Hi,
On 3/2/18 1:03 PM, Fujii Masao wrote:
> On Fri, Mar 2, 2018 at 1:07 PM, Michael Paquier <michael@paquier.xyz> wrote:
>
>> We would talk about two backups running
>> simultaneously on a standby, which would overlap with each other to
>> generate a file aimed only at being helpful for debugging purposes, and
>> we provide no information now for backups taken from standbys. We could
>> of course make that logic a bit smarter by checking if there is an
>> extsing file with the same name and create a new file with a different
>> name. But is that worth the complication? That's where I am not
>> convinced, and that's the reason why this patch is doing things this
>> way.
>
> What about including the backup history file in the base backup instead of
> creating it in the standby's pg_wal and archiving it? As the good side effect
> of this approach, we can use the backup history file for debugging purpose
> even when WAL archiving is not enabled.
I don't think this is a good idea, even if it does solve the race
condition. First, it would be different from the way primary-style
backups generate this file. Second, these files would not be excluded
from future restore / backup cycles so they could build up after a while
and cause confusion. I guess we could teach PG to delete them or
pg_basebackup to ignore them, but that doesn't seem worth the effort.
Regards,
--
-David
david@pgmasters.net