Determined to make Windows Failover Clusters as resilient as possible, Microsoft has once again made significant improvements to its quorum mechanisms in Windows Server 2012. The Dynamic Quorum Management option allows the cluster
In addition to dynamic quorum, multisite geoclusters now benefit from the ability to specify which nodes receive votes and which ones don't. This allows you to bias a particular site (e.g., the primary site) to have the controlling votes, or nodes, to maintain quorum. This also prevents a split-brain scenario from occurring as the secondary site tries to update the cluster database when the primary site is down. Here's how to configure these new quorum options with Failover Clusters.
Configuring Dynamic Quorum in Windows Server 2012
The principle behind quorum in a failover cluster environment is to ensure that only a majority of nodes can form and participate in a cluster. This prevents a second subset of nodes from forming a separate cluster that can access the same shared resources in an uncoordinated fashion, which can lead to corruption. When nodes are shut down or fail, there are fewer active nodes remaining to maintain the static quorum value of votes needed for the cluster to function. The new Dynamic Quorum Management dynamically adjusts the votes of remaining active nodes to ensure that quorum can be maintained in the event of yet another node failure or shutdown.
There are a few requirements that must be met before the Dynamic Quorum mechanism kicks in. First, Dynamic Quorum must be enabled, which it is, by default, in Windows Server 2012. The Failover Cluster Manager can be used to view or modify the Dynamic Quorum option by running the Configure Cluster Quorum Wizard. Start the wizard by highlighting the cluster in the left-hand pane, right-clicking on it, selecting More Actions and then choosing Configure Cluster Quorum Settings (Figure 1).
The Quorum Wizard prompts you to select from several different quorum configurations depending on your environment (Typical, Add/Change or Advanced). By default, the cluster will use the typical settings for your configuration to establish the quorum management options. You can also add or change the quorum witness if one was selected during the installation process.
To view or change the Dynamic Quorum Management option, use the Advanced quorum configuration option, as seen above. Stepping through the Quorum Wizard, it will prompt you to Configure Quorum Management. This is where you can view or change the Dynamic Quorum option (Figure 2).
You can also view or modify the cluster's Dynamic Quorum setting by using PowerShell cmdlets. The first cmdlet, Get-Cluster, as shown in Figure 3, reveals the current Dynamic Quorum setting (0=disabled, 1=enabled). You can then use PowerShell to enable Dynamic Quorum by establishing the variable $cluster with Get-Cluster and then set the property DynamicQuorum to a value of 1.
With Dynamic Quorum enabled, the next condition that must be met is that the cluster must be up and running and currently sustaining quorum based on the initial cluster configuration. The final condition for Dynamic Quorum to work is that any subsequent node failures or shutdowns must be experienced sequentially -- not with multiple nodes going down at the same time. A lengthier cluster regroup operation would occur if multiple nodes exited the cluster simultaneously.
You can use PowerShell to view the number of votes and observe the inner workings of the Dynamic Quorum mechanism. By default, each server in the cluster gets a single vote, or NodeWeight. When Dynamic Quorum is enabled, an additional property called DynamicWeight is used to track a server's dynamic vote toward quorum. The cluster will adjust a node's dynamic weight to zero, if necessary, to avoid losing quorum, should another node exit the cluster. The PowerShell cmdlet reveals the NodeWeight and DynamicWeight for a two-node cluster (Figure 4).
Dynamic Quorum allows cluster nodes to be individually shut down or fail to the point where just a single node is left functioning ("last man standing"). Just as quorum is dynamically adjusted downward as nodes fail or are shut down in the cluster, quorum is adjusted upward as nodes are rebooted back into the cluster.
Using weighted votes to assign nodes
The other major enhancement to the quorum mechanism in Windows Server 2012 is the ability to specify which nodes in a cluster receive a vote. As mentioned, all nodes receive a vote that contributes toward quorum by default. In multisite geocluster configurations, it may be beneficial to give nodes in the primary site a vote to ensure they keep running in the event of a network failure between sites. Nodes in the secondary site can be configured with zero votes so they cannot form a cluster.
You can use the Quorum Wizard (Advanced Quorum Configurations) to configure whether a node receives a vote. The wizard also allows you to see how Node1 is given a vote and Node2 is not (Figure 5).
Alternatively, you can use PowerShell to specify whether a node receives a vote. Use the Get-ClusterNode cmdlet to set the NodeWeight for Node2 back to 1 so that it receives a vote (Figure 6).
Windows Server 2012 has made significant improvements to the quorum mechanism, resulting in more resilient Failover Clusters. Dynamic Quorum Management takes the worry out of whether enough servers are active to achieve or maintain quorum if systems should fail or shut down. Multisite geoclusters also use weighted votes to specify which primary site should continue running in the event of intersite network failures.
About the author
Bruce Mackenzie-Low is a master consultant at Hewlett-Packard Co., providing third-level worldwide support on Microsoft Windows-based products, including clusters and crash-dump analysis. With more than 18 years of computing experience at Digital, Compaq and HP, he is a well-known resource for resolving highly complex problems involving clusters, SANs, networking and internals. He has taught extensively throughout his career, always leaving his audience energized with his enthusiasm for technology.
This was first published in February 2013