Thin disks on NFS datastores are converted to thick during restore

This is a known VMWARE behavior when restoring Virtual Machines on NFS volumes:

https://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=2137818

Cause
NFS hosted virtual disk created as thin-provisioned is first fully written and then marked as thick-provisioned eager-zeroed:
1. Virtual machine is created thin-provisioned on an NFS datastore.
2. During its first backup, the virtual machine is fully read and marked fully-written because CBT information is not available for NFS volumes.
3. On a restore, since the virtual machine virtual disk is read as fully-written, ESXi disk type decision logic marks it as thick and eager-zeroed.
Resolution
This is a known issue affecting VMware ESXi 5.5. Currently, there is no resolution.

To work around the issue, migrate thin-provisioned virtual machines from an NFS datastore to a VMFS datastore for thin-provisioned virtual disk backup or restore operations.


Also, as of vSphere 6.0, NFSv4 is being used but it still doesn’t support CBT:


https://kb.vmware.com/selfservice/search.do?cmd=displayKC&docType=kc&docTypeID=DT_KB_1_1&externalId=2077787

Cause
This issue occurs because Changed Block Tracking (CBT) is not supported when the ChangeID is set to "*" against virtual machines residing on NFS. The QueryChangedDiskAreas("*") API returns an error against NFS version 3 and earlier.

Note: vSphere 6.0 supports NFS version 4, which provides sector change information, however NFSv4 does not provide the FIEMAP ioctl, so CBT still cannot be used to query changed sectors for backup and restore operations.
Resolution
This is expected behavior. CBT with a change ID equal to "*" is supported only on VMFS file systems.
To work around this issue, migrate thin-provisioned virtual machines from an NFS datastore to a VMFS datastore for thin-provisoned disk backup or restore operations.