Re: BUG #15346: Replica fails to start after the crash

At Thu, 30 Aug 2018 18:48:55 -0700, Michael Paquier <michael@paquier.xyz> wrote in <20180831014855.GJ15446@paquier.xyz>
> On Fri, Aug 31, 2018 at 09:48:46AM +0900, Kyotaro HORIGUCHI wrote:
> > Please wait a bit.. I have a concern about this.
> 
> Sure, please feel free.

Thanks.

I looked though the patch and related code and have a concern.

The patch inhibits turning off updateMinRecoveryPoint on other
than the startup process running crash-recovery except at the end
of XLogNeedsFlush.

>     /*
>      * Check minRecoveryPoint for any other process than the startup
>      * process doing crash recovery, which should not update the control
>      * file value if crash recovery is still running.
>      */
>     if (XLogRecPtrIsInvalid(minRecoveryPoint))
>       updateMinRecoveryPoint = false;

Even if any other processes than the startup calls the function
during crash recovery, they don't have a means to turn on
updateMinRecoveryPoint again. Actually the only process that
calls the function druing crash recovery is the startup. bwriter
and checkpointer doesn't. Hot-standby backends come after
crash-recvery ends. After all, we just won't have an invalid
minRecoveryPoint value there, and updateMinRecoverypoint must be
true.

Other than that I didn't find a problem. Please find the attached
patch. I also have not come up with how to check the case
automatically..


regards.

-- 
Kyotaro Horiguchi
NTT Open Source Software Center