Yeah, but the indrelid did not change after the the REINDEX.
Agreed, but none of this makes sense. pg_dump grabs pks via a bulk pg_index scan based on table oids, so I wanted to rule out some problem there.
I'm chalking this one up to user error, not database corruption, as the OP has not actually shown us the output of how they are determining the missing index, and then how the reindex fixes it (although I appreciate the pg_index query results). To put another way, user error is a much more likely explanation than anything else given the lack of specific data.
