Horton

Re: forgotten initalization of a variable

Michael Paquier <michael@paquier.xyz>

View thread View on postgr.es
From: Michael Paquier <michael@paquier.xyz>
To: Kyotaro Horiguchi <horikyota.ntt@gmail.com>
Cc: pgsql-hackers@lists.postgresql.org
Date: 2020-04-21T08:34:26Z
Lists: pgsql-hackers

Attachments

On Tue, Apr 21, 2020 at 03:08:30PM +0900, Kyotaro Horiguchi wrote:
> The commit a7e8ece41c adds a new member restoreCommand to
> XLogPageReadPrivate. readOneRecord doesn't make use of it but forgets
> to set NULL. That can lead to illegal pointer access.

That's an oversight of the original commit.  Now, instead of failing
even if there is a restore command, wouldn't it be better to pass down
the restore_command to readOneRecord() so as we can optionally
improve the stability of a single record lookup?  This only applies to
a checkpoint record now, but this routine could be called elsewhere in
the future.  Please see the attached.
--
Michael

Commits

  1. Fix single-record reads to use restore_command if available in pg_rewind

  2. Add -c/--restore-target-wal to pg_rewind