In a vSAN cluster with VDI environment, when you must replace a failed cache device you delete disk groups from a host or in situation where you end up re-creating complete disk group.
When you recreate a disk group in the host after rebuild finish there are no objects in a disk group and hence the DG is empty initially.
When user logout from Linked Clone desktop the "checkpoint.vmdk" will be removed and recreated.
If there are no objects in the disk group of the host, checkpoint.vmdk will be deployed to the newly created disk group of the host. As all the newly created checkpoint.vmdk will be created on new diskgroup, there is possibility of heavy IO burst on the disk group once user starts log in VDI environment causing the IO congestion.