When the Service Manager on a node with the compute role starts, the Service Manager registers the node with the Resource Manager Service. Compute nodes use a heartbeat protocol to send periodic signals to the Resource Manager Service. The Resource Manager Service stores compute node details in memory. If the node stops sending heartbeat signals, the Resource Manager Service marks the node as unavailable and does not dispatch jobs to the node.
When you enable a Data Integration Service that runs on the grid, the Data Integration Service designates one node with the compute role as the master compute node. The Service Manager on the master compute node communicates with the Resource Manager Service to find an available worker compute node to run job requests.