public void OnNext(REEFMessage value)
{
if (value != null && value.evaluatorControl != null)
{
Logger.Log(Level.Verbose, "Received a REEFMessage with EvaluatorControl");
Handle(value.evaluatorControl);
}
}
public void OnNext(REEFMessage value)
{
if (value != null && value.evaluatorControl != null)
{
LOGGER.Log(Level.Info, "Received a REEFMessage with EvaluatorControl");
Handle(value.evaluatorControl);
}
}
public void Send(EvaluatorHeartbeatProto evaluatorHeartbeatProto)
{
lock (_queuedHeartbeats)
{
// Do not send a heartbeat if Evaluator has already signaled that it was done.
if (_isCompletedHeartbeatQueued)
{
LOGGER.Log(Level.Warning, "Evaluator trying to schedule a heartbeat after a completed heartbeat has already been scheduled or sent.");
return;
}
if (IsEvaluatorStateCompleted(evaluatorHeartbeatProto.evaluator_status.state))
{
_isCompletedHeartbeatQueued = true;
}
if (_evaluatorSettings.OperationState == EvaluatorOperationState.RECOVERY)
{
LOGGER.Log(Level.Warning, string.Format(CultureInfo.InvariantCulture, "In RECOVERY mode, heartbeat queued as [{0}]. ", evaluatorHeartbeatProto));
_queuedHeartbeats.Enqueue(evaluatorHeartbeatProto);
return;
}
// NOT during recovery, try to send
REEFMessage payload = new REEFMessage(evaluatorHeartbeatProto);
try
{
_observer.OnNext(payload);
_heartbeatFailures = 0; // reset failure counts if we are having intermidtten (not continuous) failures
}
catch (Exception e)
{
if (evaluatorHeartbeatProto.task_status == null || evaluatorHeartbeatProto.task_status.state != State.RUNNING)
{
Utilities.Diagnostics.Exceptions.Throw(e, "Lost communications to driver when no task is running, recovery NOT supported for such scenario", LOGGER);
}
_heartbeatFailures++;
_queuedHeartbeats.Enqueue(evaluatorHeartbeatProto);
LOGGER.Log(Level.Error, string.Format(CultureInfo.InvariantCulture, "Sending heartbeat to driver experienced #{0} failure. Hearbeat queued as: [{1}]. ", _heartbeatFailures, evaluatorHeartbeatProto), e);
if (_heartbeatFailures >= _maxHeartbeatRetries)
{
LOGGER.Log(Level.Warning, "Heartbeat communications to driver reached max of {0} failures. Driver is considered dead/unreachable", _heartbeatFailures);
LOGGER.Log(Level.Info, "=========== Entering RECOVERY mode. ===========");
ContextManager.HandleDriverConnectionMessage(new DriverConnectionMessageImpl(DriverConnectionState.Disconnected));
LOGGER.Log(Level.Info, "instantiate driver reconnect implementation: " + _driverConnection);
_evaluatorSettings.OperationState = EvaluatorOperationState.RECOVERY;
// clean heartbeat failure
_heartbeatFailures = 0;
}
}
}
}
public void Send(EvaluatorHeartbeatProto evaluatorHeartbeatProto)
{
lock (_queuedHeartbeats)
{
if (_evaluatorSettings.OperationState == EvaluatorOperationState.RECOVERY)
{
LOGGER.Log(Level.Warning, string.Format(CultureInfo.InvariantCulture, "In RECOVERY mode, heartbeat queued as [{0}]. ", evaluatorHeartbeatProto));
_queuedHeartbeats.Enqueue(evaluatorHeartbeatProto);
return;
}
// NOT during recovery, try to send
REEFMessage payload = new REEFMessage(evaluatorHeartbeatProto);
try
{
_observer.OnNext(payload);
_heartbeatFailures = 0; // reset failure counts if we are having intermidtten (not continuous) failures
}
catch (Exception e)
{
if (evaluatorHeartbeatProto.task_status == null || evaluatorHeartbeatProto.task_status.state != State.RUNNING)
{
Org.Apache.REEF.Utilities.Diagnostics.Exceptions.Throw(e, "Lost communications to driver when no task is running, recovery NOT supported for such scenario", LOGGER);
}
_heartbeatFailures++;
_queuedHeartbeats.Enqueue(evaluatorHeartbeatProto);
LOGGER.Log(Level.Error, string.Format(CultureInfo.InvariantCulture, "Sending heartbeat to driver experienced #{0} failure. Hearbeat queued as: [{1}]. ", _heartbeatFailures, evaluatorHeartbeatProto), e);
if (_heartbeatFailures >= _maxHeartbeatRetries)
{
LOGGER.Log(Level.Warning, "Heartbeat communications to driver reached max of {0} failures. Driver is considered dead/unreachable", _heartbeatFailures);
LOGGER.Log(Level.Info, "=========== Entering RECOVERY mode. ===========");
ContextManager.HandleDriverConnectionMessage(new DriverConnectionMessageImpl(DriverConnectionState.Disconnected));
try
{
_driverConnection = _evaluatorSettings.EvaluatorInjector.GetInstance <IDriverConnection>();
}
catch (Exception ex)
{
Org.Apache.REEF.Utilities.Diagnostics.Exceptions.CaughtAndThrow(ex, Level.Error, "Failed to inject the driver reconnect implementation", LOGGER);
}
LOGGER.Log(Level.Info, "instantiate driver reconnect implementation: " + _driverConnection);
_evaluatorSettings.OperationState = EvaluatorOperationState.RECOVERY;
// clean heartbeat failure
_heartbeatFailures = 0;
}
}
}
}
