- Clustered Data ONTAP 8
- ONTAP 9
Check Active IQ if this impacts your systems
'Cluster HA' refers to a concept that only applies to two node clustered Data ONTAP configurations. Please note that 'Cluster HA' is different from 'Storage Failover HA', which deals with one node's ability to take-over the partner node's disks in a storage HA pair configuration. However, in order to properly configure Cluster HA, the proper physical cabling for a 2-node clustered Data ONTAP cluster cabled for multipath HA must be adhered to. Some key concepts, restrictions, and guidelines for physical cabling can be found in the Multipath and Dual-Path Cabling Information, as well as in the respective FAS/V-Series hardware documentation.
In a clustered ONTAP cluster, there are two methods a cluster can use to determine if all of the nodes in the cluster are in quorum.
Clustered Data ONTAP configurations that are greater than 2 nodes:
For clustered Data ONTAP configurations that have greater than two nodes in the cluster, the entire cluster maintains quorum using a voting mechanism where each node has a vote that decides which node will maintain the master copy (read/write) of the replicated databases (also known as the RDB Master). This voting is accomplished using the dedicated cluster networking that all nodes in the cluster are connected to. In order to maintain cluster quorum, there must be greater than 1/2 the total number of nodes that are alive and communicating with each other. When there is an even number of nodes in the cluster, there is always a possibility that exactly 1/2 the number of nodes in the cluster can become segmented from the other 1/2 of nodes in the cluster, and thus when this happens, each half of the cluster must decide which node is the RDB Master. However, since the cluster requires more than 1/2 the number of nodes in the cluster to maintain quorum, both halves will not be able to elect an RDB Master, thus the entire cluster will be out of quorum to avoid a situation that is commonly referred in the industry as 'Split Brain', where both halves of the cluster could end up having two separate read/write copies of replicated databases. To solve this issue, the cluster has a notion of 'epsilon' node. Epsilon is a concept where a single node in the cluster (typically the first node that created the cluster during initial setup) is statically assigned a vote that has a slightly greater weight than all of the other nodes in the cluster. Thus, in a situation where there is a segmentation of exactly 1/2 of the cluster, the 1/2 that is able to communicate with the epsilon node will elect a new RDB master node and the 1/2 that is not able to communicate with the epsilon node will go out of quorum, thus protecting against a 'Split Brain' situation and also maintaining cluster quorum. This kind of cluster configuration (greater than two node) does not require 'Cluster HA' to be enabled as quorum and quorum voting utilizes cluster networking.
Clustered Data ONTAP configurations that contain exactly 2 nodes:
With clustered Data ONTAP configurations that contain exactly two nodes, both nodes are equally important. If one node goes down, the other node must be able to become the RDB Master. In order to accomplish this, the voting mechanism and epsilon that is used in cluster configurations greater than two nodes is eliminated in favor of a different mechanism. The mechanism that a two node cluster uses leverages the physical disk connectivity that both nodes share as part of a Storage HA pair. Each node will monitor the other node using this disk connectivity and should one of the node go down, the surviving node automatically becomes the RDB Master. This mechanism is referred to as 'Cluster HA'.
In clustered Data ONTAP releases prior to 8.2, the Cluster HA must be explicitly configured as part of the clustered Data ONTAP installation and setup process.
More details about this topic are covered in the follow clustered Data ONTAP documentation.
Data ONTAP 8.3 System Administration Guide for clustered Data ONTAP: Understanding quorum and epsilon
Data ONTAP 8.3 High-Availability Configuration Guide for clustered Data ONTAP: