Re: Use pg_rewind when target timeline was switched

On Fri, Sep 18, 2015 at 9:00 AM, Alexander Korotkov wrote:
> BTW, it would be an option to generate system_identifier to each new
> timeline, by analogy of initdb do for the whole WAL.
> Having such system_identifiers we can distinguish different timeline which
> have assigned same ids.
> Any thoughts?

If you mean a new field incorporated in XLogLongPageHeader and
ControlFile to ensure that a new timeline generated is unique across
the same installation identified with system_identifier, then I'm not
against it for something up to 2 bytes (even 1 byte would be fine),
though it seems that there is a very narrow need for it. Do you have
cases in mind that could use it? Even in the case of pg_rewind, it
seems to me we would not need this extra timeline-based ID. Per your
case above, it is possible to rewind node 4 on timeline 3 using node 2
on timeline 2 when both timelines have forked exactly at the same
point from timeline 1, by having node 4 beginning to replay from
timeline 1 (last checkpoint record before WAL forked), and if I am
reading your patch correctly that's what you do.
-- 
Michael