Horton

Re: Race condition in recovery?

Dilip Kumar <dilipbalaut@gmail.com>

View thread View on postgr.es
From: Dilip Kumar <dilipbalaut@gmail.com>
To: Kyotaro Horiguchi <horikyota.ntt@gmail.com>
Cc: Robert Haas <robertmhaas@gmail.com>, PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>, Heikki Linnakangas <hlinnaka@iki.fi>
Date: 2021-05-27T07:17:30Z
Lists: pgsql-hackers 
On Thu, May 27, 2021 at 12:09 PM Kyotaro Horiguchi
<horikyota.ntt@gmail.com> wrote:
>
> At Thu, 27 May 2021 11:44:47 +0530, Dilip Kumar <dilipbalaut@gmail.com> wrote in
> > Maybe we can somehow achieve that without a broken archive command,
> > but I am not sure how it is enough to just delete WAL from pg_wal?  I
> > mean my original case was that
> > 1. Got the new history file from the archive but did not get the WAL
> > file yet which contains the checkpoint after TL switch
> > 2. So the standby2 try to stream using new primary using old TL and
> > set the wrong TL in expectedTLEs
> >
> > But if you are not doing anything to stop archiving WAL files or to
> > guarantee that WAL has come to archive and you deleted those then I am
> > not sure how we are reproducing the original problem.
>
> Thanks for the reply!
>
> We're writing at the very beginning of the switching segment at the
> promotion time. So it is guaranteed that the first segment of the
> newer timline won't be archived until the rest almost 16MB in the
> segment is consumed or someone explicitly causes a segment switch
> (including archive timeout).

I agree

> > BTW, I have also tested your script and I found below log, which shows
> > that standby2 is successfully able to select the timeline2 so it is
> > not reproducing the issue.  Am I missing something?
>
> standby_2? My last one 026_timeline_issue_2.pl doesn't use that name
> and uses "standby_1 and "cascade".  In the ealier ones, standby_4 and
> 5 (or 3 and 4 in the later versions) are used in ths additional tests.
>
> So I think it shold be something different?

Yeah, I tested with your patch where you had a different test case,
with "026_timeline_issue_2.pl", I am able to reproduce the issue.

-- 
Regards,
Dilip Kumar
EnterpriseDB: http://www.enterprisedb.com

Commits

  1. Further refinement of stuck_on_old_timeline recovery test

  2. Adjust new test case to set wal_keep_segments.

  3. Adjust new test case to set wal_keep_size.

  4. Fix corner case failure of new standby to follow new primary.

  5. Back-port a few PostgresNode.pm methods.

  6. Allow PostgresNode.pm's backup method to accept backup_options.

  7. Add end-to-end testing of pg_basebackup's tar-format output.

  8. Allow PostgresNode.pm tests to wait for catchup

  9. Delay reading timeline history file until it's fetched from master.