Making index_set_state_flags() transactional

From: Michael Paquier <michael(at)paquier(dot)xyz>
To: Postgres hackers <pgsql-hackers(at)lists(dot)postgresql(dot)org>
Cc: Rahila Syed <rahila(dot)syed(at)2ndquadrant(dot)com>
Subject: Making index_set_state_flags() transactional
Date: 2020-09-03 08:04:40
Views: Raw Message | Whole Thread | Download mbox | Resend email
Lists: pgsql-hackers

Hi all,

For a couple of things I looked at lately, it would be really useful
to make index_state_set_flags() transactional and replace its use of
heap_inplace_update() by CatalogTupleUpdate():
- When dropping an index used in a replica identity, we could make the
reset of relreplident for the parent table consistent with the moment
the index is marked as invalid:
- In order to make CREATE INDEX CONCURRENTLY work for partitioned
tables, we need to be able to switch indisvalid for all the index
partitions in one final transaction so as we have a consistent
partition tree at any state of the operation. Obviously,
heap_inplace_update() is a problem here.

The restriction we have in place now can be lifted thanks to 813fb03,
that removed SnapshotNow, and as far as I looked at the code paths
doing scans of pg_index, I don't see any reasons why we could not make
this stuff transactional. Perhaps that's just because nobody had use
cases where it would be useful, and I have two of them lately. Note
that 3c84046 was the original commit that introduced the use of
heap_inplace_update() and index_state_set_flags(), but we don't have
SnapshotNow anymore. Note as well that we already make the index
swapping transactional in REINDEX CONCURRENTLY for the pg_index

I got the feeling that this is an issue different than the other
threads where this was discussed, which is why I am beginning a new
thread. Any thoughts? Attached is a proposal of patch.


Attachment Content-Type Size
reindex-transactional-v1.patch text/x-diff 1.9 KB


Browse pgsql-hackers by date

  From Date Subject
Next Message Michael Paquier 2020-09-03 08:18:29 Re: Improving connection scalability: GetSnapshotData()
Previous Message Daniel Gustafsson 2020-09-03 07:47:07 Re: Switch to multi-inserts for pg_depend