Unless you don't run cinder-volume + cinder-three-par on all compute nodes... Which turns out to be my case (running 52 instances of the 3par driver was overloading the storage controller and that is not really necessary, just a few are enough).
Unless you don't run cinder-volume + cinder-three-par on all compute nodes... Which turns out to be my case (running 52 instances of the 3par driver was overloading the storage controller and that is not really necessary, just a few are enough).