Horton

Re: Preserve attstattarget on REINDEX CONCURRENTLY

Michael Paquier <michael@paquier.xyz>

View thread View on postgr.es
From: Michael Paquier <michael@paquier.xyz>
To: Ronan Dunklau <ronan@dunklau.fr>
Cc: Tomas Vondra <tomas.vondra@enterprisedb.com>, pgsql-hackers@postgresql.org
Date: 2021-02-05T02:17:48Z
Lists: pgsql-hackers

Attachments

On Thu, Feb 04, 2021 at 03:52:44PM +0100, Ronan Dunklau wrote:
>> Hmmm, that sure seems like a bug, or at least unexpected behavior (that
>> I don't see mentioned in the docs).

Yeah, per the rule of consistency, this classifies as a bug to me.

> Please find attached a correct patch.

ConstructTupleDescriptor() does not matter much, but this patch is not
acceptable to me as it touches the area of the index creation while
statistics on an index expression can only be changed with a special
flavor of ALTER INDEX with column numbers.  This would imply an ABI
breakage, so it cannot be backpatched as-is.

Let's copy this data in index_concurrently_swap() instead.  The
attached patch does that, and adds a test cheaper than what was
proposed.  There is a minor release planned for next week, so I may be
better to wait after that so as we have enough time to agree on a
solution.
--
Michael

Commits

  1. Fix ORDER BY clause in new regression test of REINDEX CONCURRENTLY

  2. Preserve pg_attribute.attstattarget across REINDEX CONCURRENTLY