On Thu, Nov 30, 2006 at 03:59:15PM -0500, Tom Lane wrote:
> Stephen Harris <lists@spuddy.org> writes:
> > Starting up the standby database still goes back to earlier log files,
> > but I guess that's the 100 checkpoint thing you mentioned earlier.
>
> Actually I was misremembering that: the frequency of recovery
> checkpoints is time-based, and for a slave that's in sync with its
> master, they should occur about as often as checkpoints on the master.
> So the amount of log to be replayed should be comparable to what the
> master would have to replay if it crashed.
Hmm. For testing purposes I have
archive_timeout=5
so an archive log is created every 5 seconds if there's activity.
In idle periods it's created every 5 minutes as the internal checkpoint
occurs on the server. However if I replay all those "empty" logs then
shutdown then startup it will ask for some of them again.
As a test, I started a recover, recovered 11 archive logs
(000000010000000000000035 to 00000001000000000000003F), some may have had
activity; I not sure. I then waited 10 minutes after the last log file
had been replayed. Then I did a fast shutdown. Starting the standby
database again required it going back to 000000010000000000000035 again.
--
rgds
Stephen