Hi,
On 2019-01-21 18:14:31 -0500, Tom Lane wrote:
> Andres Freund <andres@anarazel.de> writes:
> > Given https://www.postgresql.org/message-id/20190121193300.gknn7p4pmmjg7nqf%40alap3.anarazel.de
> > and the concerns voiced in the thread quoted therein, I'm a bit
> > surprised that you just went ahead with this, and backpatched it to boot.
>
> I don't think that's relevant. The issues there were about whether
> a pg_index row update ought to cause an invalidation of the relcache
> entry for the index's table (not the one for the index, which it
> already takes care of). That seems very questionable to me --- the
> potentially-invalidatable info ought to be in the index's relcache entry,
> not its parent table's entry, IMO.
Well, we've plenty of information about indexes in the table's
relcache. Among other things, the list of indexes, bitmaps of indexed
attributes, which index is the primary key, etc is all maintained
there... So I don't really see a material difference between the
constraint and the index case. You can make some arguments about
superfluous invals, true. I don't see why rd_indexlist et al is
materially different from rd_fkeylist.
Greetings,
Andres Freund