Horton

Re: XLogReadRecord() error in XlogReadTwoPhaseData()

Tom Lane <tgl@sss.pgh.pa.us>

View thread View on postgr.es
From: Tom Lane <tgl@sss.pgh.pa.us>
To: Noah Misch <noah@leadboat.com>
Cc: Michael Paquier <michael@paquier.xyz>, pgsql-hackers@postgresql.org
Date: 2021-11-17T22:47:10Z
Lists: pgsql-hackers 
Noah Misch <noah@leadboat.com> writes:
> Tom Lane reported another instance today:
> https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=tadarida&dt=2021-11-11%2013%3A29%3A58

> Each of the three failures happened on a sparc64 Debian+gcc machine.  I had
> tried ~8000 iterations on thorntail, another sparc64 Debian+gcc animal,
> without reproducing this.

>>> As a first step, let's report the actual XLogReadRecord() error message.
>>> Attached.

>> Good catch!  This looks good.

> Pushed.

Well, we didn't have to wait too long [1]:

#   at t/003_cic_2pc.pl line 143.
#                   'pgbench: error: client 0 script 1 aborted in command 4 query 0: ERROR:  could not read two-phase state from WAL at 0/159EF88: unexpected pageaddr 0/0 in log segment 000000010000000000000001, offset 5890048
# pgbench: error: client 2 script 3 aborted in command 2 query 0: ERROR:  canceling statement due to lock timeout
# pgbench: fatal: Run was aborted; the above results are incomplete.

I suppose "unexpected pageaddr 0/0" is most easily explained by supposing
that XlogReadTwoPhaseData tried to read a WAL page that hadn't been
written out yet.  Have we got any synchronization around that?

			regards, tom lane

[1] https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=tadarida&dt=2021-11-17%2013%3A01%3A24

Commits

  1. Under has_wal_read_bug, skip recovery/t/032_relfilenode_reuse.pl.

  2. Under has_wal_read_bug, skip contrib/bloom/t/001_wal.pl.

  3. Use Test::Builder::todo_start(), replacing $::TODO.

  4. On sparc64+ext4, suppress test failures from known WAL read failure.

  5. Report any XLogReadRecord() error in XlogReadTwoPhaseData().