Open chucklever opened 1 month ago
Scott says that usually upon careful questioning, the customer request is for parallel I/O, which is not necessarily the same as active/active high availability. pNFS is generally the preferred approach for this, and NFSD can currently offer only pNFS block.
We would want to move the pNFS metadata service (eg, via a floating IP address). The SCSI persistent reservation metadata would reside on each SCSI target, though we think the metadata server's initiator might maintain some of this state, and that would need to follow the floating IP.
Scott says there are long-term bugs/ERs open internally at Red Hat requesting better (easier) support for NFSD active/active configurations. We of course would need a clustered file system to make this happen.