Physical mode RDMs cannot be migrated by any automated tool
What this check detects
Any disk where Raw is true and Raw Comp. Mode is physical (also written as physicalMode or pass-through).
- RVTools tab
- vDisk
- Columns read
- Raw, Raw Comp. Mode, Raw LUN ID
Why it matters
A raw device mapping in physical compatibility mode is not a virtual disk. The mapping file on the datastore is a pointer, a few kilobytes in size, and the guest’s SCSI commands — including vendor-specific ones like SCSI-3 persistent reservations — are passed directly to the underlying LUN. There is no VMDK holding the data, so there is nothing for a converter to read.
This is what separates physical mode from virtual mode. A virtual mode RDM intercepts SCSI commands and can be cloned into a VMDK during a Storage vMotion, which means most tooling can handle it. A physical mode RDM cannot, and every automated migration path — Proxmox’s importer, Nutanix Move, Microsoft’s VM Conversion — will either skip the disk silently or fail the VM.
The silent skip is the dangerous outcome. A VM migrates, boots, and appears healthy, but the application data volume is missing or empty. This is most common with clustered workloads: Windows Server Failover Cluster quorum and shared data disks are the textbook physical mode RDM use case, and they are exactly the workloads where a partial migration causes the most damage.
How to resolve it
The LUN holds ordinary application data
- Confirm the disk is not shared with another VM: check the Raw LUN ID against every other row in the vDisk tab.
- Power the VM down. Physical mode RDMs cannot be converted live.
- Storage vMotion the RDM to a VMFS or vSAN datastore, selecting the thin or thick provisioned VMDK format. This copies the LUN contents into a real virtual disk.
- Verify the disk now shows Raw = false in a fresh RVTools export before scheduling the migration wave.
- Migrate normally. Decommission the source LUN only after the target VM has been validated.
The LUN is shared by a Windows Server Failover Cluster
- Do not convert the disk in place. Shared-disk clusters need the cluster rebuilt on the target platform, not the disks moved.
- Build the replacement cluster on the target using its native shared storage — Proxmox with a shared LVM or Ceph RBD, AHV with volume groups, Hyper-V with cluster shared volumes.
- Migrate the application by joining new nodes to the cluster and failing over, rather than by converting VMs.
- Treat the cluster as a single unit in the wave plan. Its nodes must move together.
The LUN can be retired
- Where the RDM exists only for a historical reason — a P2V that was never cleaned up, or a performance decision that no longer applies — copy the data into a standard VMDK at the file level and detach the RDM.
- This is usually the cheapest resolution, and it removes the finding permanently.