Cinder migration may leave stale volumes

[Lihi Tikolsky]

When performing cinder migration, if the storage connectivity from the cinder node to the storage fails, the volume will not be found as expected.
In that case, looks like cinder attempts a rollback by trying to delete the volume it created for the migration at the destination.

*The issue*
the rollback action does not perform a 'detach' prior to 'delete volume'.
In case of some storage vendors, volume delete will fail because the volume is still attached to the host (at the storage array side).
This leaves stale volumes on the storage arrays at the destination.

*Solution*
Perform 'detach' before attempting to 'delete' the volume

*Log from the migration is attached*
migration starts at the beginning of the log

source vol: f3316fc0-2461-4220-be01-0a3a1b18d2fe
dest host to migrate to: nova3.ps.xiv.ibm.com@IBM-XIV_hostdev31a.xiv.ibm.com_Openstack_fibre_channel#IBM-XIV_hostdev31a.xiv.ibm.com_Openstack_fibre_channel

[Jay Bryant (jsbryant)]

Alon, can you please take a look at this. Appears that it was seen on XIV doing migration.

Thanks!

[Alon Marx (alonma)]

Jay,

I looked at the relevant piece of code and it has changed since the time the bug was reported. This issue may already already be fixed by these changes, but we need to do some testing to verify it.

[Jay Bryant (jsbryant)]

Alon,

Thanks for the analysis and update. Let me know what you find.

[Jay Bryant (jsbryant)]

Alon, any update on this?

[Alon Marx (alonma)]

Hi Jay,

I wasn't able to reproduce it. Can we close it as non-reproducable?

Alon

[Jay Bryant (jsbryant)]

Marking this as invalid given that it cannot be reproduced.