Using reset for hard_reboot is not reliable

[Rafi Khardalian]

Using reset for hard_reboot is not reliable, even where it is supported by libvirt. Hard reboots are one of the only ways to recover a VM in a broken state. The reset command assumes the domain is running in some capacity and will fail if it is not. Here are some steps to reproduce:

1. Create a new libvirt VM (using qemu for my testing).

2. virsh list # validate it is running

virsh # list
 Id Name State
----------------------------------
  3 instance-00000001 running

3. Find and kill -9 the pid of the qemu/kvm process. virsh list --all to confirm:

virsh # list --all
 Id Name State
----------------------------------
  - instance-00000001 shut off

4. Issue a virsh reset, as the code would do:

virsh # reset instance-00000001
error: Failed to reset domain instance-00000001
error: Requested operation is not valid: domain is not running

There is no way to recover this VM without manual intervention. Reverting to the hold behavior, by commenting out the conditional and forcing the code below works much more reliably:

            self._destroy(instance)
            self._create_domain(xml, virt_dom)

Hard reset is the current sledgehammer for fixing issues and it really needs to stay that way.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/11371

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/11371
Committed: http://github.com/openstack/nova/commit/2b7d5c783a330fbf5a54cd5f63dbf5f1004c6103
Submitter: Jenkins
Branch: master

commit 2b7d5c783a330fbf5a54cd5f63dbf5f1004c6103
Author: Rafi Khardalian <email address hidden>
Date: Tue Aug 14 20:41:23 2012 +0000

    Revert to prior method of executing a libvirt hard_reboot.

    Fixes bug 1036826.

    Using reset for hard_reboot is not reliable, even where it is supported
    by libvirt. Hard reboots are one of the only ways to recover a VM in a
    broken state. The reset command assumes the domain is running in some
    capacity and will fail if it is not.

    Hard reset is the current sledgehammer for fixing issues and it really
    needs to stay that way.

    Change-Id: I84705b72d79cf71adad066b18267fdfb199bc9cb