Re: Inconsistency in reporting checkpointer stats

At Mon, 19 Dec 2022 18:05:38 +0530, Bharath Rupireddy <bharath(dot)rupireddyforpostgres(at)gmail(dot)com> wrote in
> On Fri, Dec 16, 2022 at 2:14 PM Kyotaro Horiguchi
> <horikyota(dot)ntt(at)gmail(dot)com> wrote:
> >
> > In the first place I don't like that we count the same things twice.
> > Couldn't we count the number only by any one of them?
> >
> > If we remove CheckPointerStats.ckpt_bufs_written, CreateCheckPoint can
> > get the final number as the difference between the start-end values of
> > *the shared stats*. As long as a checkpoint runs on a single process,
> > trace info in BufferSync will work fine. Assuming single process
> > checkpointing there must be no problem to do that. (Anyway the current
> > shared stats update for checkpointer is assuming single-process).
>
> What if someone resets checkpointer shared stats with
> pg_stat_reset_shared()? In such a case, the checkpoint complete
> message will not have the stats, no?

I don't know. I don't believe the stats system doesn't follow such a
strict resetting policy.

regards.

--
Kyotaro Horiguchi
NTT Open Source Software Center