Re: Suspicious test failure - mdmon misses recovery events on loop devices

[NeilBrown <neilb@suse.de>]

[-- Attachment #1: Type: text/plain, Size: 1259 bytes --]

On Mon, 29 Jul 2013 22:42:25 +0200 Martin Wilck <mwilck@arcor.de> wrote:

> 
> > My current idea to solve this is yet another separate thread just for
> > monitoring kernel state changes. Don't have it ready yet, though.
> 
> Another idea would be in manage_member, after queueing the metadata
> update and waking up the monitor, to wait for the metadata to finish
> processing before actually starting the recovery (writing "recover" to
> sync_action).
> 
> Martin

I hope an extra thread won't be necessary :-)

I think that manage_member is the place to fix this.  However it might be
even simpler than you suggest.

We currently have

		replace_array(container, a, newa);
		sysfs_set_str(&a->info, NULL, "sync_action", "recover");

monitor subsequently takes that 'newa', looks at 'sync_action', see that it
is 'idle' and assume that the recover never happened.
Suppose we change it to:

		if (sysfs_set_str(&a->info, NULL, "sync_action", "recover") == 0)
		        newa->prev_action = newa->curr_action = recovery;
		replace_array(container, a, newa);

Then it wouldn't matter if monitor never saw the 'recovery' state as manager
explicitly told it that recovery had started.

Could you try that?

Thanks,
NeilBrown

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 828 bytes --]