LiveListenerBus

LiveListenerBus is an event bus to announce application-wide events in a Spark application. It asynchronously passes listener events to registered SparkListeners (based on spark.extraListeners configuration property).

spark sparklistener event senders
Figure 1. LiveListenerBus, SparkListenerEvents, and Senders

LiveListenerBus is a single-JVM SparkListenerBus that uses listenerThread to poll events. Emitters are supposed to use post method to post SparkListenerEvent events.

The event queue is java.util.concurrent.LinkedBlockingQueue with capacity of 10000 SparkListenerEvent events.

Creating Instance

LiveListenerBus takes the following to be created:

LiveListenerBus is created (and started) when SparkContext is requested to initialize.

Starting LiveListenerBus

start(
  sc: SparkContext): Unit

start starts processing events.

Internally, it saves the input SparkContext for later use and starts listenerThread. It makes sure that it only happens when LiveListenerBus has not been started before (i.e. started is disabled).

If however LiveListenerBus has already been started, a IllegalStateException is thrown:

[name] already started!

Posting SparkListenerEvent Event

post(
  event: SparkListenerEvent): Unit

post puts the input event onto the internal eventQueue queue and releases the internal eventLock semaphore. If the event placement was not successful (and it could happen since it is tapped at 10000 events) onDropEvent method is called.

The event publishing is only possible when stopped flag has been enabled.

FIXME Who’s enabling the stopped flag and when/why?

If LiveListenerBus has been stopped, the following ERROR appears in the logs:

[name] has already stopped! Dropping event [event]

Event Dropped Callback

onDropEvent(
  event: SparkListenerEvent): Unit

onDropEvent is called when no further events can be added to the internal eventQueue queue (while posting a SparkListenerEvent event).

It simply prints out the following ERROR message to the logs and ensures that it happens only once.

ERROR Dropping SparkListenerEvent because no remaining room in event queue. This likely means one of the SparkListeners is too slow and cannot keep up with the rate at which tasks are being started by the scheduler.
It uses the internal logDroppedEvent atomic variable to track the state.

Stopping LiveListenerBus

stop(): Unit

stop releases the internal eventLock semaphore and waits until listenerThread dies. It can only happen after all events were posted (and polling eventQueue gives nothing).

It checks that started flag is enabled (i.e. true) and throws a IllegalStateException otherwise.

Attempted to stop [name] that has not yet started!

stopped flag is enabled.

listenerThread for Event Polling

LiveListenerBus uses SparkListenerBus single-daemon thread that ensures that the polling events from the event queue is only after the listener was started and only one event at a time.

FIXME There is some logic around no events in the queue.

Registering SparkListenerInterface with Application Status Queue

addToStatusQueue(
  listener: SparkListenerInterface): Unit

addToStatusQueue simply adds the SparkListenerInterface to eventLog queue.

addToStatusQueue is used when…​FIXME

Registering SparkListenerInterface with Queue

addToQueue(
  listener: SparkListenerInterface,
  queue: String): Unit

addToQueue…​FIXME

addToQueue is used when…​FIXME