VM fails to power on with "Unable to create virtual SCSI device for scsi1:0" on WSFC / SQL Cluster shared disks

Products

• VMware vSphere ESXi
• VMware vCenter Server

Issue/Introduction

When attempting to power on Microsoft Windows Server Failover Clustering (WSFC) or SQL Server cluster virtual machines using shared VMDKs, the power-on task fails.

Error Message observed in vCenter Server Task Console:

Unable to create virtual SCSI device for scsi1:0, '/vmfs/volumes/<Datastore>/<VM_Folder>/<VM_Name>_2.vmdk'
Failed to start the virtual machine.
Module DevicePowerOn power on failed.

Environment

 

• VMware vSphere ESXi

• VMware vCenter Server

 

Cause

Shared VMDKs configured with Physical or Virtual SCSI Bus Sharing rely on strict host-level disk filters and locking mechanics to allow multi-node cluster access. This power-on failure occurs due to one of two reasons:

1. Unsupported Snapshot or Replication Filters: Snapshot-based replication solutions (such as vSphere Replication or 3rd-party backup/DR tools) attach filter drivers to the virtual SCSI controller. Because SCSI Bus Sharing explicitly prohibits snapshot and filter operations, enabling replication generates invalid or locked .vmfd sidecar metadata that prevents the SCSI controller (scsi1:0) from initializing.

2. Orphaned Disk Locks Across Hypervisors: If cluster nodes sharing the same VMDKs are powered off abruptly or co-located on the same ESXi host without DRS Host Anti-Affinity Rules, the hypervisor can fail to release the SCSI reservation lock. This leaves orphaned .vmfd sidecar files and stale ddb.sidecars references in the VMDK descriptor file, blocking the virtual machine from claiming the SCSI target at power-on.

Resolution

Step 1: Stop Replication & Power Off Cluster VMs

1. Remove or stop the replication job for all WSFC cluster VMs, ensuring shared disk controllers (scsi1:X) are excluded.

2. Power off all virtual machines participating in the WSFC cluster.

Step 2: Clean Up Stale Sidecar Files (.vmfd)

1. SSH into the ESXi host and navigate to the VM datastore folder:

   cd /vmfs/volumes/<Datastore_Name>/<VM_Directory>/
   

2. Locate and delete the orphaned sidecar (.vmfd) file(s) associated with the shared disk:

   rm <Shared_Disk_Name>-<hash>.vmfd
   

Step 3: Edit VMDK Descriptor File

1. Open the shared disk descriptor file (.vmdk) using vi:

   vi <Shared_Disk_Name>.vmdk
   

2. Delete the line starting with ddb.sidecars:

   ddb.sidecars = "SCSI3_PRINFO,..."
   

3. Save and exit (:wq!).

Step 4: Configure Host Anti-Affinity & Power On

1. In vCenter, create a vSphere DRS VM Anti-Affinity Rule to keep all WSFC cluster nodes on separate ESXi hosts.

2. Power on the cluster virtual machines sequentially.

Additional Information

Constraint: Snapshots, vSphere Replication, and Storage vMotion are strictly unsupported on virtual machines with active SCSI Bus Sharing / shared VMDKs. SQL Disaster Recovery for these workloads should be achieved via application-level mechanisms (e.g., SQL Always On Availability Groups or Array-Based LUN Replication).