Re: Database corruption event, unlockable rows, possibly bogus virtual xids? Invalid backend server xid

Ned Wolpert <ned.wolpert@imemories.com> writes:
>   Event: Running 9.1.6 with hot-standby, archiving 4 months of wal files,
> and even a nightly pg_dump all. 50G database.  Trying to update or delete a
> row in a small (21 row, but heavily used table) would lock up completely.
> Never finish. Removed all clients, restarted the db instance, no joy. Check
> pg_stat_activity, and nothing that wasn't idle.... run the delete, locked
> up.

I'm betting one of those prepared transactions had updated or deleted
this row, and thus held a row lock on it.  (Come to think of it, a
SELECT FOR UPDATE/SHARE might have been enough.)  Did you try committing
or rolling back those xacts?

            regards, tom lane