Scripts to Extract vCloud Director Chain Length and Linked Clone Mappings

There were some questions this week about extracting the chain length for a virtual machine as well as figuring out which Linked Clones were mapped to a particular vAppTemplate and/or shadowVM in vCloud Director. If you are not familiar with the concept of a Linked Clone (Fast Provisioning in vCloud Director 1.5 leverages Linked Clones) and their relationship to chain length, there is a very good white paper called VMware vCloud Director 1.5 Performance and Best Practices that goes into detail about Linked Clones and some of the things to be aware of from a performance standpoint.

Depending on how you deploy a new vApp whether that is from a vAppTemplate in your catalog or copying from an existing vApp, the chain length of a Linked Clone tree will grow either breadth wise or length wise.
The chain length can grow much faster when copying from an existing vApp and that can impact the performance of your virtual machines. Below is a diagram between deploying from a vAppTemplate in a catalog (breadth wise) and consecutive copies from an existing vApp (length wise).

vCloud Director limits the chain length to 30 and automatically spins up a full clone (called a shadow VM) when the maximum is hit and all sub-sequent Linked Clones will be based off of this new shadow VM. A shadow VM can also spin up automatically if the current datastore is getting to full capacity or if the VM needs to exists on a different vCenter Server as Linked Clones do not span datastores or different vCenter Servers.

To view the current chain length of a given virtual machine, you will need to login with a System Administrator account and this can be seen using the vCloud UI at the VM level. You can also get the chain length using the vCloud API which is part of the VCloudExtension property called VirtualDisksMaxChainLength and can be viewed by performing a GET operation on a VM.

Note: Though the property has "MaxChainLength" in the name, this property represents the current chain length and it is not configurable, as the system max is 30.

To be able to quickly get a summary of the chain length of all VMs that include vAppTemplates and Shadow VMs, I wrote a vCloud SDK for PHP script called vcloudVMChainLength.php. The script provides a few pieces of information: VM name as shown in vCloud Director, name of the vApp the VM is part of in vCloud Director, whether it is a vAppTemplate, the vCenter Server hosting the VM, the MoRef of the VM and chain length.

Before I provide the sample output, let me give you some background about my lab environment. I have two vAppTemplates called "Application Server" and "Database Server". The "Application Server" is currently residing on a datastore with close to full capacity and I will be deploying 3 new vApps from this vAppTemplate called "App-Test-1", "App-Test-2" and "App-Test-3". I will also be deploying 3 new vApps from "Database Server" but only the first clone will be from the vAppTemplate, the remainder will be a clone of the previous vApp in the workspace.

Here is a diagram of the 9 VMs (2 vAppTemplates, 1 shadow VM and 6 vApps):

Here is the sample output of the environment above:

We can see that during the deployment of the "App-Test-*" vApps, the datastore capacity was low and vCloud Director automatically spun up a shadow VM which is a full copy and created the 3 new Linked Clones off of that image and the chain length is max of 2. With the "DB-Test-*", we started out with deploying from the vAppTemplate, but all subsequent copies was from the previous vApp which grew the chain length to 4. If we continue to copy from the vApp length wise, we will quickly reach the maximum chain length and possibly degrade the performance of the VMs as the deepest node will need to traverse back to the parent to perform it's disk read operations. 

The other question I had was about mapping the Linked Clones back to their parent VMs. This is not something you can really do in vCloud Director nor using the vCloud API. However, with a little bit of creativity and information from vCloud Director, you can leverage the vSphere API to help you get a list of VMs that are linked off of a given VM. I wrote a vSphere SDK for Perl script called vCloudVMLinkedClonesMapping.pl that can help you get this information. 

Note: This was tested in my lab which had a minimal configuration, ensure you do additional validation before making any decisions based on the output.

To use the script, you will need to provide the MoRef ID for a VM that is either a vAppTemplate, shadow VM or just a regular VM which may have Linked Clones. You can get this information by using the script that was shown earlier.

Let's take a look at our "Application-Server" VM which has MoRef ID 677 and see if there are any Linked Clones:

As we expected, there are no Linked Clones as the datastore it is currently residing on is getting to full capacity and a shadow VM was spun up for the new 3 deployments.

Let's take a look at the shadow VM "shadow-App-VM" with MoRef ID 721 which is automatically prefix with "shadow-" of the source VM name:

Just as we expected, we can see that our 3 vApps are linked off of the Shadow VM that we just checked.

Note: The output includes the display name and URN ID of the vApp in vCloud Director so you can easily identify it.

Finally, let's take a look at our "Database-Server" VM which has MoRef ID 675 and we should also see the 3 Linked Clones:

You might ask if it is necessary to keep track of all this information and the answer is yes/no. Though it is good to understand how vApps are being deployed by your consumers and ensure they are efficiently doing so by deploying from a catalog. vCloud Director has built in mechanisms to automatically handle deep chain lengths or low datastore space by deploying additional shadow VMs to ensure users are still able to request new vApps without any impact.