Failover Clustering and Network Load Balancing Team Blog
A few weeks ago we gave an overview of the cluster quorum model: http://blogs.msdn.com/b/clustering/archive/2010/05/14/10012930.aspx
This week we will provide some more details about when and how the different types are used.
The two most recommended quorum modes are the Node Majority and Node and Disk Majority. For cluster with odd number of nodes, Node Majority is always recommended. For cluster with even number of nodes and Available Storage, Node and Disk Majority is the recommended type.
For example, if a cluster has 2 nodes, add a disk witness resource would create a Node and Disk Majority cluster and the cluster would have 3 votes (2 nodes + disk witness resource). The cluster would maintain quorum if either 1 node is down or if the disk witness failed. If a cluster has 2 nodes and there is a no disk witness resource (Node Majority cluster), the number of votes is 2 and the minimum number of quorum voters to remain online 2. This means that if either of the nodes goes down, the whole cluster would come down. So the minimum number of nodes needed for failover cluster with 3 nodes without disk witness resource and 2 nodes with disk witness resource (3 total votes).
For an odd number nodes cluster, a disk witness resource would not increase the survivability of a cluster. For example, if a disk witness resource is added to a 3 node to create a Node and Disk Majority cluster, the number of votes is now 4. The quorum requirement is 4/2+1= 3. This means that at least 3 votes must be up all the time. If the disk witness resource is down, any of the 3 nodes goes down would bring the cluster under the quorum and bring down the cluster. If the 3 node cluster is a Node Majority cluster, which means it has no disk witness resource, the cluster can survive when 1 of the nodes goes down. Therefore, the quorum mode should be chosen to increase the survivability of the cluster when with the maximum of nodes failure in mind.
Fortunately, Failover Cluster Manager intelligently picks the best model for you based on the number of nodes and disk availability when creating the cluster. It will use Node and Disk Majority Quorum settings when create cluster with even number of nodes, provided that cluster disk resource is available. Similarly, Failover Cluster Manager will select Node Majority Quorum when creating a cluster with odd number of nodes.
For Node and Disk Majority Quorum, the disk chosen automatically by Failover Cluster Manager during create cluster might not be the one you wish to place as a disk witness. To change the disk witness, right-click on the cluster name and select More Actions…, then Configure Cluster Quorum Settings… to get to the “Configure Cluster Quorum Wizard”. Walk through the wizard, select “Node and Disk Majority Quorum” button and select or change the disk witness. The original disk witness would automatically be place back in the available storage disk.
The Powershell commands to set the quorum are listed below and documented online: http://technet.microsoft.com/en-us/library/ee461013.aspx
For Node Majority:
For Node and Disk Majority”
Node and File Share Majority Quorum mode uses a file share location instead of disk witness as the additional vote. So Node and File Share Majority cluster would only make sense for even number nodes. The explanation for why Node and File Share Majority should be use only on even number node is exactly the same as the Node and Disk Majority Cluster. Please consult the previous section for Node and disk Majority Quorum.
This quorum configuration option is often seen in a multi-site / multi-subnet cluster with an even number of nodes at two sites, and this additional vote creates an odd number of total votes. If you have 2 nodes at Site 1, 2 nodes at Site 2 and the File Share Witness at a third site, you can sustain the failure of any site and still maintain quorum with at least 3 votes.
Node and File Share Majority Quorum are chosen because no shared disks are available or disks are needed for other tasks. To setup a file share witness disk, create a share folder on a remote machine accessible from all the nodes in the cluster. The share must be in the same forest and should not be on a node in the same cluster (as losing that node would result in the loss of 2 votes). The file share folder must granted full access to the Computer Name Account (CNO) of the cluster. From “Configure Cluster Quorum Wizard”, select the “Node and File Share Quorum” button and enter the file share path information.
\\<machine name>\<file share path and folder>
The Powershell command for Node and File Share Majority is:
Set-ClusterQuorum -NodeAndFileShareMajority \\A19C1I4X64C1\fsw
Prior to Windows Server 2003, the only available quorum type was the Disk Only. In this setup, only the Disk Witness has the single authoritative vote and if the disk witness resource fails, the entire cluster would go down. Because of this single point failure in the single disk vote, No Majority: Disk Only is not recommended. The primary scenario for using No Majority: Disk Only is when dealing with unstable software or network that constantly cause networks or nodes to fail while the disk access is stable. With this scenario, selecting No Majority: Disk Only Quorum ensures that as long as a node is up, the cluster is up and thus the resource is running. Sometimes this will also be temporarily using during an upgrade when the number of nodes in the cluster gradually increases, so as to avoid constantly changing the quorum model.
The Powershell command for No Majority: Disk Only Quorum is:
Set-ClusterQuorum -DiskOnly “cluster disk 5”
The PowerShell CMDlets are documented online at: http://technet.microsoft.com/en-us/library/ee461013.aspx
Thanks,Peter HuangSoftware Development Engineer in Test IIClustering and High-AvailabilityMicrosoft
Excellent. Thanks for give the in-depth explanation
What would happen if 1 node goes down due to os corruption for extended time in node majority cluster of 3 nodes ; should we change the quoroum model to node + disk and add a disk to avoid cluster shut down. Should we first evict the failed node before changing quorum model