Retype fails with "no valid host was found"

[Pedro Magalhães]

I have two backends pointing to the same storage, one with no limits, named delleqlx, and other with QoS frontend, named delleqlx2, allowing only 20MB/s disk limit (total_bytes_sec=20000000).

If I create a volume directly in delleqlx or delleqlx2 everything works fine, including in the delleqlx2 that inputs iotune in the XML of the instance.

The problem is when I create a volume in delleqlx, I shutted down the instance and then trying to retype the volume and get the following error.

2017-02-07 15:49:47.295 5012 ERROR cinder.scheduler.manager [req-d4d2867d-0a2d-448b-86f1-5485fea0d6a3 1063d72bd47747e9b0c23295b52eba57 ee93125f51da4a7a93b88380e61bf697 - default default] Failed to schedule_retype: No valid host was found. Current host not valid for volume 1d9004d0-f606-4cfa-bf05-8ffb308edb63 with type {u'name': u'delleqlx2', u'qos_specs_id': u'5c717946-4fca-473a-acaa-9d39136cab0b', u'deleted': False, u'created_at': u'2017-02-07T10:42:06.000000', u'updated_at': u'2017-02-07T10:44:55.000000', u'extra_specs': {u'volume_backend_name': u'limited_disk'}, u'is_public': False, u'deleted_at': None, u'id': u'efa27d5c-0a1a-4d05-b9ef-9fe3f6eb9d24', u'description': u''}, migration not allowed

P.s.-> The volume is in available mode.

[Sean McGinnis (sean-mcginnis)]

In-use isn't supported as the error says. If unattached does not error, but does not actually retype, we will need to see the log messages for that request to see what is happening.

[Pedro Magalhães (pjm6)]

My mistake, I edited the bug description.

[Sean McGinnis (sean-mcginnis)]

Log errors:

2017-02-07 15:49:47.284 5012 WARNING cinder.scheduler.host_manager [req-d4d2867d-0a2d-448b-86f1-5485fea0d6a3 1063d72bd47747e9b0c23295b52eba57 ee93125f51da4a7a93b88380e61bf697 - default default] volume service is down. (host: os-storage01@lvm)
2017-02-07 15:49:47.295 5012 ERROR cinder.scheduler.manager [req-d4d2867d-0a2d-448b-86f1-5485fea0d6a3 1063d72bd47747e9b0c23295b52eba57 ee93125f51da4a7a93b88380e61bf697 - default default] Failed to schedule_retype: No valid host was found. Current host not valid for volume 1d9004d0-f606-4cfa-bf05-8ffb308edb63 with type {u'name': u'delleqlx2', u'qos_specs_id': u'5c717946-4fca-473a-acaa-9d39136cab0b', u'deleted': False, u'created_at': u'2017-02-07T10:42:06.000000', u'updated_at': u'2017-02-07T10:44:55.000000', u'extra_specs': {u'volume_backend_name': u'limited_disk'}, u'is_public': False, u'deleted_at': None, u'id': u'efa27d5c-0a1a-4d05-b9ef-9fe3f6eb9d24', u'description': u''}, migration not allowed

[Sean McGinnis (sean-mcginnis)]

Service status:

+------------------+-----------------------------------------------------+------+----------+-------+----------------------------+-----------------+
| Binary | Host | Zone | Status | State | Updated_at | Disabled Reason |
+------------------+-----------------------------------------------------+------+----------+-------+----------------------------+-----------------+
| cinder-scheduler | os-controller01-cinder-scheduler-container-c5068e79 | nova | enabled | up | 2017-02-07T15:57:46.000000 | - |
| cinder-scheduler | os-controller02-cinder-scheduler-container-1921976c | nova | enabled | up | 2017-02-07T15:57:42.000000 | - |
| cinder-scheduler | os-controller03-cinder-scheduler-container-dc95afe8 | nova | enabled | up | 2017-02-07T15:57:44.000000 | - |
| cinder-volume | os-storage01@delleqlx | nova | enabled | up | 2017-02-07T15:57:44.000000 | - |
| cinder-volume | os-storage01@delleqlx2 | nova | enabled | up | 2017-02-07T15:57:45.000000 | - |
| cinder-volume | os-storage01@lvm | nova | disabled | down | 2017-02-07T15:57:39.000000 | - |
+------------------+-----------------------------------------------------+------+----------+-------+----------------------------+-----------------+

[Sean McGinnis (sean-mcginnis)]

Looks like this is expected behavior of retype needing to switch backends, but migrate on demand not set.

The intent was to use the same backend, but have two different QoS profiles. Currently setting up two different volume types that use the same backend to verify QoS profiles can be associated with each volume type and retype can be done without moving all of the volume data.

Will close once we validate that resolves the issue.

[Sean McGinnis (sean-mcginnis)]

Retype between volume types of the same backend works. Closing this out.