Re: Race condition in recovery?

At Mon, 17 May 2021 13:01:04 +0900 (JST), Kyotaro Horiguchi <horikyota(dot)ntt(at)gmail(dot)com> wrote in
> At Mon, 17 May 2021 12:20:12 +0900 (JST), Kyotaro Horiguchi <horikyota(dot)ntt(at)gmail(dot)com> wrote in
> > Assuming that we keep expectedTLEs synced with recoveryTargetTLI,
> > rescanLatestTimeLine updates the list properly so no need to worry
> > about the future. So the issue would be in the past timelines. After
> > reading the checkpoint record, if we need to rewind to the previous
> > timeline for the REDO point, the dummy list is inconvenient.
>
> By the way, I tried reproducing this situation, but ended in finding
> it a kind of impossible because pg_basebackup (or pg_stop_backup())
> waits for the promotion checkpoint to end.

Mmm. That's wrong. What the tool waits is not a promotion checkpoint,
but a backup-checkpoint, maybe. (I don't remember cleary about that,
sorry.)

> If we make a backup in a somewhat broken steps, that could be done but
> I didn't try.

So there might still be a way to reproduce that.

regards.

--
Kyotaro Horiguchi
NTT Open Source Software Center