Horton

Re: Recovering from detoast-related catcache invalidations

Tom Lane <tgl@sss.pgh.pa.us>

View thread
From: Tom Lane <tgl@sss.pgh.pa.us>
To: Xiaoran Wang <fanfuxiaoran@gmail.com>
Cc: pgsql-hackers@lists.postgresql.org, Andres Freund <andres@anarazel.de>
Date: 2024-01-13T19:12:23Z
Lists: pgsql-hackers

Commits

Same data as JSON: GET /api/v1/messages/:b64id/commits the thread's linked commits as JSON, with link sources. API reference →

  1. Fix catcache invalidation of a list entry that's being built

  2. Cope with inplace update making catcache stale during TOAST fetch.

  3. Add previous commit to .git-blame-ignore-revs.

  4. Re-pgindent catcache.c after previous commit.

  5. Cope with catcache entries becoming stale during detoasting.

I wrote:
> Xiaoran Wang <fanfuxiaoran@gmail.com> writes:
>> Hmm, how about first checking if any invalidated shared messages have been
>> accepted, then rechecking the tuple's visibility?
>> If there is no invalidated shared message accepted during
>> 'toast_flatten_tuple',
>> there is no need to do then visibility check, then it can save several
>> CPU cycles.

> Meh, I'd just as soon not add the additional dependency/risk of bugs.
> This is an expensive and seldom-taken code path, so I don't think
> shaving a few cycles is really important.

It occurred to me that this idea might be more interesting if we
could encapsulate it right into systable_recheck_tuple: something
like having systable_beginscan capture the current
SharedInvalidMessageCounter and save it in the SysScanDesc struct,
then compare in systable_recheck_tuple to possibly short-circuit
that work.  This'd eliminate one of the main bug hazards in the
idea, namely that you might capture SharedInvalidMessageCounter too
late, after something's already happened.  However, the whole idea
only works for catalogs that have catcaches, and the other users of
systable_recheck_tuple are interested in pg_depend which doesn't.
So that put a damper on my enthusiasm for the idea.

			regards, tom lane