Comment 9 for bug 1334345

Noticing this in the trace of that unit test failure:

libvirt: error : internal error: could not initialize domain event timer
...
WARNING [nova.compute.manager] Instance is not stopped. Calling the stop API.

It blows up when trying to (re)start the instance in cloud.py:

http://git.openstack.org/cgit/openstack/nova/tree/nova/api/ec2/cloud.py#n1746

Note there is a TODO comment in the ec2 create_image code:

    # TODO(yamahata): race condition
    # At the moment there is no way to prevent others from
    # manipulating instances/volumes/snapshots.
    # As other code doesn't take it into consideration, here we don't
    # care of it for now. Ostrich algorithm

By ostrich I'm assuming put our heads in the sand. :(

Looking at the code, it stops the instance and then waits for it to be stopped, there is a hard-coded 1 hour timeout in the code and then when it's vm_state is stopped it moves on to creating the snapshot. Once the snapshot is done it goes to start the instance and that's where we hit the invalid task_state failure.

e-r query here: https://review.openstack.org/#/c/103119/