Re: Issues with ON CONFLICT UPDATE and REINDEX CONCURRENTLY

Mihail Nikalayeu <michail.nikolaev@gmail.com>

From: Michail Nikolaev <michail.nikolaev@gmail.com>

To: Michael Paquier <michael@paquier.xyz>

Cc: PostgreSQL Hackers <pgsql-hackers@postgresql.org>

Date: 2024-06-14T10:06:13Z

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 →

Replace flaky CIC/RI isolation tests with a TAP test
- e1c971945d62 19 (unreleased) landed
Disable recently added CIC/RI isolation tests
- 77038d6d0b49 19 (unreleased) landed
Fix infer_arbiter_index for partitioned tables
- 81f72115cf18 19 (unreleased) landed
Stabilize tests some more
- a4a0fa0c7587 19 (unreleased) landed
Put back alternative-output expected files
- be25c7767728 19 (unreleased) landed
Remove doc and code comments about ON CONFLICT deficiencies
- 758479213d57 19 (unreleased) landed
Avoid use of NOTICE to wait for snapshot invalidation
- 5dee7a603f66 19 (unreleased) landed
Fix ON CONFLICT with REINDEX CONCURRENTLY and partitions
- 90eae926abbb 19 (unreleased) landed
Fix ON CONFLICT ON CONSTRAINT during REINDEX CONCURRENTLY
- 2bc7e886fc1b 19 (unreleased) landed
Fix new test for CATCACHE_FORCE_RELEASE builds
- 9e8fa05d3412 19 (unreleased) landed
Improve test case stability
- 417ac9c1eeeb 19 (unreleased) landed
Fix infer_arbiter_index during concurrent index operations
- bc32a12e0db2 19 (unreleased) landed
Doc: cover index CONCURRENTLY causing errors in INSERT ... ON CONFLICT.
- 438ccd33a37a 13.23 landed
- faba2595d619 14.20 landed
- e1dd1f924e19 15.15 landed
- 4c2895629f6f 16.11 landed
- 46524d519a2d 18.1 landed
- 27f714b5dad7 17.7 landed
- 8b18ed6dfbb8 19 (unreleased) landed
Fix infer_arbiter_indexes() to not assume resultRelation is 1.
- 915de706d28c 17.0 cited
Revert temporal primary keys and foreign keys
- 8aee330af55d 17.0 cited

Hello, Michael.

> Isn't the issue that we may select as arbiter indexes stuff that's
!indisvalid?
As far as I can see (1) !indisvalid indexes are filtered out.

But... It looks like this choice is not locked in any way (2), so
index_concurrently_swap or index_concurrently_set_dead can change this
index after the decision is made, even despite WaitForLockersMultiple (3).
In some cases, it may cause a data loss...
But I was unable to reproduce that using some random usleep(), however -
maybe it is a wrong assumption.

[1]:
https://github.com/postgres/postgres/blob/915de706d28c433283e9dc63701e8f978488a2b9/src/backend/optimizer/util/plancat.c#L804

[2]:
https://github.com/postgres/postgres/blob/915de706d28c433283e9dc63701e8f978488a2b9/src/backend/optimizer/util/plancat.c#L924-L928
[3]:
https://github.com/postgres/postgres/blob/8aee330af55d8a759b2b73f5a771d9d34a7b887f/src/backend/commands/indexcmds.c#L4153