You can start a new node manager on a machine. The node is identified by the unqualified host name of the machine where the command is run. Alternatively, a name can be provided as an argument to the start-node command. Note that node names must be unique across the cluster.

To simplify node recovery, starting a node manager using an already registered node name replaces the existing node in the cluster.

To shut down the cluster, use the stop-cluster command. This stops all daemon processes, starting with the node managers and then Cluster Manager.

If only a single node must be removed from the cluster, use the stop-node command. This action shuts down the node manager on the machine where it is run. Providing a node name as an additional argument shuts down the specified node, which may be remote.

To maintain a cluster, we recommend you configure the machines running a cluster daemon to automatically start them when the machine is started. To do this, you can script the required commands above and run them during system startup. For scripting options, see your operating system documentation.

A node manager can be started before Cluster Manager. If the node manager cannot contact Cluster Manager to register during initialization, then it will reattempt registration a number of times before ending. These attempts reduce the possibility of a race condition between Cluster Manager and the node manager.