Github user pwendell commented on a diff in the pull request:
https://github.com/apache/spark/pull/1056#discussion_r15506095
--- Diff: core/src/main/scala/org/apache/spark/executor/Executor.scala ---
@@ -348,4 +353,48 @@ private[spark] class Executor(
}
}
}
+
+ def stop() {
+ isStopped = true
+ threadPool.shutdown()
+ }
+
+ def startDriverHeartbeater() {
+ val interval = conf.getInt("spark.executor.heartbeatInterval", 2000)
+ val timeout = AkkaUtils.lookupTimeout(conf)
+ val retryAttempts = AkkaUtils.numRetries(conf)
+ val retryIntervalMs = AkkaUtils.retryWaitMs(conf)
+ val heartbeatReceiverRef =
AkkaUtils.makeDriverRef("HeartbeatReceiver", conf, env.actorSystem)
+
+ val t = new Thread() {
+ override def run() {
+ // Sleep a random interval so the heartbeats don't end up in sync
+ Thread.sleep(interval + (math.random * interval).asInstanceOf[Int])
+
+ while (!isStopped) {
+ val tasksMetrics = new ArrayBuffer[(Long, TaskMetrics)]()
+ for (taskRunner <- runningTasks.values()) {
+ if (taskRunner.running) {
+ Option(taskRunner.task).flatMap(_.metrics).foreach { metrics
=>
+ tasksMetrics += ((taskRunner.taskId, metrics))
+ }
+ }
+ }
+
+ val message = Heartbeat(executorId, tasksMetrics.toArray,
+ env.blockManager.blockManagerId)
+ val response =
AkkaUtils.askWithReply[HeartbeatResponse](message, heartbeatReceiverRef,
--- End diff --
This could cause executors to take a long time (e.g. minutes) to shut down
becuase the akka timeout could be high (i.e. if the driver cannot respond to
these heartbeats because it has shut down it's actor system). It depends a bit
on the order in which things shut down. Have you tested this at all on a
cluster? I wonder whether the driver actor system that respods to these
heartbeats shuts down before the executors do.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
