Management services for each member machine in a DataFlow cluster are provided by a daemon known as the node manager. When a distributed graph executes, the node manager will be contacted to create an executor for the partitions that will be run on that node. As with the cluster manager, the node manager is involved only in the compilation of the graph; the node manager plays no role in the execution of the graph.

Nodes are uniquely identified by name with the cluster. Typically, this is the unqualified hostname of the machine where the node manager runs, but it also can be explicitly specified. When started, the node manager registers its name with the cluster manager; when stopped, it unregisters itself.