On Thu, Sep 13, 2012 at 07:29:44AM +1000, NeilBrown wrote:
> Hi Marc,
>
> md registers a reboot notifier. When that is called it tries to checkpoint
> everything.
> All varieties of the 'reboot' system call seem to call the reboot notifiers.
> alt-sysrq-b doesn't use the same path. It calls machine_emergency_restart,
> bypassing all the reboot handling.
>
> Once upon a time I had the idea that killing the md threads would lead to
> proper checkpointing, so alt-sysrq-I would do-the-right-thing.
> I'm not sure if it does though.
>
> But alt-sysrq-o (power off) seems to use the normal reboot handling and so
> works - as you noticed. So that should always be safe and seems to be the
> only safe approach.
Thank you for confirming what I thought I saw.
For my specific case, I'll be ok because I have a controllable power supply,
but obviously for the hosted server case, this sucks since sending a
poweroff makes sure the machine won't come back.
Is there a chance you can contact whoever is responsible for the sysrq-b
codepath to make the right call so that you get a chance to sync arrays, or
should admins just know to use sysrq-i and then sysrq-r ?
Thanks,
Marc
--
"A mouse is a device used to point at the xterm you want to type in" - A.S.R.
Microsoft is to operating systems ....
.... what McDonalds is to gourmet cooking
Home page: http://marc.merlins.org/
Attachment:
signature.asc
Description: Digital signature
