vm_state inaccurate when operate virsh command

[Eric Xie]

"vm_state" of instance is inaccurate when user operate "virsh pause DOM" command.

One case:
1. Use "nova pause INSTANCE", and get:
| e4ce6895-a0db-4b43-9483-c4b8401fdd1c | instance_test | PAUSED | - | Paused | |
2. Use "virsh pause DOM", and get:
| e4ce6895-a0db-4b43-9483-c4b8401fdd1c | instance_test | ACTIVE | - | Paused | |

Another case:
1. nova pause INSTANCE;
| e4ce6895-a0db-4b43-9483-c4b8401fdd1c | instance_test | PAUSED | - | Paused | |
2. virsh resume INSTANCE;
| e4ce6895-a0db-4b43-9483-c4b8401fdd1c | instance_test | PAUSED | - | Running | |

[Eric Xie (mark-xiett)]

I check the related code:
nova.compute.manager.py
_sync_instance_power_state()
reason of case 1:
elif vm_state == vm_states.ACTIVE:
            elif vm_power_state == power_state.PAUSED:
                # Note(maoy): a VM may get into the paused state not only
                # because the user request via API calls, but also
                # due to (temporary) external instrumentations.
                # Before the virt layer can reliably report the reason,
                # we simply ignore the state discrepancy. In many cases,
                # the VM state will go back to running after the external
                # instrumentation is done. See bug 1097806 for details.
                LOG.warn(_("Instance is paused unexpectedly. Ignore."),
                         instance=db_instance)
reason of case 2: (Why not process power_state 'PAUSED')
        elif vm_state == vm_states.PAUSED:
            if vm_power_state in (power_state.SHUTDOWN,
                                  power_state.CRASHED):
                self.heal_instance_state(context, db_instance,
                    vm_state,vm_power_state,default_action = instance_actions.FORCE_STOP)

[Sean Dague (sdague)]

This is working as designed. virsh commands outside of openstack are not supported.