MIME-Version: 1.0
References: <CAKygsHQHhd4rzYcNCd8OJMH46saT1E+9+5MT2CgOCGgf_yaQAw@mail.gmail.com>
 <CAECtzeUQ3hJYPM6k6_XqZLq=DygHbQ3+5YQLrv-ngcht2rQjQg@mail.gmail.com>
In-Reply-To: <CAECtzeUQ3hJYPM6k6_XqZLq=DygHbQ3+5YQLrv-ngcht2rQjQg@mail.gmail.com>
From: PetSerAl <petseral@gmail.com>
Date: Mon, 20 May 2024 13:36:08 +0300
Message-ID: <CAKygsHRBRUhjGAQtqnbYPaXxJPvN2J49yS4ceee62iVG_jcuQQ@mail.gmail.com>
Subject: Re: pg_dump and not MVCC-safe commands
To: Guillaume Lelarge <guillaume@lelarge.info>
Cc: pgsql-general@lists.postgresql.org
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Archived-At: <https://www.postgresql.org/message-id/CAKygsHRBRUhjGAQtqnbYPaXxJPvN2J49yS4ceee62iVG_jcuQQ%40mail.gmail.com>
Precedence: bulk

My question:
What happens if not MVCC-safe command committed after snapshot but before l=
ock?

On Mon, May 20, 2024 at 12:33=E2=80=AFPM Guillaume Lelarge
<guillaume@lelarge.info> wrote:
>
> Hi,
>
> Le lun. 20 mai 2024 =C3=A0 11:27, PetSerAl <petseral@gmail.com> a =C3=A9c=
rit :
>>
>> How pg_dump interact with not MVCC-safe commands?
>>
>> As I understand, pg_dump first take snapshot and then lock all tables
>> it intended to dump. What happens if not MVCC-safe command committed
>> after snapshot but before lock? From comment to pg_dump.c I understand
>> that it may fail with 'cache lookup failed' error. But, can it happen,
>> that pg_dump not fail, but instead capture inconsistent dump? For
>> example TRUNCATE committed after snapshot and pg_dump will see result
>> of TRUNCATE but not result of other commands in TRUNCATE transaction?
>>
>>
>
> You can't truncate an already existing table while pg_dump is running. TR=
UNCATE needs an exclusive lock, and pg_dump already has a lock on all table=
s of the database it's dumping. So TRUNCATE will be blocked until pg_dump f=
inishes all its work.
>
> (The same will happen for VACUUM FULL, CLUSTER and some (all?) ALTER TABL=
E commands.)
>
>
> --
> Guillaume.