The following terms are used throughout the remainder of this chapter and in the Serviceguard documentation.
Cluster: a set of two or more nodes that have enough hardware and software redundancy that a hardware or software failure will not significantly disrupt service
Package: a workload that has been configured to run on a cluster. A package is capable of running on more than one node in the cluster and can be migrated from node to node for maintenance or in the event of a hardware or software failure.
Primary Node: the node where a package is configured to run most of the time. Generally a package runs on its primary node unless the node fails or requires maintenance, in which case the package is moved to an adoptive node.
Adoptive Node: a node where a package is configured to run when the primary node is unavailable
Cluster Heartbeat: a message sent from each node in the cluster to every other node in the cluster using either the network or a serial line. The cluster heartbeat is used to determine when a node has failed or is no longer accessible. In the event that a node becomes unreachable, the remaining nodes in the cluster will reform the cluster without the failed node.
Cluster Lock: required when a cluster is forming and exactly half of the nodes in the cluster are available. The cluster lock prevents two separate instances of the same cluster from creating a situation known as split-brain syndrome. Only the set of nodes that acquires the lock is able to form the cluster. A cluster can be configured to use either a cluster lock disk or a quorum server for locking purposes.
Lock Disk: a specific storage device available to all nodes in the cluster that holds the cluster lock
Quorum Server: a system separate from any node in the cluster; it runs a daemon that is responsible for giving a cluster lock to only one set of nodes in the cluster
Split-Brain Syndrome: the situation where two halves of a cluster each form a distinct instance of the same cluster. This is problematic because generally the package file systems are not capable of being mounted by multiple systems simultaneously. In addition, package IP addresses should only be active on a single node, which would not be the case when a cluster experiences a split-brain syndrome.
