Cinder fails to backup/snapshot/clone/extend volumes when pool is full

[Walt Boring]

Cinder isn't very 'cloud' like with respect to volume actions being taken on volumes when the pool that the source volume is on is full. Cinder should handle this case when there are other pools in the same backend that have space available.

The problem with all this is that a customer using an OpenStack cloud has no way to resolve this issue without involving the deployer/admin of the cinder deployment.

In a very large cloud deployment, this manual intervention becomes very cumbersome. This is a terrible experience for customers using OpenStack. Cinder actually has space available in the same backend, but since the pool is full all operations fail.

Volumes should automatically get migrated if the operation fails due to insufficient space in the existing pool.

The following actions should just work.
extend volume
backup (which typically requires a snapshot/clone being taken for attached volumes prior to bits being copied to the backend backup storage location.
clone volume
snapshot volume - some backends do a full cone as a snapshot as their only option.

[Sofia Enriquez (lsofia-enriquez)]

This bug was discussed on the Antelope PTG and it's going to be treated as an RFE. Walt is going to write a spec for the extended case (it can be updated later when other operations are also ready).

https://etherpad.opendev.org/p/antelope-ptg-cinder#L242