Getting rid of unused virtual disks on XenServer

 Posted on February 1, 2017  |  4 minutes

A continuous test server I’d set up had stopped working. The XenServer on which it was running had a 1TB disk: and it was full. What’s going on?

When destroy does not destroy everything

My test would provision a new virtual machine (VM), run its test, and then delete the VM. I used xe vm-destroy to get rid of the VM. However, turns out this command does not delete the virtual hard disk associated with the VM! The correct command is:

xe vm-uninstall force=true uuid=[VM_UUID]

But what if you’ve already made this mistake? There is no built-in command to show you such orphan disks. I had to spend half a day understanding XenServer abstractions, which is a veritable alphabet soup: VDI, VHD, VBD, PBD, SR …

Some terminology

Let me simplify some of those terms, because we’ll have to work with them.

SR

Let’s start with SR. An SR is a storage location for the XenServer. It could be a local hard drive, an NFS mount, a DVD, and so on. SR stands for “Storage Repository”. Here is a sample SR from xe sr-list:

uuid ( RO)                : 6d2685a4-7783-ec05-a1c1-aa2fe2e4891e
          name-label ( RW): Local storage
    name-description ( RW):
                host ( RO): xs.eng.example.com
                type ( RO): ext
        content-type ( RO): user

VDI

What is a VDI? When you create a virtual machine, you can attach a “virtual hard disk”. This is a file system on the VM, but it is stored as a flat file on the underlying controlling operating system (Dom0 in Xen parlance). The file is a .vhd file.

A VDI is a wrapper for a .vhd file (there are other possible formats), and is what you operate on. You never directly work with .vhd files. VDI stands for “Virtual Disk Image”. Here is a sample VDI from xe vdi-list:

uuid ( RO)                : c701a2f3-b5b2-442b-a118-5e9e89c37a26
          name-label ( RW): ubuntu-trusty-disk-image
    name-description ( RW): Ubuntu Trusty Disk Image
             sr-uuid ( RO): 6d2685a4-7783-ec05-a1c1-aa2fe2e4891e
        virtual-size ( RO): 128849018880
            sharable ( RO): false
           read-only ( RO): false

The UUID we see above becomes the name of the .vhd file that actually stores the virtual disk content:

# ls -l c701a2f3-b5b2-442b-a118-5e9e89c37a26.vhd
-rw-r--r-- 1 root root 125865009664 Feb  1 19:15 c701a2f3-b5b2-442b-a118-5e9e89c37a26.vhd

You may notice that the file is actually smaller than the size listed in the VDI. This is a feature, not a bug. The file starts out small and grows as we fill in the disk.

VBD

Also, note that the VDI has no information on which VM it belongs to. Until now, we’re only looking at an independent virtual disk file.

That is where the VBD comes in. A VBD provides additional metadata for XenServer to connect a VDI with a VM: e.g., What is the name of the VM that uses this (virtual) disk? VBD stands for “Virtual Block Device”. Here is a sample VBD from xe vbd-list:

uuid ( RO)             : bb8ecebe-bffe-316d-415f-437f05bacc88
          vm-uuid ( RO): ecce131d-5848-8ee4-e490-3d595ddcb2fe
    vm-name-label ( RO): ubuntu-trusty
         vdi-uuid ( RO): c701a2f3-b5b2-442b-a118-5e9e89c37a26
            empty ( RO): false
           device ( RO): xvda

Finding orphan disks

OK, back to the problem at hand. So we want to find all the virtual disks that do not have a corresponding VM, because the VM was, uhm, destroyed.

First, we list all the virtual disks known to the XenServer, by executing the following command on Dom0:

xe vdi-list sr-uuid=$sr_uuid params=uuid managed=true

You can obtain the ID for the storage that is running full via xe sr-list, as shown earlier.

Next, for each virtual disk listed above, we check if there exists a corresponding VM. The VBD is a mapping from virtual disk to VM, so it is the perfect place to look for this information. If the VBD comes back empty, then we’re looking at an orphan disk.

xe vbd-list vdi-uuid=$vdi params=vm-name-label

We can then delete such orphan disks via xe vdi-destroy command.

The script

The entire shell script is on my Github. You can run it without any options to see if there are any orphan virtual disk files. For each disk file, it prints either OK or DEL. The DEL files are orphans.

If you run it with -x option, the script also deletes orphan files. It now prints DO_DEL for files it deleted.

Conclusion

In this problem, the interesting piece is not the script, or even the solution. What’s interesting is the technology of virtualization and all its intricacies, which we see through the window of Xen.

xenserver  virtualization  shell-scripts 

See also