Windows Server cluster quorum

Group and the purpose of arbitration

One of the purposes of cluster quorum is to prevent the cluster from causing a split-up when the cluster has a network partition, which is the result of partitioning (or grouping) of the cluster, which means that both partitions assume that the other party is no longer present or invalid, thus competing for control over the cluster resources. The consequences of a brain fissure are that two partitions are simultaneously and independently read and written to the shared disk, causing confusion in disk data.

The second purpose of the arbitration is to limit the maximum number of failures that the cluster can withstand, and the quorum requires a majority of the cluster to exist, or the cluster will fail, such as a 5-node cluster that can tolerate a maximum of two nodes at a time of failure.

Cluster voting

Voting algorithms are based on the majority of voting results, and each node in the cluster needs a heartbeat mechanism to communicate the "health state" of each other, and each node receives a "notification" representing a vote. For a cluster of three nodes (assuming three nodes are a,b,c), each node will have three votes when it is running normally. When Node A has a heartbeat failure but is still running, B and C heartbeats are normal, and the entire cluster splits into two partitions. Node A is in a partition, B and C in another partition. A partition must be removed to ensure the healthy operation of the cluster. According to the voting algorithm, B and C in a partition, there are two votes, a only one vote, B and C composed of the cluster gain control, a is removed.

The cluster node, the disk witness resource, and the file share resource can all be assigned voting rights, and the voting rights can be allocated in the cluster manager. A cluster node can be assigned a poll, as well as a disk witness or a file share witness resource can also be assigned a poll.

The right to vote does not represent the ability to participate in the voting arbitration, and must ensure the validity of the vote. In Cluster Administrator, you can see whether the voting rights are assigned and whether they have a valid vote, only the assigned votes are valid, as shown in 1.

650) this.width=650; "title=" 002-figure 1.jpg "style=" Float:none; "alt=" Wkiol1igxfwscp-taaa1sfjapzo706.jpg-wh_50 "src=" Https://s2.51cto.com/wyfs02/M02/8D/86/wKioL1igXFWSCp-TAAA1SfjApzo706.jpg-wh_500x0-wm_3-wmp_4-s_3342398168.jpg " />

Figure 1 Cluster voting rights and valid votes

Arbitration Witness

If the cluster has only two nodes, because there is only one vote on each node, the voting algorithm does not work, and each zone after the cluster partition has a valid poll causing the cluster to split. Similarly, if the cluster nodes are even, and the cluster partition is just about half the nodes, it causes the cluster to crack. To prevent this from happening, a third device needs to be introduced: the witness of the arbitration. The purpose of the arbitration witness is similar to the example of increasing voting in life, both shared disks and shared folder resources can act as quorum witnesses, and shared disks are also called quorum disks when used for quorum. The disk and shared files for the witness used for the quorum can also be assigned a poll, when the heartbeat between nodes in a two-node cluster is a problem, two nodes simultaneously go for the quorum witness this ticket, the first arrival request is satisfied first. The same is true for clusters with even-numbered nodes rip after partitioning, so that more than half of the votes are obtained from the nodes and partitions that get the quorum witness first, and the other node or partition is rejected.

In a Windows Server cluster, the default is to assign voting rights to each node, and when the cluster server node is odd and the network partition is unfortunate, because the poll results can form a majority node partition, the cluster rejects nodes of a small number of node partitions. However, if the cluster node is an even number and the network partition occurs, the cluster may have node rip, which is a good time to configure the quorum witness for the cluster.

Currently, the cluster supports 4 types of quorum models, including node majority, node and disk majority, node and file share majority, and disk only (Legacy model). Node most models only allocate votes for cluster nodes, node and disk (file share) Most models are voting for both cluster nodes and cluster disks (file shares), and only the disk model is a traditional model. The number of votes for the 4 arbitration models and the corresponding table of arbitration instructions are as follows:

Arbitration Model	Components with voting rights (number of votes)	arbitration ( v is the number of votes, v/2 rounding down)
Node most	Node (number of nodes)	V/2 + 1
node and disk most	Nodes and disks (node Count + 1)	V/2 + 1
node and file share most	Node file Share (nodes + 1)	V/2 + 1
Disk only (Legacy)	Disk Witness resources (1)	V

Witness Resources

Both shared disks and shared folders can be used as quorum witness resources, but they are slightly different, except that the disk witness can vote for the quorum, and the disk that acts as the quorum witness also stores a continuously updated copy of the cluster database, helping the cluster maintain the cluster's normal state and configuration information in the event of a node failure. Make sure that each node holds the most recent database copy. The file share quorum witness cannot store the cluster database.

Cluster quorum evolution

The quorum technology continues to evolve with the iterative updating of Windows Server and the complexity of the environment, starting with the initial cluster to a Windows Server 2016 cluster, where the quorum model has changed a little. Next, briefly describe the evolution of the Windows Server cluster quorum approach.

