Re: Stop IMSM verify?

On Wed, 12 Sep 2012 18:34:41 -0700 Daniel Frey [off-list ref] wrote:

On Wed, Sep 12, 2012 at 6:22 PM, NeilBrown [off-list ref] wrote:

quoted

On Wed, 12 Sep 2012 18:00:21 -0700 Daniel Frey [off-list ref] wrote:

quoted

Is there any way to stop the verify of an IMSM raid through mdadm?

No.

Darn.

quoted

quoted

I found https://raid.wiki.kernel.org/index.php/RAID_Administration,
but the imsm raid does not seem to have sync_action anywhere in the
/sys/block tree.

It certainly should be there.
  find /sys/block -name sync_action
??

I'm using mdadm exclusively for my imsm raid, and:

 ~ # find /sys/block -name sync_action
 ~ #

Of course... it only exists in that tree via a symlink that find won't follow.

% find /sys/devices/ -name sync_action

will work.

I take it the imsm raid should have that? I used that exact same
command (and found nothing) before I posted to the list.

quoted

quoted

I've been having issues with suspend which I'm trying to troubleshoot,
and I know I can stop it if I boot into Windows. Just wondering if
that functionality is in mdadm.

I suspect your problem is that mdmon is being killed at an awkward time
during shutdown.
What distro?  Is 'systemd' being used?

The distro is gentoo (no systemd) - suspend works (as in it goes into
suspend properly) but 8 out of 10 times it fails to wake out of
suspend by either the keyboard or power button. The only way to "wake"
it is to yank the power plug from the PSU for a few seconds and plug
it in. As you can imagine, the imsm raid isn't too happy about that...
it boots in Verify mode. I've found I can stop the raid verify in
Windows, so I thought perhaps mdadm could do the same.

I haven't quite figured the suspend issue out yet. It could be
hardware (thinking maybe PSU?) but I'm trying to troubleshoot this
issue methodologically. It started after an upgrade to a kernel >
3.0.2, I have yet to try a 2.6.x to see if it goes back to normal.
When it does resume properly, everything is fine (mdadm doesn't
resync.) It's related to the sleep of death (or eternal sleep?) issue.

I guess it would be nice if we could somehow arrange that the array were
marked 'clean' just before suspend.... it looks like that should be possible.
'suspend notifiers' are called after the 'sync' and before user-space is
frozen.  I could maybe get md to register a 'suspend notifier' which poked
mdmon somehow so that it would mark the array as clean.  It might not even be
that hard.
If I were to create a patch against mainline (or maybe some earlier kernel if
you really wanted), would you be able/willing to test it out?

NeilBrown