Looking for a strategy to have Cinder manage local storage on a per-compute node basis. Ideally, Cinder would not attach volumes over iSCSI since logical volumes would be local.

The current operating assumption is having each nova compute node tied to a matching configured cinder availability zone (volume_group). Volume creation then takes place on e.g availability zone cinder-compute01 where, as part of a host aggregate, compute01 node is a member of.

This works to the extent that volumes are being attached locally but over iSCSI on instance creation. This feels suboptimal and introduces a point of failure that simply is not needed.

End goal is to leverage SSD local storage on all compute nodes.

Running Mitaka on CentOS 7.