[SPARK-4498][core] Don't transition ExecutorInfo to RUNNING until Driver adds Executor

The ExecutorInfo only reaches the RUNNING state if the Driver is alive to send the ExecutorStateChanged message to master.  Else, appInfo.resetRetryCount() is never called and failing Executors will eventually exceed ApplicationState.MAX_NUM_RETRY, resulting in the application being removed from the master's accounting.

JoshRosen

Author: Mark Hamstra <[email protected]>

Closes #3550 from markhamstra/SPARK-4498 and squashes the following commits:

8f543b1 [Mark Hamstra] Don't transition ExecutorInfo to RUNNING until Executor is added by Driver

Author: markhamstra

core/src/main/scala/org/apache/spark/deploy/client/AppClient.scala

@@ -134,6 +134,7 @@ private[spark] class AppClient(
         val fullId = appId + "/" + id
         logInfo("Executor added: %s on %s (%s) with %d cores".format(fullId, workerId, hostPort,
           cores))
+        master ! ExecutorStateChanged(appId, id, ExecutorState.RUNNING, None, None)
         listener.executorAdded(fullId, workerId, hostPort, cores, memory)
 
       case ExecutorUpdated(id, state, message, exitStatus) =>

core/src/main/scala/org/apache/spark/deploy/worker/ExecutorRunner.scala

@@ -144,8 +144,6 @@ private[spark] class ExecutorRunner(
       Files.write(header, stderr, UTF_8)
       stderrAppender = FileAppender(process.getErrorStream, stderr, conf)
 
-      state = ExecutorState.RUNNING
-      worker ! ExecutorStateChanged(appId, execId, state, None, None)
       // Wait for it to exit; executor may exit with code 0 (when driver instructs it to shutdown)
       // or with nonzero exit code
       val exitCode = process.waitFor()