On Fri, Nov 29, 2019 at 7:23 AM Tom Lane <tgl@xxxxxxxxxxxxx> wrote: > The most likely "corruption" explanation is something wrong with the > indexes on the referenced and/or referencing column, causing rows to > not be found when referential actions should have found them. Random > querying of the tables wouldn't necessarily expose that --- you'd need > to be sure that your queries use the questionable indexes, and maybe > even search for some of the specific rows that seem mis-indexed. Or try using contrib/amcheck, which is available in Postgres 10. Perhaps try the query here, modified to verify all B-Tree indexes (not just those indexes in the pg_catalog schema): https://www.postgresql.org/docs/10/amcheck.html -- Peter Geoghegan
