This article discusses the configuration and risks associated with sharing an RDM (Raw Device Mapping) virtual disk between two or more virtual machines without utilizing a clustering service on the guest operating system.
If a shared disk is not managed by a clustering service (such as Microsoft Cluster Service (MSCS), Symantec Veritas Cluster Server (VCS), or equivalent), attaching the virtual disk to multiple powered-on virtual machines prevents the disk from functioning as a properly synchronized shared storage device.
Guest operating systems cache disk changes locally; therefore, a write operation executed by one virtual machine is not detected by the others. This lack of synchronization allows subsequent modifications by other virtual machines to overwrite initial changes, ultimately leading to severe data corruption.
To safely configure and use a shared virtual disk without a clustering service, the following guidelines and workarounds must be applied:
A RAW LUN must be provisioned, and an RDM pointer in virtual compatibility mode should be created for the respective virtual machines.
To prevent data corruption, one of the following operational methods must be utilized:
Read-Write / Read-Only Configuration: Only one virtual machine is permitted to access the device in read-write mode, while all other virtual machines must be restricted to read-only access.
Note: The read-only virtual machines are required to unmount and remount the volume before any new data written by the read-write host can be recognized.
Exclusive Power-On Execution: If multiple virtual machines are configured to use the volume in read-write mode, it must be ensured that no more than one of these virtual machines is powered on at any given time.
For instructions on building and configuring a Microsoft Cluster, refer to the Broadcom documentation - Setup for Windows Server Failover Clustering on VMware vSphere.
For further details regarding Raw Device Mappings, refer to the Broadcom documentation - Raw Device Mapping in vSphere.