YARN-6486. FairScheduler: Deprecate continuous scheduling. (Contributed by Wilfred Spiegelenburg)

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

@@ -943,6 +943,7 @@ public class FSAppAttempt extends SchedulerApplicationAttempt
     return false;
   }
 
+  @SuppressWarnings("deprecation")
   private Resource assignContainer(FSSchedulerNode node, boolean reserved) {
     if (LOG.isTraceEnabled()) {
       LOG.trace("Node offered to app: " + getName() + " reserved: " + reserved);

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

@@ -41,6 +41,7 @@ import org.apache.hadoop.metrics2.lib.MutableRate;
 @Metrics(context="fairscheduler-op-durations")
 public class FSOpDurations implements MetricsSource {
 
+  @Deprecated
   @Metric("Duration for a continuous scheduling run")
   MutableRate continuousSchedulingRun;
 
@@ -90,6 +91,7 @@ public class FSOpDurations implements MetricsSource {
     registry.snapshot(collector.addRecord(registry.info()), all);
   }
 
+  @Deprecated
   public void addContinuousSchedulingRunDuration(long value) {
     continuousSchedulingRun.add(value);
   }

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

@@ -48,6 +48,7 @@ class FSPreemptionThread extends Thread {
   private final Timer preemptionTimer;
   private final Lock schedulerReadLock;
 
+  @SuppressWarnings("deprecation")
   FSPreemptionThread(FairScheduler scheduler) {
     setDaemon(true);
     setName("FSPreemptionThread");

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

@@ -153,6 +153,7 @@ public class FairScheduler extends
   private final int UPDATE_DEBUG_FREQUENCY = 25;
   private int updatesToSkipForDebug = UPDATE_DEBUG_FREQUENCY;
 
+  @Deprecated
   @VisibleForTesting
   Thread schedulingThread;
 
@@ -167,15 +168,19 @@ public class FairScheduler extends
 
   protected boolean sizeBasedWeight; // Give larger weights to larger jobs
   // Continuous Scheduling enabled or not
+  @Deprecated
   protected boolean continuousSchedulingEnabled;
   // Sleep time for each pass in continuous scheduling
+  @Deprecated
   protected volatile int continuousSchedulingSleepMs;
   // Node available resource comparator
   private Comparator<FSSchedulerNode> nodeAvailableResourceComparator =
           new NodeAvailableResourceComparator();
   protected double nodeLocalityThreshold; // Cluster threshold for node locality
   protected double rackLocalityThreshold; // Cluster threshold for rack locality
+  @Deprecated
   protected long nodeLocalityDelayMs; // Delay for node locality
+  @Deprecated
   protected long rackLocalityDelayMs; // Delay for rack locality
   protected boolean assignMultiple; // Allocate multiple containers per
                                     // heartbeat
@@ -284,6 +289,7 @@ public class FairScheduler extends
    * Thread which attempts scheduling resources continuously,
    * asynchronous to the node heartbeats.
    */
+  @Deprecated
   private class ContinuousSchedulingThread extends Thread {
 
     @Override
@@ -389,18 +395,44 @@ public class FairScheduler extends
     return rackLocalityThreshold;
   }
 
+  /**
+   * Delay in milliseconds for locality fallback node to rack.
+   * @deprecated linked to {@link #isContinuousSchedulingEnabled} deprecation
+   * @return delay in ms
+   */
+  @Deprecated
   public long getNodeLocalityDelayMs() {
     return nodeLocalityDelayMs;
   }
 
+  /**
+   * Delay in milliseconds for locality fallback rack to other.
+   * @deprecated linked to {@link #isContinuousSchedulingEnabled} deprecation
+   * @return delay in ms
+   */
+  @Deprecated
   public long getRackLocalityDelayMs() {
     return rackLocalityDelayMs;
   }
 
+  /**
+   * Whether continuous scheduling is turned on.
+   * @deprecated Continuous scheduling should not be turned ON. It is
+   * deprecated because it can cause scheduler slowness due to locking issues.
+   * Schedulers should use assignmultiple as a replacement.
+   * @return whether continuous scheduling is enabled
+   */
+  @Deprecated
   public boolean isContinuousSchedulingEnabled() {
     return continuousSchedulingEnabled;
   }
 
+  /**
+   * The sleep time of the continuous scheduler thread.
+   * @deprecated linked to {@link #isContinuousSchedulingEnabled} deprecation
+   * @return sleep time in ms
+   */
+  @Deprecated
   public int getContinuousSchedulingSleepMs() {
     return continuousSchedulingSleepMs;
   }
@@ -894,6 +926,7 @@ public class FairScheduler extends
     }
   }
 
+  @Deprecated
   void continuousSchedulingAttempt() throws InterruptedException {
     long start = getClock().getTime();
     List<FSSchedulerNode> nodeIdList;
@@ -1253,6 +1286,7 @@ public class FairScheduler extends
     this.rmContext = rmContext;
   }
 
