On Fri, Sep 21, 2012 at 10:41 AM, Andres Freund <andres@2ndquadrant.com> wrote:
> Hrm. I retract my earlier statement about the low likelihood of corruption due
> to this.
Yeah. :-(
We've recently had at least one report of autovacuum failing to
terminate due to a series of index pages forming a circular loop, and
at least one case where it appears that the data became not-unique on
a column upon which a unique index existed, in releases that contain
this bug.
It seems therefore that REINDEX + VACUUM with
vacuum_freeze_table_age=0 is not quite sufficient to recover from this
problem. If your index has come to contain a circularity, vacuum will
fail to terminate, and you'll need to drop it completely to recover.
And if you were relying on your index to enforce a unique constraint
and it didn't, you'll need to do manual data repair before it will be
possible to rebuild or replace that index.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company