Albe Laurenz wrote:
> As demonstrated here:
> http://archives.postgresql.org/pgsql-general/2010-02/msg01057.php
>
> it can happen that you end up with a corrupt database
> if you use PITR to recover past a hardware failure.
>
> It seems that unique constraints are not verified during
> WAL replay.
>
> Is that intentional or an oversight that should be fixed?
It's intentional. WAL replay works at a lower level, it has no knowledge
of operators or the concept of uniqueness.
-- Heikki Linnakangas EnterpriseDB http://www.enterprisedb.com