Instance destroy and Instance snapshot are locking operations, meaning
that they can only occur sequentially. This is the result of the commit
6d8903b5d9609305d63aba0067a032a66f5ce3ee
The problem is that the destroy instance can occur before snapshoting,
resulting in a VM NotFound exception being raised while snapshoting, as
the VM no longer exists. In the logs it can be observed that the lock was
being held by "do_terminate_instance", which was then aquired by
"instance_synchronized_snapshot".
Nova ComputeManager expects an InstanceNotFound exception during the
snapshot operation.
Changes exception raised by VMUtils._check_lookup_vm to InstanceNotFound.
Change-Id: I3e4d4f1874c6ee4b410e8bda7412d49f079b5b8f
Closes-Bug: #1486458
6e4fdbb122