VMDK Recovery Kit Overview

The primary focus of the LifeKeeper for Linux VMDK Recovery Kit is to offer LifeKeeper users an alternative storage method for shared storage and data replication. The VMDK Recovery Kit enables the creation of LifeKeeper resource hierarchies on LifeKeeper protected servers. A virtual hard disk provided by the VMware hypervisor is connected to this resource hierarchy and the file system created on that disk is mounted. When a failure is detected on a node in the cluster where the virtual hard disk is connected, the VMDK Recovery Kit initiates a failover to the predetermined backup node and connects the same virtual hard disk to the backup node.

Once the file system configured on the virtual disk is mounted on a LifeKeeper server, it can be fully utilized as additional storage for LifeKeeper hierarchies. Resource hierarchies for the VMDK Recovery Kit are created using the existing File System Recovery Kit available with the LifeKeeper Core product (steeleye-lk package).

VMDK Recovery Kit Restrictions

• This version of the VMDK Recovery Kit does not include support for a local recovery when access to the virtual hard disk fails. When a failure is detected the default action is to initiate a transfer of the hierarchy to a backup server. Depending on the makeup of the resource hierarchy, this action can result in hung processes. To avoid hung processes, the default action can be changed to halt the server and force a failover to a backup server. To change the default switchover behavior, alter the LKVMDKERROR setting in /etc/default/LifeKeeper. See Configuring the LifeKeeper for Linux VMDK Recovery Kit for more information on LKVMDKERROR.

• All guests participating in the cluster must have the same SCSI controller configuration. The VMDK Recovery Kit reconnects the virtual hard disk to the SCSI controller on the virtual hard disk that was connected when the resource was created.

• Snapshots cannot be created or restored on a LifeKeeper protected VMDK. When the virtual hard disk is switched between nodes, consistency of the snapshots cannot be guaranteed.