Horton

Re: prevent immature WAL streaming

Robert Haas <robertmhaas@gmail.com>

View thread View on postgr.es
From: Robert Haas <robertmhaas@gmail.com>
To: Amul Sul <sulamul@gmail.com>
Cc: Alvaro Herrera <alvherre@alvh.no-ip.org>, "Bossart, Nathan" <bossartn@amazon.com>, "Jakub.Wartak@tomtom.com" <Jakub.Wartak@tomtom.com>, Kyotaro Horiguchi <horikyota.ntt@gmail.com>, Ryo Matsumura <matsumura.ryo@fujitsu.com>, "andres@anarazel.de" <andres@anarazel.de>, "masao.fujii@oss.nttdata.com" <masao.fujii@oss.nttdata.com>, "mengjuan.cmj@alibaba-inc.com" <mengjuan.cmj@alibaba-inc.com>, "pgsql-hackers@lists.postgresql.org" <pgsql-hackers@lists.postgresql.org>
Date: 2021-10-13T17:14:32Z
Lists: pgsql-hackers 
On Wed, Oct 13, 2021 at 2:01 AM Amul Sul <sulamul@gmail.com> wrote:
> Instead of abortedRecPtr point, isn't enough to write
> overwrite-contrecord at XLogCtl->lastReplayedEndRecPtr?  I think both
> are pointing to the same location then can't we use
> lastReplayedEndRecPtr instead of abortedRecPtr to write
> overwrite-contrecord and remove need of extra global variable, like
> attached?

I think you mean missingContrecPtr, not abortedRecPtr. If I understand
correctly, abortedRecPtr is going to be the location in some WAL
segment which we replayed where a long record began, but
missingContrecPtr seems like it would have to point to the beginning
of the first segment we were unable to find to continue replay; and
thus it ought to be the same as lastReplayedEndRecPtr. But the
committed code doesn't seem to check that these are the same or verify
the relationship between them in any way, so I'm worried there is some
other case here. The comments in XLogReadRecord also suggest this:

         * We get here when a record that spans multiple pages needs to be
         * assembled, but something went wrong -- perhaps a contrecord piece
         * was lost.  If caller is WAL replay, it will know where the aborted

Saying that "perhaps" a contrecord piece was lost seems to imply that
other explanations are possible as well, but I'm not sure what.

-- 
Robert Haas
EDB: http://www.enterprisedb.com

Commits

  1. Fix determination of broken LSN in OVERWRITTEN_CONTRECORD

  2. Fix instability in 026_overwrite_contrecord.pl test.

  3. Fix typo

  4. Change recently added test code for stability

  5. Remove unstable, unnecessary test; fix typo

  6. Repair two portability oversights of new test

  7. Fix WAL replay in presence of an incomplete record

  8. Document XLOG_INCLUDE_XID a little better

  9. Revert "Avoid creating archive status ".ready" files too early"

  10. Avoid creating archive status ".ready" files too early

  11. Fix scenario where streaming standby gets stuck at a continuation record.

  12. Teach xlogreader to follow timeline switches

  13. Forbid using pg_xlogfile_name() and pg_xlogfile_name_offset() during