One domain controller configured in a failover cluster? Or two DCs, one on each cluster node?

Question

I am currently planning a high available Windows Server 2012 R2 environment within a Hyper-V-Cluster with two nodes. I am not sure about how to deploy the virtual domain controller for the work domain, to keep the forest available when a node goes down.

I see two options:

1. Create one virtual machine with the domain services and configure the vm as a cluster resource in the failover cluster. Let the cluster worry about the availability of the virtual machine (domain controller).

2. Create one VM on Hyper-V-Node 1. Create a second VM on Hyper-V-Node 2. VMs are not configured as a cluster resource (no redundancy per VM). Install domain services on both VMs. Let the domain controllers worry about offering the domain services, if a node and therefore one VM goes down.

My Questions are:

• Is there a preferred method?
• What advantages and disadvantages do both methods have?

I appreciate high quality answers, please back up your responses with sources.

Solution 1

Creating a single, clustered Domain Controller isn't creating highly available AD services. It's creating a highly available VM, regardless of the services that VM is serving. If you lose the VM (OS failure, data corruption, etc.) then you've lost the services that VM was serving, so those services were not highly available to begin with.

Don't confuse a clustered, highly available VM with a clustered, highly available service. A clustered, highly available VM makes the VM highly available and indirectly makes the services on the VM highly available BUT only so long as the VM itself is up and running. if the VM itself is down, so are the services it provides. Clustered VM's protect against HOST failures, not VM or service level failures.

The proper course of action would be to create two Domain Controllers. Place them where you wish and cluster them if you want to. If it were me, I'd create two and place them as you've suggested (one DC on each host) and I'd cluster them as well. There's no reason to have one of them be unavailable if one of the hosts goes down. Clustering them allows both to continue serving clients regardless of which host is up or down.

Solution 2

Domain controllers and failover clusters are mutually exclusive.

"It is not supported to combine the Active Directory Domain Services role and the Failover Cluster feature on Windows Server 2012"

"It is not recommend to combine the Active Directory Domain Services role and the Failover Cluster feature on Windows Server 2003, Windows Server 2008, or Windows Server 2008 R2"

http://support.microsoft.com/kb/281662

Whatever you are doing with Active Directory, it needs to do what a Windows client would do. Find an available domain controller and use it, and if it has an issue with a domain controller, find the next available domain controller and use it.

Assuming you actually read the entire article, it is really a cautionary tale. The message is "don't do it".

Comments

[Matt]

Domain Controllers are inherently highly available anyway, you don't need to cluster them (nor should you). Microsoft's guidance is to place at least two domain controllers per Active Directory site; this is enough and is what you should be doing.

[krisFR]

I appreciate high quality answers, please back up your responses with sources Where are sources ? (i am kidding)

[Daniel]

Reading your response I realise that my question was ambiguous. I know domain services on cluster nodes are not supported. I do not want to install domain services on the cluster nodes, but put a VM on each node and not configuring them as cluster resources. See my edited question for details. I hope it's clearer now. Still, your answer is not wrong, so have an upvote :)

[Daniel]

You addressed some important points, having a VM that is highly available but not protected against faults in the OS.

[Greg Askew]

You need a minimum of two domain controllers, so option 2.

[joeqwerty]

This is incorrect. AD-less cluster bootstrapping in Windows Server 2012 and forward allow a failover cluster to bootstrap without a DC. - blogs.technet.com/b/wincat/archive/2012/08/29/…