[SPARK-937] adding EXITED executor state and not relaunching cleanly exited executors

There seems to be 2 issues.

1. When job is done, driver asks executor to shutdown. However, this clean exit was assigned FAILED executor state by Worker. I introduced EXITED executor state for executors who voluntarily exit (both normal and abnormal exit depending on the exit code).

2. When Master gets notified an executor has exited, it launches another one to replace it, regardless of reason why the executor had exited. When the reason was job has finished, the unnecessary replacement got subsequently killed when App disassociates. This launching and killing of unnecessary executors shows up in the log and is confusing to users. I added check for executor exit status and avoid launching (and subsequent killing) of unnecessary replacements when executors exit cleanly.

One could ask the scheduler to tell Master job is done so that Master wouldn't launch the replacement executor. However, there is a race condition between App telling Master job is done and Worker telling Master an executor had exited. There is no guarantee the former will happen before the later. Instead, I chose to check the exit code when executor exits. If the exit code is 0, I assume executor has been asked to shutdown by driver and Master will not launch replacements.

Due to race condition, it could also happen that (although didn't happen on my local cluster), Master detects App disassociation event before the executor exits by itself. In such cases, the executor will be rightfully killed and labeled as KILLED, while the App state will show FINISHED.

Author: Kan Zhang <[email protected]>

Closes #306 from kanzhang/SPARK-1118 and squashes the following commits:

cb0cc86 [Kan Zhang] [SPARK-937] adding EXITED executor state and not relaunching cleanly exited executors

Author: kanzhang

core/src/main/scala/org/apache/spark/deploy/ExecutorState.scala

@@ -19,9 +19,9 @@ package org.apache.spark.deploy
 
 private[spark] object ExecutorState extends Enumeration {
 
-  val LAUNCHING, LOADING, RUNNING, KILLED, FAILED, LOST = Value
+  val LAUNCHING, LOADING, RUNNING, KILLED, FAILED, LOST, EXITED = Value
 
   type ExecutorState = Value
 
-  def isFinished(state: ExecutorState): Boolean = Seq(KILLED, FAILED, LOST).contains(state)
+  def isFinished(state: ExecutorState): Boolean = Seq(KILLED, FAILED, LOST, EXITED).contains(state)
 }

core/src/main/scala/org/apache/spark/deploy/master/Master.scala

@@ -303,10 +303,11 @@ private[spark] class Master(
             appInfo.removeExecutor(exec)
             exec.worker.removeExecutor(exec)
 
+            val normalExit = exitStatus.exists(_ == 0)
             // Only retry certain number of times so we don't go into an infinite loop.
-            if (appInfo.incrementRetryCount < ApplicationState.MAX_NUM_RETRY) {
+            if (!normalExit && appInfo.incrementRetryCount < ApplicationState.MAX_NUM_RETRY) {
               schedule()
-            } else {
+            } else if (!normalExit) {
               logError("Application %s with ID %s failed %d times, removing it".format(
                 appInfo.desc.name, appInfo.id, appInfo.retryCount))
               removeApplication(appInfo, ApplicationState.FAILED)

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

@@ -154,11 +154,10 @@ private[spark] class ExecutorRunner(
       Files.write(header, stderr, Charsets.UTF_8)
       stderrAppender = FileAppender(process.getErrorStream, stderr, conf)
 
-      // Wait for it to exit; this is actually a bad thing if it happens, because we expect to run
-      // long-lived processes only. However, in the future, we might restart the executor a few
-      // times on the same machine.
+      // 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()
-      state = ExecutorState.FAILED
+      state = ExecutorState.EXITED
       val message = "Command exited with code " + exitCode
       worker ! ExecutorStateChanged(appId, execId, state, Some(message), Some(exitCode))
     } catch {