Re: [PATCH] Avoid open and lock the table Extendend Statistics (src/backend/commands/statscmds.c)

From: Tomas Vondra <tomas(dot)vondra(at)enterprisedb(dot)com>
To: Andres Freund <andres(at)anarazel(dot)de>, Ranier Vilela <ranier(dot)vf(at)gmail(dot)com>, Tomas Vondra <tv(at)fuzzy(dot)cz>
Cc: Pg Hackers <pgsql-hackers(at)postgresql(dot)org>
Subject: Re: [PATCH] Avoid open and lock the table Extendend Statistics (src/backend/commands/statscmds.c)
Date: 2022-02-13 22:59:13
Message-ID: 79817e57-d670-8eb4-b3bc-0683f2688adc@enterprisedb.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

On 2/13/22 22:43, Andres Freund wrote:
> Hi,
>
> On 2022-02-13 18:21:38 -0300, Ranier Vilela wrote:
>> Why open and lock the table Extended Statistics if it is not the owner.
>> Check and return to avoid this.
>
> I was about to say that this opens up time-to-check-time-to-use type
> issues. But it's the wrong thing to lock to prevent those.
>

I doubt optimizing this is worth any effort - ALTER STATISTICS is rare,
not doing it with owner rights is even rarer.

> Having looked briefly at it, I don't understand what the locking scheme is?
> Shouldn't there be at least some locking against concurrent ALTERs and between
> ALTER and ANALYZE etc?
>

I don't recall the details of the discussion (if at all), but if you try
to do concurrent ALTER STATISTICS, that'll end up with:

ERROR: tuple concurrently updated

reported by CatalogTupleUpdate. AFAIK that's what we do for other object
types that don't have a relation that we might lock (e.g. try to co
CREATE OR REPLACE FUNCTION).

ANALYZE reads the statistics from the catalog, so it should see the last
committed stattarget value, I think.

regards

--
Tomas Vondra
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Andres Freund 2022-02-13 23:22:49 bailing out in tap tests nearly always a bad idea
Previous Message Chapman Flack 2022-02-13 22:57:24 Re: Mark all GUC variable as PGDLLIMPORT