svn merge -c 1355124 FIXES: MAPREDUCE-4376. TestClusterMRNotification times out (Kihwal Lee via bobby)

git-svn-id: https://svn.apache.org/repos/asf/hadoop/common/branches/branch-0.23@1358418 13f79535-47bb-0310-9956-ffa450edef68

hadoop-mapreduce-project/CHANGES.txt

@@ -290,6 +290,8 @@ Release 0.23.3 - UNRELEASED
 
     MAPREDUCE-4372. Deadlock in Resource Manager (Devaraj K via bobby)
 
+    MAPREDUCE-4376. TestClusterMRNotification times out (Kihwal Lee via bobby)
+
 Release 0.23.2 - UNRELEASED
 
   INCOMPATIBLE CHANGES

hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-jobclient/src/test/java/org/apache/hadoop/mapred/NotificationTestCase.java

@@ -91,26 +91,22 @@ public abstract class NotificationTestCase extends HadoopTestCase {
 
   public static class NotificationServlet extends HttpServlet {
     public static int counter = 0;
+    public static int failureCounter = 0;
     private static final long serialVersionUID = 1L;
     
     protected void doGet(HttpServletRequest req, HttpServletResponse res)
       throws ServletException, IOException {
+      String queryString = req.getQueryString();
       switch (counter) {
         case 0:
-        {
-          assertTrue(req.getQueryString().contains("SUCCEEDED"));
-        }
-        break;
+          verifyQuery(queryString, "SUCCEEDED");
+          break;
         case 2:
-        {
-          assertTrue(req.getQueryString().contains("KILLED"));
-        }
-        break;
+          verifyQuery(queryString, "KILLED");
+          break;
         case 4:
-        {
-          assertTrue(req.getQueryString().contains("FAILED"));
-        }
-        break;
+          verifyQuery(queryString, "FAILED");
+          break;
       }
       if (counter % 2 == 0) {
         res.sendError(HttpServletResponse.SC_BAD_REQUEST, "forcing error");
@@ -120,6 +116,15 @@ public abstract class NotificationTestCase extends HadoopTestCase {
       }
       counter++;
     }
+
+    protected void verifyQuery(String query, String expected) 
+        throws IOException {
+      if (query.contains(expected)) {
+        return;
+      }
+      failureCounter++;
+      assertTrue("The request (" + query + ") does not contain " + expected, false);
+    }
   }
 
   private String getNotificationUrlTemplate() {
@@ -147,10 +152,12 @@ public abstract class NotificationTestCase extends HadoopTestCase {
   }
 
   public void testMR() throws Exception {
+
     System.out.println(launchWordCount(this.createJobConf(),
                                        "a b c d e f g h", 1, 1));
     Thread.sleep(2000);
     assertEquals(2, NotificationServlet.counter);
+    assertEquals(0, NotificationServlet.failureCounter);
     
     Path inDir = new Path("notificationjob/input");
     Path outDir = new Path("notificationjob/output");
@@ -168,12 +175,14 @@ public abstract class NotificationTestCase extends HadoopTestCase {
                                                 outDir).getID());
     Thread.sleep(2000);
     assertEquals(4, NotificationServlet.counter);
+    assertEquals(0, NotificationServlet.failureCounter);
     
     // run a job with FAILED status
     System.out.println(UtilsForTests.runJobFail(this.createJobConf(), inDir,
                                                 outDir).getID());
     Thread.sleep(2000);
     assertEquals(6, NotificationServlet.counter);
+    assertEquals(0, NotificationServlet.failureCounter);
   }
 
   private String launchWordCount(JobConf conf,

hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-jobclient/src/test/java/org/apache/hadoop/mapred/UtilsForTests.java

@@ -606,9 +606,14 @@ public class UtilsForTests {
     conf.setReducerClass(IdentityReducer.class);
     
     RunningJob job = UtilsForTests.runJob(conf, inDir, outDir);
+    long sleepCount = 0;
     while (!job.isComplete()) {
       try {
+        if (sleepCount > 300) { // 30 seconds
+          throw new IOException("Job didn't finish in 30 seconds");
+        }
         Thread.sleep(100);
+        sleepCount++;
       } catch (InterruptedException e) {
         break;
       }
@@ -626,9 +631,14 @@ public class UtilsForTests {
     conf.setMaxMapAttempts(1);
     
     RunningJob job = UtilsForTests.runJob(conf, inDir, outDir);
+    long sleepCount = 0;
     while (!job.isComplete()) {
       try {
+        if (sleepCount > 300) { // 30 seconds
+          throw new IOException("Job didn't finish in 30 seconds");
+        }
         Thread.sleep(100);
+        sleepCount++;
       } catch (InterruptedException e) {
         break;
       }
@@ -646,17 +656,27 @@ public class UtilsForTests {
     conf.setReducerClass(IdentityReducer.class);
     
     RunningJob job = UtilsForTests.runJob(conf, inDir, outDir);
+    long sleepCount = 0;
     while (job.getJobState() != JobStatus.RUNNING) {
       try {
+        if (sleepCount > 300) { // 30 seconds
+          throw new IOException("Job didn't finish in 30 seconds");
+        }
         Thread.sleep(100);
+        sleepCount++;
       } catch (InterruptedException e) {
         break;
       }
     }
     job.killJob();
+    sleepCount = 0;
     while (job.cleanupProgress() == 0.0f) {
       try {
+        if (sleepCount > 2000) { // 20 seconds
+          throw new IOException("Job cleanup didn't start in 20 seconds");
+        }
         Thread.sleep(10);
+        sleepCount++;
       } catch (InterruptedException ie) {
         break;
       }