For high availability, Context Store supports a maximum of one failed node regardless of how many nodes are assigned to the cluster. To avoid service interruption, the remaining nodes must be able to support the throughput and capacity of the contribution of the failed node.