Re: Vacuumlo improvements

From: Nathan Bossart <nathandbossart(at)gmail(dot)com>
To: Shawn McCoy <shawn(dot)the(dot)mccoy(at)gmail(dot)com>
Cc: pgsql-hackers(at)postgresql(dot)org
Subject: Re: Vacuumlo improvements
Date: 2026-05-13 03:00:15
Message-ID: agPpPyLs-61qFw88@nathan
Views: Whole Thread | Raw Message | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

On Tue, May 12, 2026 at 11:34:10AM -0600, Shawn McCoy wrote:
> Ideally, vacuumlo could be improved to:
> - Resolve domain types back to their base types when scanning columns
> (using pg_type.typbasetype), or
> - At least emit a WARNING when it encounters columns with domains over
> oid/lo that it is skipping, so the user is aware.

Commit 64c604898e added the note about domains to the docs. Unfortunately,
neither that nor the corresponding thread [0] offer any clues as to why
vacuumlo doesn't resolve domains. The commit history for vacuumlo has been
pretty quiet for a long time, so maybe it's just been overlooked.

> At minimum, I can submit a documentation improvement to make the
> data-loss risk more prominent. The current parenthetical note is easy
> to miss.

Improving the documentation seems reasonable, too. Another thing we could
explore is allowing users to specify which tables/columns refer to LOs,
perhaps with a user-provided query. One wrinkle is that dblink allows
specifying multiple databases, and presumably each database will be a
little different.

Separately, do you know whether users are using lo_manage() at all? And if
not, why?

[0] https://postgr.es/m/BAY164-W265A089BD32F8901A686C9FF430%40phx.gbl

--
nathan

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Tatsuo Ishii 2026-05-13 03:11:14 Re: Row pattern recognition
Previous Message Fujii Masao 2026-05-13 02:49:25 Re: Call EndCopyFrom() after initial table sync in logical replication