(1) disk-only quorum

In earlier Windows Server cluster arbitrations, the disk had a unique quorum vote, and if the disk witness resource failed, the cluster would fail, and the quorum model was too dependent on the cluster disk and was rarely adopted. Although the disk quorum mechanism is still retained in the cluster, the arbitration mechanism is gradually discarded. This kind of arbitration mechanism is only considered when the disk network is stable, but the network instability between nodes leads to intermittent failure of network and node communication.

(2) Witness and node arbitration

the Windows server cluster Development group found that disk-only quorum was unable to accommodate most environments, and began to explore new arbitration mechanisms. Starting with Windows Server 2008 the cluster nodes also have the right to vote, optionally assigning voting rights to nodes, and the node and resource witness can also be combined into a more flexible voting model. Node voting and disk voting combine several quorum options such as node majority, node and disk majority, node and file share majority.  

(3) Dynamic arbitration

A dynamic witness quorum was introduced in Windows Server R2, the cluster is configured for dynamic quorum by default, and the cluster is dynamically voting based on the current number of cluster voting nodes. Dynamic quorum means that when the cluster increases, deletes, shuts down, opens cluster nodes, or increases the deletion witness resource, the cluster dynamically adjusts the number of votes and the validity of the poll to keep its number of valid votes odd. The quorum witness vote is also dynamically adjusted based on the witness resource status, and if the witness resource is offline or fails, the cluster will set the witness vote to 0, which means that the witness will no longer participate in the poll.

For example, in a 4-node cluster, although 4 nodes are assigned to vote, the cluster to keep its number of votes odd, "take" a node of the vote, the remaining 3 valid votes, 2 is shown.

650) this.width=650; "title=" 002-figure 2.jpg "style=" Float:none; "alt=" Wkiol1igxffipl2iaab2csgddqs365.jpg-wh_50 "src=" Https://s2.51cto.com/wyfs02/M01/8D/86/wKioL1igXFfiPL2IAAB2CSGddQs365.jpg-wh_500x0-wm_3-wmp_4-s_3792602252.jpg " />

Figure 2 Number of votes and valid votes in a dynamic quorum

If a disk witness or file share witness quorum is configured for the cluster at this point, then 3 of the votes allocated in the cluster will be 5, and a valid vote of 4, when the cluster keeps its number of valid votes odd, the "take Away" vote is "returned" to the node. At this time 4 nodes plus 1 witness resource quorum, a total of 5 votes, 4 is shown.

650) this.width=650; "title=" 002-figure 3.jpg "style=" Float:none; "alt=" Wkiom1igxfjwl7btaaa-zpuao_g408.jpg-wh_50 "src=" Https://s4.51cto.com/wyfs02/M00/8D/89/wKiom1igXFjwL7BTAAA-ZPuAO_g408.jpg-wh_500x0-wm_3-wmp_4-s_2260828450.jpg " />

Figure 3 Cluster Increase share witness

650) this.width=650; "title=" 002-figure 4.jpg "style=" Float:none; "alt=" Wkiol1igxfuhnhuxaabumizxysm842.jpg-wh_50 "src=" Https://s5.51cto.com/wyfs02/M00/8D/86/wKioL1igXFuhnHUXAABumiZXySM842.jpg-wh_500x0-wm_3-wmp_4-s_3688765745.jpg " />

Figure 4 Number of votes re-adjusted in dynamic arbitration

If a node is manually closed at this point, then the allocation of a valid vote in the cluster will be 4 remaining, when the cluster to keep its number of valid votes is odd, then "take" a node of the vote, only two nodes have a valid vote. At this time two nodes plus a witness resource quorum, a total of 3 votes, 5 shows.

650) this.width=650; "title=" 002-figure 5.jpg "style=" Float:none; "alt=" Wkiom1igxf2hchvlaabekq6u4ec865.jpg-wh_50 "src=" Https://s3.51cto.com/wyfs02/M02/8D/89/wKiom1igXF2hCHvLAABEKq6U4ec865.jpg-wh_500x0-wm_3-wmp_4-s_1605254045.jpg " />

Figure 5 Closing the node dynamic quorum re-adjustment poll number

(4) Cloud Arbitration

The Cloud quorum model was introduced in Windows Server 2016, as shown in 6. The cloud quorum model can use Azure BLOB storage as a cluster quorum witness resource for Windows Server 2016, as shown in 7.

650) this.width=650; "title=" 002-figure 6.jpg "style=" Float:none; "alt=" Wkiom1igxf6wucepaaboti0nkxu059.jpg-wh_50 "src=" Https://s3.51cto.com/wyfs02/M01/8D/89/wKiom1igXF6wuCEpAABOtI0NKxU059.jpg-wh_500x0-wm_3-wmp_4-s_2817215856.jpg " />

Figure 6 New Cloud quorum Witness

650) this.width=650; "title=" 002-figure 7.jpg "src=" https://s2.51cto.com/wyfs02/M01/8D/89/ Wkiom1igxzvwsxrdaabedn8hqee964.jpg-wh_500x0-wm_3-wmp_4-s_488597764.jpg "alt=" Wkiom1igxzvwsxrdaabedn8hqee964.jpg-wh_50 "/>

Figure 7 Configuring the Cloud Quorum Witness

Windows Server cluster quorum