Re: type cache cleanup improvements - Mailing list pgsql-hackers

From Artur Zakirov
Subject Re: type cache cleanup improvements
Date
Msg-id CAKNkYnxaok+EKoGvq0oNQ+jJP6fAphDNEcYxri1JNqPKdMiBhg@mail.gmail.com
Whole thread Raw
In response to Re: type cache cleanup improvements  (Teodor Sigaev <teodor@sigaev.ru>)
List pgsql-hackers
On Tue, 15 Oct 2024 at 10:09, Alexander Korotkov <aekorotkov@gmail.com> wrote:
> >     /* Call check_delete_rel_type_cache() if we actually cleared something */
> >     if (hadTupDescOrOpclass)
> >         delete_rel_type_cache_if_needed(typentry);
> >
> >         /*
> >          * Call check_delete_rel_type_cache() if we cleaned
> >          * TCFLAGS_HAVE_PG_TYPE_DATA flag previously.
> >          */
> >         if (hadPgTypeData)
> >             delete_rel_type_cache_if_needed(typentry);
> >
> > check_delete_rel_type_cache don't exist, so these comments are wrong?
>
> Yep, they didn't get updated.  Fixed in the attached patchset.

Thank you Alexander for the fixes. The last version of the patch looks
good to me.

> I'm not sure I get the point.  This check ensures that type entry has
> something to be cleared.  In this case we need to keep
> RelIdToTypeIdCacheHash entry to find this item on invalidation
> message.  I'm not sure how TCFLAGS_DOMAIN_BASE_IS_COMPOSITE is
> relevant here, because it's valid only for TYPTYPE_DOMAIN while this
> patch deals with TYPTYPE_COMPOSITE.

Regarding this discussion earlier, I assumed that TYPTYPE_DOMAIN also
needs to be handled by `insert_rel_type_cache_if_needed()`. And it
seems that handling of TYPTYPE_DOMAIN will remain the same as before.

-- 
Kind regards,
Artur



pgsql-hackers by date:

Previous
From: Jingtang Zhang
Date:
Subject: Re: Use WALReadFromBuffers in more places
Next
From: Tender Wang
Date:
Subject: Re: Improve node type forward reference