On Mon, 2004-07-05 at 23:40, Mike Mascari wrote:
> Simon Riggs wrote:
>
> >
> > ...Nobody is shouting YES, so its a dodo...
>
> The point at which the above process becomes
> too complex (or less than obvious) for hand-recovery is precisely
> when unforeseen consequences of nixing a single transaction become
> too great.
>
Agreed. The potential for unforeseen consequences is just too high, and
although I'm fairly sure they would always be spotted during recovery
and cause an error - I think it is something that requires proof. And I
don't have that. So, lets leave that idea alone for 100 years.
> ... hand-recovery ...
hmmm...not sure I know what you mean.
It is very-very-close-to-impossible to edit the transaction logs
manually, unless some form of special-format editor were written for the
purpose.
Is it clear that the PITR features are completely different from
pg_dump? (Which would allow a manual edit and recover). The xlogs are
binary files that refer to all changes to all tables in a cluster
ordered by time, rather than by table.
Best regards, Simon Riggs
