model-writtenrun 8393 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Andrey Klychkov reported that repeated reindexdb --concurrently failures on duplicate keys accumulated invalid indexes in PostgreSQL 12.3 #1; Peter Eisentraut investigated whether an operating-system upgrade could explain the underlying index corruption #2, but Andrey said PostgreSQL was installed after the system update #3.
- Tom Lane distinguished the unexplained duplicate data from the expected requirement to remove invalid indexes after concurrent reindexing fails #4, and Michael Paquier pointed to the existing recovery instructions on the REINDEX documentation page #5.
- Andrey asked why one unique-index failure left other indexes invalid #6; Michael explained that concurrent table reindexing creates all replacement indexes before building and validating them, so a build failure leaves multiple invalid replacements #8.
- Tom identified that reindexdb users needed clearer guidance on its own documentation page #7; Michael proposed strengthening the link to REINDEX without duplicating its documentation #9, and Tom supplied the final sentence #10.
- Michael posted the documentation patch #11, Tom gave a brief assent #12, and Michael committed it unchanged as 9b60c4b979, with a backpatch to PostgreSQL 12, then announced application #13.
- After the commit, Andrey requested documentation showing how to discover invalid TOAST indexes and drop them #14; Michael supported adding cleanup queries near the existing recovery instructions, while noting the need to handle schema qualification #15.
- Alvaro Herrera subsequently questioned the invalid nonunique replacement index and suggested that reindexdb remove indexes it could not finish #16; Michael offered a tentative explanation and welcomed automatic cleanup #17, leaving these follow-up ideas without an attached patch or resolution in this thread.