The state of the cluster holds all of the configuration and internal data.
The cluster state information is held in nonvolatile memory. If the mainline power fails, the uninterruptible power supply units maintain the internal power long enough for the cluster state information to be stored on the internal disk drive of each node. If a power failure occurs, the write cache data and configuration information that is held in memory is stored on the internal disk drive of the node. If the partner node is still online, it attempts to flush the cache and continue operation with the write cache disabled.
Figure 1 shows an example of a cluster that contains four nodes. The cluster state shown in the grey box does not actually exist. Instead, each node in the cluster maintains an identical copy of the cluster state. When a change is made to the configuration or internal cluster data, the same change is applied to all nodes.
The cluster contains a single node that is elected as the configuration node. The configuration node can be thought of as the node that controls the updating of cluster state. For example, a user request is made (1), that results in a change being made to the configuration. The configuration node controls updates to the cluster (2). The configuration node then forwards the change to all nodes (including Node 1), and they all make the state-change at the same point in time (3). Using this state-driven model of clustering ensures that all nodes in the cluster know the exact cluster state at any one time. If the configuration node fails, the cluster can elect a new node to take over its responsibilities.
