Nathan Bossart <nathandbossart@gmail.com> writes:
> On Wed, Apr 08, 2026 at 02:53:20PM -0400, Tom Lane wrote:
>> ... Aside from this problem, the elog's that
>> relation_needs_vacanalyze emits seem 100% inappropriate and misleading
>> when it's being called from the view.
> FWIW that elog() shouldn't be emitting anything from the view, unless
> something is broken.
[ looks again... ] Oh, you mean because pg_stat_get_autovacuum_scores
passes 0 for elevel instead of any of the valid elevel constants.
Can you say "undocumented cowboy hack"? I'd be happier if we added
something like "#define LOG_NEVER 0 /* Never emit this message */"
to elog.h and used that.
>> 2. Add a "bool *should_free" parameter, like we have in tuplestores
>> and some other places. It's on the caller to pfree if should_free
>> gets set, but since we'd have to touch every caller, we'd not miss
>> any.
> This sounds most similar to the "bool *may_free" idea that Andres just
> posted. IIUC the idea is that callers can free the result if they want,
> but they aren't required to do so.
Hmm, yeah I suppose a caller that doesn't care about leakage could
skip the pfree. But are there really any of those? The complaint
that prompted 02502c1bc concerned databases with many many thousands
of relations.
I now realize that what you said upthread about caching the results
might be a bigger problem, ie if the pgstats code does retain all
these values then we'd have a memory bloat problem there. Maybe
we need a more aggressive API change that includes a way to specify
"don't cache this result".
regards, tom lane
