go down to all the stable
branches, it is much easier to think about utility statements in 16~
now that we compile the query IDs depending on their parsed tree, so
will apply down to that.
pg_stat_statements tests have also been refactored in 16~, but that's
a nit here..
--
Michael