Horton

Re: [BUG] non archived WAL removed during production crash recovery

Michael Paquier <michael@paquier.xyz>

View thread View on postgr.es
From: Michael Paquier <michael@paquier.xyz>
To: Jehan-Guillaume de Rorthais <jgdr@dalibo.com>
Cc: Kyotaro Horiguchi <horikyota.ntt@gmail.com>, masao.fujii@oss.nttdata.com, pgsql-bugs@lists.postgresql.org
Date: 2020-04-21T02:15:01Z
Lists: pgsql-bugs, pgsql-hackers 
On Mon, Apr 20, 2020 at 02:22:35PM +0200, Jehan-Guillaume de Rorthais wrote:
> The problem is that we would have to read the controldata file each time we
> wonder if a segment should be archived/removed. Moreover, the controldata
> file might not be in sync quickly enough with the real state for some other
> code path or futur needs.

I don't think that this is what Horiguchi-san meant here.  What I got
from his previous message would be to be to copy the shared value from
the control file when necessary, and have the shared state use only a
subset of the existing values of DBState, aka:
- DB_IN_CRASH_RECOVERY
- DB_IN_ARCHIVE_RECOVERY
- DB_IN_PRODUCTION
Still, that sounds wrong to me because then somebody would be tempted
to change the shared value thinking that things like DB_SHUTDOWNING,
DB_SHUTDOWNED_* or DB_STARTUP are valid but we don't want that here.
Note that there may be a case for DB_STARTUP to be used in
XLOGShmemInit(), but I'd rather let the code use the safest default,
DB_IN_CRASH_RECOVERY to control that we won't remove .ready files by
default until the startup process sees fit to do the actual switch
depending on the checkpoint record lookup, if archive recovery was
actually requested, etc.

> Indeed, Benoît Lobréau reported this behavior to me.

Noted.  Thanks for the information.  I don't think that I have ever
met Benoît in person, do I?  Tell him that I owe him one beer or a
beverage of his choice when we meet IRL, and that he had better use
this message-id to make me keep my promise :)
--
Michael

Commits

  1. Fix interpolation in test name.

  2. Add more TAP coverage for archive status with crash recovery of standbys

  3. Fix handling of WAL segments ready to be archived during crash recovery

  4. Fix WAL recycling on standbys depending on archive_mode