With the introduction of the NodeStageVolume call in CSI v0.2, we might want to introduce a filesystem/csi isolator to perform per-workload volume publish, and make Slave::publishResources() perform the per-node publish (staging).
The benefit is that we can support NFS through this isolator.
The major problems is that both SLRP and this isolator need to talk to the CSI plugin. We need a design for plugin management.
Alternatively, we could just keep using filesystem/linux to do the bind mount, and implement most logic in Slave::publishResources(). The per-workload publish can be done through passing LAUNCH or LAUNCH_GROUP to SLRP.