+  @SuppressWarnings("deprecation")
   private void initScheduler(Configuration conf) throws IOException {
     try {
       writeLock.lock();
@@ -1299,6 +1333,10 @@ public class FairScheduler extends
       }
 
       if (continuousSchedulingEnabled) {
+        // Contiuous scheduling is deprecated log it on startup
+        LOG.warn("Continuous scheduling is turned ON. It is deprecated " +
+            "because it can cause scheduler slowness due to locking issues. " +
+            "Schedulers should use assignmultiple as a replacement.");
         // start continuous scheduling thread
         schedulingThread = new ContinuousSchedulingThread();
         schedulingThread.setName("FairSchedulerContinuousScheduling");
@@ -1374,6 +1412,7 @@ public class FairScheduler extends
     super.serviceStart();
   }
 
+  @SuppressWarnings("deprecation")
   @Override
   public void serviceStop() throws Exception {
     try {

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

@@ -97,20 +97,49 @@ public class FairSchedulerConfiguration extends Configuration {
   protected static final float  DEFAULT_LOCALITY_THRESHOLD_RACK =
 		  DEFAULT_LOCALITY_THRESHOLD;
 
-  /** Delay for node locality. */
-  protected static final String LOCALITY_DELAY_NODE_MS = CONF_PREFIX + "locality-delay-node-ms";
+  /**
+   * Delay for node locality.
+   * @deprecated Continuous scheduling is known to cause locking issue inside
+   * Only used when {@link #CONTINUOUS_SCHEDULING_ENABLED} is enabled
+   */
+  @Deprecated
+  protected static final String LOCALITY_DELAY_NODE_MS = CONF_PREFIX +
+      "locality-delay-node-ms";
+  @Deprecated
   protected static final long DEFAULT_LOCALITY_DELAY_NODE_MS = -1L;
 
-  /** Delay for rack locality. */
-  protected static final String LOCALITY_DELAY_RACK_MS = CONF_PREFIX + "locality-delay-rack-ms";
+  /**
+   * Delay for rack locality.
+   * @deprecated Continuous scheduling is known to cause locking issue inside
+   * Only used when {@link #CONTINUOUS_SCHEDULING_ENABLED} is enabled
+   */
+  @Deprecated
+  protected static final String LOCALITY_DELAY_RACK_MS = CONF_PREFIX +
+      "locality-delay-rack-ms";
+  @Deprecated
   protected static final long DEFAULT_LOCALITY_DELAY_RACK_MS = -1L;
 
-  /** Enable continuous scheduling or not. */
-  protected static final String CONTINUOUS_SCHEDULING_ENABLED = CONF_PREFIX + "continuous-scheduling-enabled";
+  /**
+   * Enable continuous scheduling or not.
+   * @deprecated Continuous scheduling is known to cause locking issue inside
+   * the scheduler in larger cluster, more than 100 nodes, use
+   * {@link #ASSIGN_MULTIPLE} to improve  container allocation ramp up.
+   */
+  @Deprecated
+  protected static final String CONTINUOUS_SCHEDULING_ENABLED = CONF_PREFIX +
+      "continuous-scheduling-enabled";
+  @Deprecated
   protected static final boolean DEFAULT_CONTINUOUS_SCHEDULING_ENABLED = false;
 
-  /** Sleep time of each pass in continuous scheduling (5ms in default) */
-  protected static final String CONTINUOUS_SCHEDULING_SLEEP_MS = CONF_PREFIX + "continuous-scheduling-sleep-ms";
+  /**
+   * Sleep time of each pass in continuous scheduling (5ms in default).
+   * @deprecated Continuous scheduling is known to cause locking issue inside
+   * Only used when {@link #CONTINUOUS_SCHEDULING_ENABLED} is enabled
+   */
+  @Deprecated
+  protected static final String CONTINUOUS_SCHEDULING_SLEEP_MS = CONF_PREFIX +
+      "continuous-scheduling-sleep-ms";
+  @Deprecated
   protected static final int DEFAULT_CONTINUOUS_SCHEDULING_SLEEP_MS = 5;
 
   /** Whether preemption is enabled. */
@@ -282,18 +311,45 @@ public class FairSchedulerConfiguration extends Configuration {
     return getFloat(LOCALITY_THRESHOLD_RACK, DEFAULT_LOCALITY_THRESHOLD_RACK);
   }
 
+  /**
+   * Whether continuous scheduling is turned on.
+   * @deprecated use {@link #ASSIGN_MULTIPLE} to improve container allocation
+   * ramp up.
+   * @return whether continuous scheduling is enabled
+   */
+  @Deprecated
   public boolean isContinuousSchedulingEnabled() {
-    return getBoolean(CONTINUOUS_SCHEDULING_ENABLED, DEFAULT_CONTINUOUS_SCHEDULING_ENABLED);
+    return getBoolean(CONTINUOUS_SCHEDULING_ENABLED,
+        DEFAULT_CONTINUOUS_SCHEDULING_ENABLED);
   }
 
