The connector remains in constant communication with the Tungsten Manager during operation. This enables the connector to respond to failures and errors, whether automatically identified, or manually triggered. For example, when a manual switch operation occurs, the manager communicates this information to all of the connectors. Each connector then responses according to the rules outline in Section 7.8, “Connector Operational States”.

The connector remains in communication with one, and only one, manager at a time. If the manager becomes unavailable, connector tries to communicate another manager within the dataservice.

Communication from the manager to the connectors is made in parallel using multiple threads, this ensures that all connectors are made aware of a change in the topology of the cluster at the same time, rather than a round-robin or staged distribution. When a change has been requested, the manager waits for a response from the cluster before confirming that switch and operational change has taken place.

Communication between managers and the connectors is handled on ports 11999 (managed by --router-gateway-port) and 12000 (managed by --mgr-rmi-remote-port). The connection is used to exchange cluster status and individual datasource availability as identified by the manager so that decisions about active connections can be effectively made by the connector.

In the event that the connection between the connector and the manager is broken, the connector enters a failsafe mode called On Hold. In this state, connections to and from the connector and datasources will continue as normal until a timeout, configured by the delayBeforeOfflineIfNoManager property, is reached. By default, this timeout is 600 seconds (10 minutes). Once the timeout has been reached, the connector reaches the Offline state.

All of the information about the dataservice, including the other nodes, topology and individual node states and roles are entirely determined by the Connector by requesting this information from the Manager. No on-disk record or description is stored or stored, created, or read by the Connector. When the Connector is first started, it connects to a manager and requests the full cluster configuration.

If the Connector cannot communicate with a manager, the connector remains in the Offline state until a manager can be reached.