> Date: Tue, 15 Mar 2016 17:21:50 +0100
> Subject: Re: how to switch old replication Master to new Standby after promoting old Standby - pg_rewind log file
missing
> From: michael.paquier@gmail.com
> To: johnlumby@hotmail.com
> CC: pgsql-general@postgresql.org; oleksandr.shulgin@zalando.de; adrian.klaver@aklaver.com
>
> On Tue, Mar 15, 2016 at 5:14 PM, John Lumby <johnlumby@hotmail.com> wrote:
>> But my question is, given that the divergence point was 2B000060,
>> why is it looking for a file earlier than that?
>
> (please do not top-post, this is annoying as it breaks the logic of the thread)
sorry, ok
> pg_rewind begins scanning WAL records not at the point where WAL
> forked, but from the previous checkpoint record before WAL forked.
In that case, I wonder if it would be useful to extend the replication_slots concept
to be able to specify to retain all WALs back to
"previous checkpoint before most recent timeline switch"
in order to guarantee that pg_rewind shall always be able to find the ones it needs
in source without requiring archiving and archive-retrieval?
> --
> Michael