On Fri, May 28, 2010 at 7:58 PM, Heikki Linnakangas
<heikki.linnakangas@enterprisedb.com> wrote:
> At PGCon, several people asked me about restarting an old master as a
> standby after failover has happened. And it wasn't the first time people ask
> me about it, even before 9.0. We have no mention of that in the docs, which
> is a pretty serious oversight. What can we say about it?
>
> I believe the current official policy is that you have to take a new base
> backup and restore from that. Rsync can be used to speed that up.
>
> However, someone once asked me for a comment on a script he wrote to do that
> in a smarter way. I forget who and when and how exactly it worked, but it
> seems possible to do safely.
>
> First of all, you have to shut down the master cleanly for this to work,
> otherwise there can be changes in the old master that never made it to the
> standby.
>
> Assuming controlled shutdown and that the standby received all WAL from the
> old master before it was promoted, I think you can simply create a
> recovery.conf in the old master's data directory to turn it into a standby
> server, and restart. Am I missing something?
Failover always increments the timeline ID of the old standby (i.e.,
new master).
Can that procedure work around the gap of the timeline ID between servers?
Regards,
--
Fujii Masao
NIPPON TELEGRAPH AND TELEPHONE CORPORATION
NTT Open Source Software Center