5:32 p.m.
Did you attempt to start a VM with this disk and it failed, or you
didn't try at all? If it's the latter then the error is strange...
If it's the former there is a known issue with multipath at the
moment, see[1] for a workaround, since you might have issues with
detaching volumes which later, because multipath grabs the rbd devices
which would fail `rbd unmap`, it will be fixed soon by automatically
blacklisting rbd in multipath configuration.
Regarding editing, you can submit an RFE for this, but it is currently
not possible. The options are indeed to either recreate the storage
domain or edit the database table
[1] https://bugzilla.redhat.com/show_bug.cgi?id=1881832#c8
On Wed, Oct 14, 2020 at 3:40 PM Michael Thomas <wart(a)caltech.edu> wrote:
On 10/14/20 3:30 AM, Benny Zlotnik wrote:
> Jeff is right, it's a limitation of kernel rbd, the recommendation is
> to add `rbd default features = 3` to the configuration. I think there
> are plans to support rbd-nbd in cinderlib which would allow using
> additional features, but I'm not aware of anything concrete.
>
> Additionally, the path for the cinderlib log is
> /var/log/ovirt-engine/cinderlib/cinderlib.log, the error in this case
> would appear in the vdsm.log on the relevant host, and would look
> something like "RBD image feature set mismatch. You can disable
> features unsupported by the kernel with 'rbd feature disable'"
Thanks for the pointer! Indeed,
/var/log/ovirt-engine/cinderlib/cinderlib.log has the errors that I was
looking for. In this case, it was a user error entering the RBDDriver
options:
2020-10-13 15:15:25,640 - cinderlib.cinderlib - WARNING - Unknown config
option use_multipath_for_xfer
...it should have been 'use_multipath_for_image_xfer'.
Now my attempts to fix it are failing... If I go to 'Storage -> Storage
Domains -> Manage Domain', all driver options are unedittable except for
'Name'.
Then I thought that maybe I can't edit the driver options while a disk
still exists, so I tried removing the one disk in this domain. But even
after multiple attempts, it still fails with:
2020-10-14 07:26:31,340 - cinder.volume.drivers.rbd - INFO - volume
volume-5419640e-445f-4b3f-a29d-b316ad031b7a no longer exists in backend
2020-10-14 07:26:31,353 - cinderlib-client - ERROR - Failure occurred
when trying to run command 'delete_volume': (psycopg2.IntegrityError)
update or delete on table "volumes" violates foreign key constraint
"volume_attachment_volume_id_fkey" on table "volume_attachment"
DETAIL: Key (id)=(5419640e-445f-4b3f-a29d-b316ad031b7a) is still
referenced from table "volume_attachment".
See https://pastebin.com/KwN1Vzsp for the full log entries related to
this removal.
It's not lying, the volume no longer exists in the rbd pool, but the
cinder database still thinks it's attached, even though I was never able
to get it to attach to a VM.
What are my options for cleaning up this stale disk in the cinder database?
How can I update the driver options in my storage domain (deleting and
recreating the domain is acceptable, if possible)?
--Mike