On 28/02/11 03:57, Sean Laurent wrote:
> Right. I read all of that. I guess I just assumed it was possible to create
> a snapshot on the standby so that a longer running on the standby could
> complete. In particular, I was really hoping to run database dumps against
> the standby, not the master.
This makes me wonder if a possible extension to HS might (one day) be a
side log for rows that are still needed for transactions running on the
slave, but have been overwritten on the master.
When a log replay overwrites a page containing tuples still needed by a
transaction running on the slave, it could be copied to a temporary heap
that's deleted when the last slave transaction needing those tuples
finishes.
That'd be a LOT of work and would introduce differences in query
execution that'd make the standby more different to the master
(increasing testing burden) so it's not exactly an obvious no-brainer
thing to do. I'm sure it'd also be seriously complicated to get right,
and isn't anything I'd ever ask anybody to do unless I was paying them
well for their time. Maybe someone who really needs functionality like
this will fund it at some point - personally, I don't presently even use
HS, so it's well beyond my needs.
--
Craig Ringer
