On Thu, 21 Mar 2024 at 07:17, Andrey M. Borodin <x4...@yandex-team.ru> wrote: > > > On 29 Jun 2022, at 17:43, Robins Tharakan <thara...@gmail.com> wrote: > > Sorry to bump ancient thread, I have some observations that might or might > not be relevant. > Recently we noticed a corruption on one of clusters. The corruption at hand > is not in system catalog, but in user indexes. > The cluster was correctly configured: checksums, fsync, FPI etc. > The cluster never was restored from a backup. It’s a single-node cluster, so > it was not ever promoted, pg_rewind-ed etc. VM had never been rebooted. > > But, the cluster had been experiencing 10 OOMs a day. There were no torn > pages, no checsum erros at log at all. Yet, B-tree indexes became corrupted.
Would you happen to have a PostgreSQL version number (or commit hash) to help debugging? Has it always had that PG version, or has the version been upgraded? Considering the age of this thread, and thus potential for v14 pre-.4: Did this cluster use REINDEX (concurrently) for the relevant indexes? Kind regards, Matthias van de Meent