2. RunGovernor state diagram

The different RunGovernor states can be generalized into three different groups which are also color coded in the diagram.

1. Initializing and starting the experiment run (green)

2. Shutting down the experiment run and clean up (blue)

3. Error handling (orange)

2.1. Initialization and Startup

When a new instance of the RunGovernor gets launched by the Executor it always gets launched as RGSPristine which as the name suggests resembles a fresh and unused RunGovernor this state however is almost instantly changed to RGSInitializing and then to RGSTransceiving. During a normal experiment run the RunGovernor would now stay in this state till it receives an ExperimentRunStartRequest from the Executor upon receiving it the RunGovernor will attempt to start the experiment run and changes its state to RGSStartingSetup during this state the RunGovernor checks if it can run the experiment at all and if it’s running already when it can run the experiment its state changes to RGSStartingSimControllers, RGSStartingEnvConductors and RGSStartingAgentConductors in that order to create all processes for SimulationControllers, EnvConductors and AgentConductors that are needed for the experiment run. Once all those processes are created the RunGovernor changes states to RGSStartingRun which simply starts all previously created processes. At this point the initialization and start up of the experiment run are done and the RunGovernor changes back into the RGSTransceiving state.

2.2. Shutdown and clean up

If a RunGovernor in the state RGSTransceiving receives an ExperimentRunShutdownRequest it will change its state to RGSStoppingSimulation and then send a SimulationStopRequest to the SimulationController of the experiment run the SimulationController now handles the shutdown of all Conductors, Enviornments and Agents and responds with a SimulationStopResponse. The RunGovernor then shuts down the SimulationController and sends an ExperimentRunShutdownResponse to the Executor afterwards it changes its state to RGSStoppingTransceiving which stops it from receiving new requests, then it enters the state RGSStoppingRun which stops all remaining processes of Conductors and SimulationControllers, if the experiment has a next phase the RunGovernor will now enter RGSStartingSetup again. Afterwards the states RGSFinalizing and RGSDone are entered. Finalizing terminates all subprocesses that might still run and with RGSDone the RunGovernor terminates itself.

2.3. Error handling

The plethora of errors that can occur while the RunGovernor runs are handled by four different Error handling states. However only three of them do actually have functionality at the moment since RGSErrorHandlingInitializing can not be entered. Another state RGSHandlingDeadChildren is only entered from the error state RGSErrorHandlingTransceiving and basically only checks if a premature task death has occured, in this case the RunGovernor will jump straight into the RGSStoppingTransceiving state, or not in which case the RunGovernor will initialize shutdown by entering the state RGSStoppingSimulation.

Most errors that are handled in the state RGSErrorHandlingTransceiving will result in the RunGovernor reentering the state RGSTransceiving with the exception of SignalInterruptError, which will cause the RunGovernor to stop the current experiment phase by entering RGSStoppingSimulation, and DeadChildrenRisingAsZombiesError, which lets the RunGovernor switch to the RGSHandlingDeadChildren state.

The state RGSErrorHandlingStarting causes the RunGovernor to either change its state back to RGSTransceiving if the error that was caught was an ExperimentAlreadyRunningError or to RGSStoppingTransceiving if the error was an ExperimentSetupFailedError all other errors that are handled by RGSErrorHandlingStarting will let the RunGovernor change to the RGSStoppingSimulation state.