On Tuesday 11 August 2009 09:49:14 Tom Lane wrote:
> Kevin Kempter <kevink@consistentstate.com> writes:
> > due to operator error at one of my client sites we're trying to restore a
> > deleted db.
> >
> > We did a restore with a pgdump that went fine but the dump is 3weeks old.
> > they do have WAL segments and they claim that the WAL segments are
> > exactly in line with the needed transactions - from a timing perspective.
> > However no base backup with pg_start_backup() and pg_stop_backup() was
> > ever run.
> >
> > Is there any way to trick postgres into thinking that it needs to play
> > fwd these WAL's? Would it work to simply place them in a dir, create a
> > recovery.conf file pointing to that dir and restart the cluster?
>
> No, there is exactly 0 hope here. You would need a filesystem-level
> backup to work from. A restore from a pg_dump is not going to reproduce
> any of the low-level details (such as OID assignments or tuple
> placement) that WAL replay needs to have match.
>
> regards, tom lane
What if we could do a pg_start_backup on the running cluster (the one that's 3
weeks behind, tar up the file system, do a pg_stop_backup and walk thru the
recovery steps?
Would there be a way to force postgres to recognize the wal segments that
werte created earlier (which according to the client include all tx from 3
weeks ago thru today) ?