[long] Recovery handling in TXN branch

[Matthias Andree]

David Relson <relson at osagesoftware.com> writes:

> On Sun, 15 Aug 2004 00:25:47 +0200
> Matthias Andree wrote:
>
> ...[snip]...
>
>> Practically, such deadlock situations as I have observed have always
>> happened across as hasty reboot so I think this is the common cause
>> for such deadlocks.
>
> The mention of reboots makes me think of how fsck gets run.  rc.sysinit
> creates a /.autofsck file after the system is up.  As part of normal
> shutdown/reboot, the file is removed.  If the file is present when
> rc.sysinit starts, the system wasn't shutdown cleanly and fsck needs to
> be run.

I was under the impression that ext2, ext3, jfs, xfs, ufs and reiserfs
track "needs fsck" (dirty) or "needs recovery" themselves.

> Would a "needs.recovery" file be of value?

It would be of very limited value.

We need to:
- detect a process has exited uncleanly QUICKLY so the other processes
  on the same data base don't lock dead.
- abort the other processes, so we need a reliable way to determine
  which PIDs to SIGTERM without aborting non-bogofilter processes
- detect reboot

I believe the complex scheme I've described handles all that. Your
bogolock parent process might help with the /quick/ detection and might
allow us to simplify the whole setup a tiny bit.

-- 
Matthias Andree

NOTE YOU WILL NOT RECEIVE MY MAIL IF YOU'RE USING SPF!
Encrypted mail welcome: my GnuPG key ID is 0x052E7D95 (PGP/MIME preferred)