Horton

Re: 'ERROR: attempted to update invisible tuple' from 'ALTER INDEX ... ATTACH PARTITION' on parent index

Michael Paquier <michael@paquier.xyz>

View thread
From: Michael Paquier <michael@paquier.xyz>
To: Shruthi Gowda <gowdashru@gmail.com>
Cc: PostgreSQL Development <pgsql-hackers@postgresql.org>
Date: 2023-07-12T00:38:41Z
Lists: pgsql-hackers

Commits

Same data as JSON: GET /api/v1/messages/:b64id/commits the thread's linked commits as JSON, with link sources. API reference →

  1. Add indisreplident to fields refreshed by RelationReloadIndexInfo()

  2. Fix updates of indisvalid for partitioned indexes

On Tue, Jul 11, 2023 at 10:52:16PM +0530, Shruthi Gowda wrote:
> While testing some use cases, I encountered 'ERROR:  attempted to update
> invisible tuple' when a partitioned index is attached to a parent index
> which is also a replica identity index.
> Below is the reproducible test case. The issue is seen only when the
> commands are executed inside a transaction.

Thanks for the report, reproduced here.

> The 'ALTER INDEX pk_foo ATTACH PARTITION foo_2023_id_ts_ix' returns
> "*ERROR:  attempted to update invisible tuple"*

While working recently on what has led to cfc43ae and fc55c7f, I
really got the feeling that there could be some command sequences that
lacked some CCIs (or CommandCounterIncrement calls) to make sure that
the catalog updates are visible in any follow-up steps in the same
transaction.

> The 'indisreplident' is false, the ctid field value is old and it does
> not reflect the ctid changes made by  'ALTER TABLE ONLY foo REPLICA
> IDENTITY USING INDEX pk_foo'.

Your report is telling that we are missing a CCI somewhere in this
sequence.  I would have thought that relation_mark_replica_identity()
is the correct place when the pg_index entry is dirtied, but that does
not seem correct.  Hmm.
--
Michael