volume migration requires #xxx part in destination host specification

[Yuriy Taraday]

When I run "cinder service-list", I get "Host" column with names like "host1@lvmdriver-1", but Cinder's scheduler selects host to migrate by HostState's "host" attribute which also includes "#lvmdriver-1" part.

We should either explicitly state that we need "#lvmdriver-1" part in host specification and add it to "cinder service-list" output or ignore these fragments in scheduler.

Steps to reproduce:
$ cinder create --name vol1 1
$ cinder migrate vol1 somehost@lvmdriver-1

Expected result:
Migration gets scheduled to somehost

Actual result:
In nova-scheduler logs:
Failed to schedule_migrate_volume_to_host: No valid host was found. Cannot place volume xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx on somehost@lvmdriver-1

[John Griffith (john-griffith)]

I assume you meant s/nova-scheduler/cinder-scheduler/

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/143234

[Nikesh Kumar Mahalka (nikeshmahalka)]

$ cinder extra-specs-list
+--------------------------------------+---------+------------------------------------------+
| ID | Name | extra_specs |
+--------------------------------------+---------+------------------------------------------+
| 4b8217ed-2320-4118-924a-472a01be9fe4 | lvm-4 | {u'volume_backend_name': u'lvmdriver-4'} |
| ae3b8e2b-6913-4116-88ef-59519a1ba2a9 | lvm-3 | {u'volume_backend_name': u'lvmdriver-3'} |
+--------------------------------------+---------+------------------------------------------+

$ cinder service-list
+------------------+---------------------------------+------+---------+-------+----------------------------+-----------------+
| Binary | Host | Zone | Status | State | Updated_at | Disabled Reason |
+------------------+---------------------------------+------+---------+-------+----------------------------+-----------------+
| cinder-scheduler | juno-devstack-controller | nova | enabled | up | 2014-12-20T18:05:40.000000 | None |
| cinder-volume | juno-devstack-block@lvmdriver-3 | nova | enabled | up | 2014-12-20T18:06:04.000000 | None |
| cinder-volume | juno-devstack-block@lvmdriver-4 | nova | enabled | up | 2014-12-20T18:06:04.000000 | None |
+------------------+---------------------------------+------+---------+-------+----------------------------+-----------------+

when i am creating a volume without volume type(default volume host for me is juno-devstack-block@lvmdriver-3#lvmdriver-3 and default volume type for me is lvm-3) and trying migration to volume host juno-devstack-block@lvmdriver-4#lvmdriver-4,it is working for me

But when i am creating a volume with volume type lvm-3 or lvm-4 and trying to migrate to other volume host,it is failing by showing " No valid host was found. Cannot place volume xxxxx on juno-devstack-block@lvmdriver-x#lvmdriver-x" in cinder-scheduler log.

[Mike Perez (thingee)]

As mentioned in the discussion in the patch review, this is by design. Release note update to follow.

[Yuriy Taraday (yorik-sar)]

I think this bug should be fixed in one of ways:
- either code should be changed to make 'cider migrate' accept what 'cinder service-list' provides;
- or 'cinder service-list' should include #xxx pool part in its output;
- or documentation should be changed accordingly to allow user to determine how to migrate volumes.

Otherwise in current state I (with user hat on) can't possibly migrate volumes without looking into debug logs and code with pdb. This makes this bug valid.

[Yuriy Taraday (yorik-sar)]

Note that for last option to work we need some extra API and CLI calls to determine list of available pools.

[Mitsuhiro Tanino (mitsuhiro-tanino)]

>>- or 'cinder service-list' should include #xxx pool part in its output;

An API which shows pool information was already suppoted at commit https://review.openstack.org/#/c/119938/.
For more detail, please see the commit. However, this API is not in API guide, so the document should be updated.
(http://developer.openstack.org/api-ref-compute-v2.html)

Also I will try to add CLI command "cinder get-pools" which shows pool information using the API which was added commit 119938.
I think this CLI also closes the gap for end users.

[Mitsuhiro Tanino (mitsuhiro-tanino)]

Fix proposed.
Review: https://review.openstack.org/#/c/144814/

[Mitsuhiro Tanino (mitsuhiro-tanino)]

Hi Yuriy,

These fixes are proposed to solve this bug.

   (a) Allow the w/o pool format for "migrate" and "manage" commands
       https://review.openstack.org/#/c/143234/

   (b) Modify help docstring of "cinder migrate". The #pool is not mentioned in the help doc string.
       https://review.openstack.org/#/c/145064/

   (c) Add "cinder get-pools" command(for Admin only) using API.
       https://review.openstack.org/#/c/144814/

   (d) Documentation
        https://wiki.openstack.org/wiki/ReleaseNotes/Juno#Limitations.2FKnown_Issues

Please feel free to comment for these fixes.

[OpenStack Infra (hudson-openstack)]

Change abandoned by Mike Perez (<email address hidden>) on branch: master
Review: https://review.openstack.org/143234
Reason: inactive for a month

[Mitsuhiro Tanino (mitsuhiro-tanino)]

I believe these fixes solved this issue. If you still have a problem, please reopen this.

(a) Add "cinder get-pools" command(for Admin only) using API.
    https://review.openstack.org/#/c/144814/

(b) Expose cinder's scheduler pool API
    https://review.openstack.org/#/c/140142/

(c) Documentation
    https://wiki.openstack.org/wiki/ReleaseNotes/Juno#Limitations.2FKnown_Issues