On 2/19/18 10:32 AM, Don Seiler wrote:
> On Mon, Feb 19, 2018 at 9:21 AM, David Steele <da...@pgmasters.net
> <mailto:da...@pgmasters.net>> wrote:
> Yes, they are typically very small. The general exception to this rule
> is if logs are stored in pg_log. I recommend storing logs out of the
> PGDATA dir as they can be quite large and don't really make sense to
> restore to another server.
> Files copied from the master will be marked as such in backup.manifest
> (master:true) so you can check for yourself.
> Good to know. And fortunately for this DB we do have pg_log (and
> pg_xlog) symlinked to different volumes outside of $PGDATA.
If pg_log is symlinked to PGDATA it will be copied. pg_xlog is not
copied in any backup.
> > I did come up with a sort of Rube Goldberg-esque workaround for now
> > involving using a clone of the prod standby VM from Veeam backup to use
> > as the backup source (after stopping recovery and opening it as a
> > standalone DB).
> You don't get PITR that way, of course, but at least it's a backup. As
> long as your clone is consistent.
> Yes it's a crash-consistent snapshot-based backup. I've done quite a few
> restores from it and it works great. It can do PITR as well since I
> would have all the WAL files from prod needed to keep recovering. But
> for these cases I just recover it to the first consistent point and open
> it for testing (or backups in this case).
I don't think it would be safe to do PITR on a backup taken in this way.
The WAL diverges even if you suppress a timeline switch.