+  /**
+   * The sleep time of the continuous scheduler thread.
+   * @deprecated linked to {@link #CONTINUOUS_SCHEDULING_ENABLED} deprecation
+   * @return sleep time in ms
+   */
+  @Deprecated
   public int getContinuousSchedulingSleepMs() {
-    return getInt(CONTINUOUS_SCHEDULING_SLEEP_MS, DEFAULT_CONTINUOUS_SCHEDULING_SLEEP_MS);
+    return getInt(CONTINUOUS_SCHEDULING_SLEEP_MS,
+        DEFAULT_CONTINUOUS_SCHEDULING_SLEEP_MS);
   }
 
+  /**
+   * Delay in milliseconds for locality fallback node to rack.
+   * @deprecated linked to {@link #CONTINUOUS_SCHEDULING_ENABLED} deprecation
+   * @return delay in ms
+   */
+  @Deprecated
   public long getLocalityDelayNodeMs() {
     return getLong(LOCALITY_DELAY_NODE_MS, DEFAULT_LOCALITY_DELAY_NODE_MS);
   }
 
+  /**
+   * Delay in milliseconds for locality fallback rack to other.
+   * @deprecated linked to {@link #CONTINUOUS_SCHEDULING_ENABLED} deprecation
+   * @return delay in ms
+   */
+  @Deprecated
   public long getLocalityDelayRackMs() {
     return getLong(LOCALITY_DELAY_RACK_MS, DEFAULT_LOCALITY_DELAY_RACK_MS);
   }

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

@@ -61,10 +61,12 @@ import java.util.Map;
 import java.util.Set;
 import java.util.concurrent.ThreadLocalRandom;
 
+@Deprecated
 public class TestContinuousScheduling extends FairSchedulerTestBase {
   private ControlledClock mockClock;
   private static int delayThresholdTimeMs = 1000;
 
+  @SuppressWarnings("deprecation")
   @Override
   public Configuration createConfiguration() {
     Configuration conf = super.createConfiguration();
@@ -77,6 +79,7 @@ public class TestContinuousScheduling extends FairSchedulerTestBase {
     return conf;
   }
 
+  @SuppressWarnings("deprecation")
   @Before
   public void setup() {
     mockClock = new ControlledClock();
@@ -187,6 +190,7 @@ public class TestContinuousScheduling extends FairSchedulerTestBase {
     Assert.assertEquals(2, nodes.size());
   }
 
+  @SuppressWarnings("deprecation")
   @Test
   public void testWithNodeRemoved() throws Exception {
     // Disable continuous scheduling, will invoke continuous
@@ -240,6 +244,7 @@ public class TestContinuousScheduling extends FairSchedulerTestBase {
     }
   }
 
+  @SuppressWarnings("deprecation")
   @Test
   public void testInterruptedException()
           throws Exception {
@@ -284,6 +289,7 @@ public class TestContinuousScheduling extends FairSchedulerTestBase {
     }
   }
 
+  @SuppressWarnings("deprecation")
   @Test
   public void testSchedulerThreadLifeCycle() throws InterruptedException {
     scheduler.start();
@@ -300,6 +306,7 @@ public class TestContinuousScheduling extends FairSchedulerTestBase {
     assertNotEquals("The Scheduling thread is still alive", 0, numRetries);
   }
 
+  @SuppressWarnings("deprecation")
   @Test
   public void TestNodeAvailableResourceComparatorTransitivity() {
     ClusterNodeTracker<FSSchedulerNode> clusterNodeTracker =
@@ -384,6 +391,7 @@ public class TestContinuousScheduling extends FairSchedulerTestBase {
     assertEquals(delayThresholdTimeMs, initSchedulerTime);
   }
 
+  @SuppressWarnings("deprecation")
   private void triggerSchedulingAttempt() throws InterruptedException {
     Thread.sleep(
         2 * scheduler.getConf().getContinuousSchedulingSleepMs());

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

@@ -194,6 +194,7 @@ public class TestFairScheduler extends FairSchedulerTestBase {
 
   // TESTS
 
+  @SuppressWarnings("deprecation")
   @Test(timeout=2000)
   public void testLoadConfigurationOnInitialize() throws IOException {
     conf.setBoolean(FairSchedulerConfiguration.ASSIGN_MULTIPLE, true);
@@ -4410,6 +4411,7 @@ public class TestFairScheduler extends FairSchedulerTestBase {
     verifyQueueNumRunnable("queue1.sub3", 0, 0);
   }
 
+  @SuppressWarnings("deprecation")
   @Test
   public void testSchedulingOnRemovedNode() throws Exception {
     // Disable continuous scheduling, will invoke continuous scheduling manually