Horton

Thread

Commits

  1. Improve error messages for some callers of XLogReadRecord()

  1. Improve error context after some failed XLogReadRecord()

    Michael Paquier <michael@paquier.xyz> — 2021-11-09T01:47:11Z

    Hi all,

I have noticed $subject while looking at a patch in this area:
https://www.postgresql.org/message-id/YYiqxtEqhgb5gIWL@paquier.xyz

While we don't expect the four callers of XLogReadRecord() related to
logical decoding to fail in the code paths changed by the patch
attached, I think that it would be good to provide more context to
users rather than just emitting an elog(ERROR) coming directly from
the WAL reader.  This way, it becomes a bit easier to guess from where
the failure is coming.

A patch is attached to improve all those elog()s.  I have tried not to
be too fancy about those new error strings, but I am sure that these
could be tweaked.

Thoughts?
--
Michael

  2. Re: Improve error context after some failed XLogReadRecord()

    Kyotaro Horiguchi <horikyota.ntt@gmail.com> — 2021-11-09T07:58:11Z

    At Tue, 9 Nov 2021 10:47:11 +0900, Michael Paquier <michael@paquier.xyz> wrote in 
> Hi all,
> 
> I have noticed $subject while looking at a patch in this area:
> https://www.postgresql.org/message-id/YYiqxtEqhgb5gIWL@paquier.xyz
> 
> While we don't expect the four callers of XLogReadRecord() related to
> logical decoding to fail in the code paths changed by the patch
> attached, I think that it would be good to provide more context to
> users rather than just emitting an elog(ERROR) coming directly from
> the WAL reader.  This way, it becomes a bit easier to guess from where
> the failure is coming.
> 
> A patch is attached to improve all those elog()s.  I have tried not to
> be too fancy about those new error strings, but I am sure that these
> could be tweaked.
> 
> Thoughts?

As the whole, I agree that any substantial message is far better than
"%s" in regard to identifying the line that emitted a log line.

+				elog(ERROR, "could not find record to advance replication slot: %s",

I prefer "could not read record while advancing replcation slot",
because it is not finding a record but it is reading successive
records.  However, I don't strongly object to the current wording.

+		elog(ERROR, "could not find record to send logically-decoded data: %s", errm);

I prefer something like the above for the same reason.  How about
"could not read record while sending logically-decoded data:".  The
same with above, I'm not strongly opposed to the current wording.

Otherwise it looks good to me.

regards.

-- 
Kyotaro Horiguchi
NTT Open Source Software Center

  3. Re: Improve error context after some failed XLogReadRecord()

    Michael Paquier <michael@paquier.xyz> — 2021-11-10T06:59:16Z

    On Tue, Nov 09, 2021 at 04:58:11PM +0900, Kyotaro Horiguchi wrote:
> Otherwise it looks good to me.

What you are proposing is better than what I had in those two
contexts, so I have just used what you have suggested for both of
them.  Thanks!
--
Michael