YARN-4546. ResourceManager crash due to scheduling opportunity overflow. Contributed by Jason Lowe.

hadoop-yarn-project/CHANGES.txt

@@ -1215,6 +1215,9 @@ Release 2.8.0 - UNRELEASED
     YARN-1382. Remove unusableRMNodesConcurrentSet (never used) in NodeListManager
     to get rid of memory leak. (Rohith Sharma K S via junping_du)
 
+    YARN-4546. ResourceManager crash due to scheduling opportunity overflow.
+    (Jason Lowe via junping_du)
+
 Release 2.7.3 - UNRELEASED
 
   INCOMPATIBLE CHANGES
@@ -1270,6 +1273,9 @@ Release 2.7.3 - UNRELEASED
 
     YARN-4510. Fix SLS startup failure caused by NPE. (Bibin A Chundatt via wangda)
 
+    YARN-4546. ResourceManager crash due to scheduling opportunity overflow.
+    (Jason Lowe via junping_du)
+
 Release 2.7.2 - UNRELEASED
 
   INCOMPATIBLE CHANGES
@@ -2130,6 +2136,9 @@ Release 2.6.4 - UNRELEASED
   YARN-3697. FairScheduler: ContinuousSchedulingThread can fail to shutdown.
   (Zhihai Xu via kasha)
 
+  YARN-4546. ResourceManager crash due to scheduling opportunity overflow.
+  (Jason Lowe via junping_du)
+
 Release 2.6.3 - 2015-12-17
 
   INCOMPATIBLE CHANGES

hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/SchedulerApplicationAttempt.java

@@ -625,8 +625,10 @@ public synchronized int addMissedNonPartitionedRequestSchedulingOpportunity(
 
   
   public synchronized void addSchedulingOpportunity(Priority priority) {
-    schedulingOpportunities.setCount(priority,
+    int count = schedulingOpportunities.count(priority);
-        schedulingOpportunities.count(priority) + 1);
+    if (count < Integer.MAX_VALUE) {
+      schedulingOpportunities.setCount(priority, count + 1);
+    }
   }
   
   public synchronized void subtractSchedulingOpportunity(Priority priority) {
@@ -661,6 +663,11 @@ public synchronized void resetSchedulingOpportunities(Priority priority,
     schedulingOpportunities.setCount(priority, 0);
   }
 
+  @VisibleForTesting
+  void setSchedulingOpportunities(Priority priority, int count) {
+    schedulingOpportunities.setCount(priority, count);
+  }
+
   synchronized AggregateAppResourceUsage getRunningAggregateAppResourceUsage() {
     long currentTimeMillis = System.currentTimeMillis();
     // Don't walk the whole container list if the resources were computed

hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/TestSchedulerApplicationAttempt.java

@@ -249,4 +249,26 @@ public void testAppPercentagesOnswitch() throws Exception {
     assertEquals(0.0f, app.getResourceUsageReport().getClusterUsagePercentage(),
         0.0f);
   }
+
+  @Test
+  public void testSchedulingOpportunityOverflow() throws Exception {
+    ApplicationAttemptId attemptId = createAppAttemptId(0, 0);
+    Queue queue = createQueue("test", null);
+    RMContext rmContext = mock(RMContext.class);
+    when(rmContext.getEpoch()).thenReturn(3L);
+    SchedulerApplicationAttempt app = new SchedulerApplicationAttempt(
+        attemptId, "user", queue, queue.getActiveUsersManager(), rmContext);
+    Priority priority = Priority.newInstance(1);
+    assertEquals(0, app.getSchedulingOpportunities(priority));
+    app.addSchedulingOpportunity(priority);
+    assertEquals(1, app.getSchedulingOpportunities(priority));
+    // verify the count is capped at MAX_VALUE and does not overflow
+    app.setSchedulingOpportunities(priority, Integer.MAX_VALUE - 1);
+    assertEquals(Integer.MAX_VALUE - 1,
+        app.getSchedulingOpportunities(priority));
+    app.addSchedulingOpportunity(priority);
+    assertEquals(Integer.MAX_VALUE, app.getSchedulingOpportunities(priority));
+    app.addSchedulingOpportunity(priority);
+    assertEquals(Integer.MAX_VALUE, app.getSchedulingOpportunities(priority));
+  }
 }