cinder retype in-use volume hangs volume access
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
Cinder |
New
|
Undecided
|
Unassigned |
Bug Description
CentOS deployment using Kolla Ansible 7.0.2 (Rocky)
Goal:
to migrate an "in-use" volume, whether it be the boot volume or not, to a different volume type
Observations:
cinder re-type --migration-policy on-demand <volume> <volume type> works fine when the volume is in status "available"
The same command appears to work while the volume is attached to a running VM, however, when the migration finishes, the volume becomes unavailable in the VM, cannot be removed from the VM (using openstack server remove volume), and leaves the server in a bad state (it can be stopped, but not powered on again - task state is powering-on, but vm_state is stopped and power_state is shutdown).
Running "openstack server remove volume" does not return an error, but the volume is not removed. Note that this is the volume whose volume type was changed.
This test was done with a few Windows VMs, but I can only imagine this happens with any operating system.
Eric
I just noticed that I didn't specify that I was using iSCSI-connected volumes with backend qos specs (not specified in OpenStack, rather the backend storage platform performs all qos).
We have multiple volume types with Ceph, but they have differing front-end qos specs, which is unsupported by cinder retype. So I am unable to test retyping Ceph volumes that are attached to a running VM - which is where the original problem exists.
Eric