Cinder Volume Backend Migration in HA Scenario

In cases of Highly Available storage backends, a cinder volume should be easily managed by multiple cinder nodes. However, cinder will associate this volume with one cinder node. If this cinder node goes down, the volume is no longer manageable.

We should be able to reassign the volume to another cinder node, without using Cinder Migrate so that this volume can be manageable via another cinder node.
This should only be relevant to HA storage backends such as SolidFire, Ceph, Netapp, Gluster, Scale.IO

This issue will also prevent HA of VMs which are attached to volumes on a cinder node that goes down. See the following scenario:

*Cinder Node01*
- VOL-01 (Attached to VM-01)
- VM-01
- VM-02

*Non-Cinder Node02*

Cinder Node01 goes down, VMHA begins and evacuates VMs to Node02. VM-02 will evacuate cleanly. VM-01 will fail to evacuate, because as part of the migration/evacuation workflow it will attempt to detach the volume, which will fail because Node01 is down. VM-01 will be left on Node01.
  • Guest
  • Jun 3 2019
  • Attach files