Re: A query in Streaming Replication

At Sat, 13 Jun 2020 23:12:25 +0530, Sreerama Manoj <manoj(dot)sreerama973(at)gmail(dot)com> wrote in
> Hello,
>
> Forgot to add the version details. I'm using postgres 10.7 .
>
> On Sat, Jun 13, 2020, 20:26 Sreerama Manoj <manoj(dot)sreerama973(at)gmail(dot)com>
> wrote:
>
> > Hello,
> > I use streaming replication in async mode. When master gets down,
> > slave will be promoted using a trigger file. During this process ".partial"
> > file will be created and a WAL file with same ID will be created in a
> > different time line in slave. When master comes back as slave, it will be
> > synced to the current master. This is the normal procedure as far as I
> > understood. But in some cases, ".partial" file is not getting created and

It is not guaranteed that a crashed master can be used as a new
standby as-is, since there can be unsent WAL records on the old master
after the LSN where the new master promoted. If files are assumed to
be sound, pg_rewind will adjust the old master as a new standby.

The .partial file is created when timeline diverges at midst of a WAL
segment. It is useful when performing PITR to the end of the same
timeline, rather than going into the next timeline. I don't have an
idea at hand of how this can be relevant to the reusability of the old
master..

> > peer DB which comes back as slave is unable to sync when this happens..
> > Please suggest if this happens in any scenario and how to overcome this.

regards.

--
Kyotaro Horiguchi
NTT Open Source Software Center