Incorrect relation locked at beginning of REINDEX CONCURRENTLY

Michael Paquier <michael@paquier.xyz>

From: Michael Paquier <michael@paquier.xyz>
To: Postgres hackers <pgsql-hackers@lists.postgresql.org>
Date: 2019-10-21T07:43:23Z
Lists: pgsql-hackers

Attachments

Hi all,

While digging into the issues reported lately about REINDEX
CONCURRENTLY, I have bumped into the following, independent, issue:
/* Now open the relation of the new index, a lock is also needed on it */
newIndexRel = index_open(indexId, ShareUpdateExclusiveLock)

In this code path, indexId is the OID od the old index copied, and
newIndexId is the OID of the new index created.  So that's clearly
incorrect, and the comment even says the intention.  This causes for
example the same session lock to be taken twice on the old index, with
the new index remaining unprotected.

Any objections if I fix this issue as per the attached?
--
Michael

Commits

  1. Acquire properly session-level lock on new index in REINDEX CONCURRENTLY