Do not set vm error state when raise MigrationError

[Xiang BZ Zhou]

Control Node: 101.0.0.20(also has compute service , but do not use it)
Compute Node: 101.0.0.30

nova version:
2014.1.b2-847-ga891e04

in control node nova.conf
allow_resize_to_same_host = True
and
in compute node nova.conf
allow_resize_to_same_host = False

detail:
1. boot an instance in compute node
nova boot --image 51c4a908-c028-4ce2-bbd1-8b0e15d8d829 --flavor 84 --nic net-id=308840da-6440-4599-923a-2edd290971d3 --availability-zone nova:compute.localdomain migrate_test

2. resize it to flavor type 1
nova resize migrate_test 1

3.the instance has set to error state when resize failed.

#nova list
+--------------------------------------+----------+--------+-------------+-------------+-------------------+
| a1424990-182a-4bc2-8c17-aa4808a49472 | migrate_test | ERROR | resize_prep | Running | private=20.0.0.15 |
+--------------------------------------+----------+--------+-------------+-------------+-------------------+

#nova show
....
| config_drive | |
| created | 2014-06-09T09:31:35Z |
| fault | {"message": "<class 'nova.exception.MigrationError'>", "code": 500, "details": " File \"/opt/stack/nova/nova/compute/manager.py\", line 3104, in prep_resize |
| | node) |
| | File \"/opt/stack/nova/nova/compute/manager.py\", line 3058, in _prep_resize |
| | raise exception.MigrationError(msg) |
| | ", "created": "2014-06-10T03:54:39Z"}
                                                                     |
| flavor | m1.micro (84) |
| hostId | f73013b029032929598a4a54586e4469c2c7cd676c147f6601f73c58
....

error log in compute node:

2014-06-10 11:54:48.372 ERROR nova.compute.manager [req-6a4ac25a-7d24-40c6-9f8d-435b4adb6fff admin admin] [instance: a1424990-182a
-4bc2-8c17-aa4808a49472] Setting instance vm_state to ERROR
2014-06-10 11:54:48.372 TRACE nova.compute.manager [instance: a1424990-182a-4bc2-8c17-aa4808a49472] Traceback (most recent call la
st):
2014-06-10 11:54:48.372 TRACE nova.compute.manager [instance: a1424990-182a-4bc2-8c17-aa4808a49472] File "/opt/stack/nova/nova/c
ompute/manager.py", line 5231, in _error_out_instance_on_exception
2014-06-10 11:54:48.372 TRACE nova.compute.manager [instance: a1424990-182a-4bc2-8c17-aa4808a49472] yield
2014-06-10 11:54:48.372 TRACE nova.compute.manager [instance: a1424990-182a-4bc2-8c17-aa4808a49472] File "/opt/stack/nova/nova/c
ompute/manager.py", line 3111, in prep_resize
2014-06-10 11:54:48.372 TRACE nova.compute.manager [instance: a1424990-182a-4bc2-8c17-aa4808a49472] filter_properties)
2014-06-10 11:54:48.372 TRACE nova.compute.manager [instance: a1424990-182a-4bc2-8c17-aa4808a49472] File "/opt/stack/nova/nova/compute/manager.py", line 3104, in prep_resize
2014-06-10 11:54:48.372 TRACE nova.compute.manager [instance: a1424990-182a-4bc2-8c17-aa4808a49472] node)
2014-06-10 11:54:48.372 TRACE nova.compute.manager [instance: a1424990-182a-4bc2-8c17-aa4808a49472] File "/opt/stack/nova/nova/compute/manager.py", line 3058, in _prep_resize
2014-06-10 11:54:48.372 TRACE nova.compute.manager [instance: a1424990-182a-4bc2-8c17-aa4808a49472] raise exception.MigrationError(msg)
2014-06-10 11:54:48.372 TRACE nova.compute.manager [instance: a1424990-182a-4bc2-8c17-aa4808a49472] MigrationError: destination same as source!
2014-06-10 11:54:48.372 TRACE nova.compute.manager [instance: a1424990-182a-4bc2-8c17-aa4808a49472]

bug reason:
1. nova-scheduler is allowed to scheduler to compute node (due to controller nova.conf)

2. but nova-compute is not allowed to resize in same host (due to compute node nova.conf)

3.
a)compute side _prep_resize() function set instance into error state:
....
self._set_instance_error_state(context, instance['uuid'])
...
then raise exception

b)
compute node reschedule the instance again, failed again
....
self._reschedule_resize_or_reraise(context, image, instance,
     exc_info, instance_type, reservations, request_spec,
     filter_properties)
...
c)compute store instance fault info
....
compute_utils.add_instance_fault_from_exc(context, self.conductor_api,
    instance, exc_info[0], exc_info=exc_info)

additional:
no matter what the scheduler filter is using, instance should not be set to ERROR status just because scheduler doesn't find a appropriate host to do resize.
and we can not deal with vm in ERROR state unless we change it's state in db

[Matt Riedemann (mriedem)]

You should be able to use the reset-state API to get it out of ERROR state, unless it's deleting in which case you've got bug 1299139.

[Davanum Srinivas (DIMS) (dims-v)]

fyi, there's an effort to eliminate the flag - https://review.openstack.org/#/c/118604/

[jichenjc (jichenjc)]

The patch mentioned above are merged , I believe we can close the bug ?