Horton

Re: Revisiting {CREATE INDEX, REINDEX} CONCURRENTLY improvements

Mikhail Nikalayeu <mihailnikalayeu@gmail.com>

View thread
From: Mihail Nikalayeu <mihailnikalayeu@gmail.com>
To: Antonin Houska <ah@cybertec.at>
Cc: Hannu Krosing <hannuk@google.com>, Matthias van de Meent <boekewurm+postgres@gmail.com>, Sergey Sargsyan <sergey.sargsyan.2001@gmail.com>, alvherre@kurilemu.de, Andres Freund <andres@anarazel.de>, Michael Paquier <michael@paquier.xyz>, PostgreSQL Hackers <pgsql-hackers@postgresql.org>, Andrey Borodin <amborodin86@gmail.com>, Melanie Plageman <melanieplageman@gmail.com>
Date: 2025-12-01T10:49:37Z
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. Revert changes to CONCURRENTLY that "sped up" Xmin advance

  2. VACUUM: ignore indexing operations with CONCURRENTLY

  3. Avoid spurious waits in concurrent indexing

Hello, Antonin!

On Mon, Dec 1, 2025 at 11:29 AM Antonin Houska <ah@cybertec.at> wrote:
> With logical replication, we cannot really use multiple snapshots as Mihail is
> proposing elsewhere in the thread, because the logical decoding system only
> generates the snapshot for non-catalog tables once (LR uses that snapshot for
> the initial table synchronization). Only snapshots for system catalog tables
> are then built as the WAL decoding progresses. It can be worked around by
> considering regular table as catalog during the processing, but it currently
> introduces quite some overhead:

My idea related to REPACK is a little bit different. I am not talking
about snapshots generated by LR - just GetLatestSnapshot.

> The core problem here is that the snapshot you need for the first pass
> restricts VACUUM on all tables in the database

We might use it only for a few seconds - it is required only to
*start* the scan (to ensure we will not miss anything in the table).
After we may throw it away and ask GetLatestSnapshot a fresh one for
next N pages. We just need to synchronize scan position in the table
and logical decoding.

The same is possible for CIC too. In that case we should do the same
and just store all incoming tuples the same way as STIR does it.

Best regards,
Mikhail.