Hello all,
I finally managed to reproduce the bug on a demo database as shown below.
The file err_demo.sql shows all the steps to reproduce. The table creation is done with errdb_noerr_231008.sql (unchanged from last mail).
The clue is:
- you have an inherited table
- on the archiv part, you have only (one or more) brin indexes
- to reorder the tuples before moving to a new major version (with pg_dumpo/restore) the archiv part is clustered.
this requires to create a btree index, which is removed immediately after the cluster statement.
when all this has been done, a write access to the archiv-part of the table hierarchy causes the error.
A direct update access to or_followup_archiv clears the error (as shown in the final comment part of err_demo.sql).
To make the error reappear, you have to redo the 3 recluster steps.
PS: please apply the steps of err_demo.sql manually because it was not tested for full automation.
Hope this enables you to investigate the root cause of the error.