At Tue, 22 Feb 2022 01:59:45 +0900, Fujii Masao <masao.fujii@oss.nttdata.com> wrote in
>
> > Of the following, I think we should do (a) and (b) to make future
> > backpatchings easier.
> > a) Use RedoRecPtr and PriorRedoPtr after they are assigned.
> > b) Move assignment to PriorRedoPtr into the ControlFileLock section.
>
> I failed to understand how (a) and (b) can make the backpatching
> easier. How easy to backpatch seems the same whether we apply (a) and
> (b) or not...
That premises that the patch applied to master contains (a) and (b).
So if it doesn't, those are not need by older branches.
> > c) Skip udpate of minRecoveryPoint only when the checkpoint gets old.
>
> Yes.
>
>
> > d) Skip call to UpdateCheckPointDistanceEstimate() when RedoRecPtr <=
> > PriorRedoPtr.
>
> But "RedoRecPtr <= PriorRedoPtr" will never happen, will it? Because a
I didn't believe that it happens. (So, it came from my
convervativeness, or laziness, or both:p) The code dates from 2009 and
StartupXLOG makes a concurrent checkpoint with bgwriter. But as of at
least 9.5, StartupXLOG doesn't directly call CreateCheckPoint. So I
think that won't happen.
So, in short, I agree to remove it or turn it into Assert().
> restartpoint is skipped at the beginning of CreateRestartPoint() in
> that case. If this understanding is right, the check of "RedoRecPtr <=
> PriorRedoPtr" is not necessary before calling
> UpdateCheckPointDistanceEstimate().
>
>
> + ControlFile->minRecoveryPoint = InvalidXLogRecPtr;
> + ControlFile->minRecoveryPointTLI = 0;
>
> Don't we need to update LocalMinRecoveryPoint and
> LocalMinRecoveryPointTLI after this? Maybe it's not necessary, but
> ISTM that it's safer and better to always update them whether the
> state is DB_IN_ARCHIVE_RECOVERY or not.
Agree that it's safer and tidy.
> if (flags & CHECKPOINT_IS_SHUTDOWN)
> ControlFile->state = DB_SHUTDOWNED_IN_RECOVERY;
>
> Same as above. IMO it's safer and better to always update the state
> (whether the state is DB_IN_ARCHIVE_RECOVERY or not) if
> CHECKPOINT_IS_SHUTDOWN flag is passed.
That means we may exit recovery mode after ShutdownXLOG called
CreateRestartPoint. I don't think that may happen. So I'd rather add
Assert ((flags&CHECKPOINT_IS_SHUTDOWN) == 0) there instaed.
I'll post the new version later.
regards.
--
Kyotaro Horiguchi
NTT Open Source Software Center