The real problem is that cinder-volume service running on a controller node without cinder VG still gets scheduled by cinder-scheduler, which causes volume creation to fail. Making sure cinder-volume is stopped on controllers is a band-aid around that Cinder bug.
The real problem is that cinder-volume service running on a controller node without cinder VG still gets scheduled by cinder-scheduler, which causes volume creation to fail. Making sure cinder-volume is stopped on controllers is a band-aid around that Cinder bug.