When volume are detached for example for a live migration, faulty devices are left on compute nodes. 

Must I configure iscsi initiator on both compute and controllers nodes?

At this time I configured it only on compute nodes.

Faulty could be an os brick module issue ?

Thanks 

Ignazio