On Thu, Jan 22, 2015 at 06:04:24PM -0600, Jim Nasby wrote:
> On 1/22/15 5:43 PM, Bruce Momjian wrote:
> >This brings up the other problem that the mod times of the files
> >are likely to be different between master and slave --- should I
> >recommend to only use rsync --checksum?
>
> I don't think so. AIUI if the timestamps are different the very next
> thing it does is run the checksum (which is expensive). So --checksum
> is just going to hurt.
Oh, OK, good.
> >I am going to now conclude that rsync is never going to work for
> >this, unless we have pg_upgrade preserve relfilenodes as well.
> >However, I am not even sure that is possible due to conflicts with
> >system table relfilenodes created in the new cluster.
>
> We've previously talked about required steps before an upgrade;
> perhaps we need a way to force an OID/relfilenode change on the old
> server prior to upgrade.
Actually, the idea I had forgotten is that we are not rsyncing between
old and new clusters here, but between two servers who are both new
after running pg_upgrade. Their relfilenodes match their oid, and the
oids match, so we should be fine.
-- Bruce Momjian <bruce@momjian.us> http://momjian.us EnterpriseDB
http://enterprisedb.com
+ Everyone has their own god. +