Rebooting instance doesn't restore mounted volume

Hi Tushar,

You use KVM on Ubuntu, right?
Also, you rebooted the instance on the guest OS?
I mean, not using euca-reboot-instances.

I'm feeling that the root cause of this issue would be a KVM problem.
If this issue is reproducible, please check collect information before and after rebooting instances on compute node.

  # virsh dumpxml VM_NAME

I guess the device you attached to your VM vanished from VM configuration after guest OS reboot.
In the case, what we can do anyway would be logging the exeption and cleaning up database, I think.

-Masanori

BTW, if you used euca-reboot-instances on KVM based systems, the issue gets back to nova side.
At this moment, libvirt does not support rebooting KVM instances, and the current implementation of
RebootInstance is like the following.

  trunk/nova/virt/libvirt_conn.py
    473 def reboot(self, instance):
    474 self.destroy(instance, False) # DESTROY ONCE
    475 xml = self.to_xml(instance)

One idea could be calling virsh dumpxml to the instance to be rebooted and updating the above xml here.

    476 self.firewall_driver.setup_basic_filtering(instance)
    477 self.firewall_driver.prepare_instance_filter(instance)
    478 self._conn.createXML(xml, 0) # CREATE AGAIN, AND THERE IS NO CODE TO RE-ATTACH EBSs.
    479 self.firewall_driver.apply_instance_filter(instance)
    480
    481 timer = utils.LoopingCall(f=None)
    482
    483 def _wait_for_reboot():
    484 try:
    485 state = self.get_info(instance['name'])['state']
    486 db.instance_set_state(context.get_admin_context(),
    487 instance['id'], state)
    488 if state == power_state.RUNNING:
    489 LOG.debug(_('instance %s: rebooted'), instance['name'])
    490 timer.stop()
    491 except Exception, exn:
    492 LOG.exception(_('_wait_for_reboot failed: %s'), exn)
    493 db.instance_set_state(context.get_admin_context(),
    494 instance['id'],
    495 power_state.SHUTDOWN)
    496 timer.stop()
    497
    498 timer.f = _wait_for_reboot
    499 return timer.start(interval=0.5, now=True)

-Masanori

Hi,

I wrote an ultimately STUPID workaround fix on this issue and linked my branch here.
Actually, this issue is a problematic one if we want to resolve it in an elegant way, I think.

Anyway, it looks working on my Ubuntu 10.10 box at least.

Tushar, if you are testing volume driver other than ISCSI or multiple multi-node nova installation,
could you try the patch below?

  lp:~itoumsn/nova/lp747922

Thanks,

I will volunteer as the assignee of this issue until someone with more elegant resolution appears...

-Masanori

I tested using your branch lp:~itoumsn/nova/lp747922 and it seems to be working as expected now.

After rebooting the instance, I see the volume is still attached and I can see all files intact.

Thank you.

Hi Tushar,

Thanks for testing. :)
I will post a merge request soon after the cactus release.

Thanks,
Masanori

pardon me, i'm a newbie here, how can i apply above patch into my existing cloud? :)