YARN-6546. SLS is slow while loading 10k queues. (Yufei Gu via Haibo Chen)

hadoop-tools/hadoop-sls/src/main/java/org/apache/hadoop/yarn/sls/SLSRunner.java

@@ -731,6 +731,7 @@ public class SLSRunner extends Configured implements Tool {
     }
 
     queueAppNumMap.put(queueName, appNum);
+    wrapper.getSchedulerMetrics().trackQueue(queueName);
   }
 
   private void runNewAM(String jobType, String user,

hadoop-tools/hadoop-sls/src/main/java/org/apache/hadoop/yarn/sls/scheduler/CapacitySchedulerMetrics.java

@@ -28,9 +28,4 @@ public class CapacitySchedulerMetrics extends SchedulerMetrics {
   public CapacitySchedulerMetrics() {
     super();
   }
-
-  @Override
-  public void trackQueue(String queueName) {
-    trackedQueues.add(queueName);
-  }
 }

hadoop-tools/hadoop-sls/src/main/java/org/apache/hadoop/yarn/sls/scheduler/FairSchedulerMetrics.java

@@ -167,8 +167,9 @@ public class FairSchedulerMetrics extends SchedulerMetrics {
   }
 
   @Override
-  public void trackQueue(String queueName) {
+  protected void registerQueueMetrics(String queueName) {
-    trackedQueues.add(queueName);
+    super.registerQueueMetrics(queueName);
+
     FairScheduler fair = (FairScheduler) scheduler;
     final FSQueue queue = fair.getQueueManager().getQueue(queueName);
     registerQueueMetrics(queue, Metric.DEMAND);
@@ -209,16 +210,4 @@ public class FairSchedulerMetrics extends SchedulerMetrics {
       }
     );
   }
-
-  @Override
-  public void untrackQueue(String queueName) {
-    trackedQueues.remove(queueName);
-
-    for (Metric metric: Metric.values()) {
-      metrics.remove("variable.queue." + queueName + "." +
-          metric.value + ".memory");
-      metrics.remove("variable.queue." + queueName + "." +
-          metric.value + ".vcores");
-    }
-  }
 }

hadoop-tools/hadoop-sls/src/main/java/org/apache/hadoop/yarn/sls/scheduler/FifoSchedulerMetrics.java

@@ -35,8 +35,9 @@ public class FifoSchedulerMetrics extends SchedulerMetrics {
   }
 
   @Override
-  public void trackQueue(String queueName) {
+  protected void registerQueueMetrics(String queueName) {
-    trackedQueues.add(queueName);
+    super.registerQueueMetrics(queueName);
+
     FifoScheduler fifo = (FifoScheduler) scheduler;
     // for FifoScheduler, only DEFAULT_QUEUE
     // here the three parameters doesn't affect results

hadoop-tools/hadoop-sls/src/main/java/org/apache/hadoop/yarn/sls/scheduler/SLSCapacityScheduler.java

@@ -323,25 +323,6 @@ public class SLSCapacityScheduler extends CapacityScheduler implements
         queueName);
   }
 
-  private void initQueueMetrics(CSQueue queue) {
-    if (queue instanceof LeafQueue) {
-      schedulerMetrics.initQueueMetric(queue.getQueueName());
-      return;
-    }
-
-    for (CSQueue child : queue.getChildQueues()) {
-      initQueueMetrics(child);
-    }
-  }
-  @Override
-  public void serviceInit(Configuration configuration) throws Exception {
-    super.serviceInit(configuration);
-
-    if (metricsON) {
-      initQueueMetrics(getRootQueue());
-    }
-  }
-
   @Override
   public void serviceStop() throws Exception {
     try {

hadoop-tools/hadoop-sls/src/main/java/org/apache/hadoop/yarn/sls/scheduler/SLSFairScheduler.java

@@ -306,25 +306,6 @@ public class SLSFairScheduler extends FairScheduler
         queueName);
   }
 
-  private void initQueueMetrics(FSQueue queue) {
-    if (queue instanceof FSLeafQueue) {
-      schedulerMetrics.initQueueMetric(queue.getQueueName());
-      return;
-    }
-
-    for (FSQueue child : queue.getChildQueues()) {
-      initQueueMetrics(child);
-    }
-  }
-
-  @Override
-  public void serviceInit(Configuration conf) throws Exception {
-    super.serviceInit(conf);
-    if (metricsON) {
-      initQueueMetrics(getQueueManager().getRootQueue());
-    }
-  }
-
   @Override
   public void serviceStop() throws Exception {
     try {

hadoop-tools/hadoop-sls/src/main/java/org/apache/hadoop/yarn/sls/scheduler/SchedulerMetrics.java

@@ -234,14 +234,35 @@ public abstract class SchedulerMetrics {
     }
   }
 
-  public abstract void trackQueue(String queueName);
+  /**
-  
+   * Track a queue by registering its metrics.
-  public void untrackQueue(String queueName) {
+   *
-    for (String m : queueTrackedMetrics) {
+   * @param queue queue name
-      metrics.remove("variable.queue." + queueName + "." + m);
+   */
+  public void trackQueue(String queue) {
+    queueLock.lock();
+    try {
+      if (!isTracked(queue)) {
+        trackedQueues.add(queue);
+        registerQueueMetrics(queue);
+      }
+    } finally {
+      queueLock.unlock();
     }
   }
-  
+
+  protected void registerQueueMetrics(String queueName) {
+    SortedMap<String, Counter> counterMap = metrics.getCounters();
+
+    for (QueueMetric queueMetric : QueueMetric.values()) {
+      String metricName = getQueueMetricName(queueName, queueMetric);
+      if (!counterMap.containsKey(metricName)) {
+        metrics.counter(metricName);
+        queueTrackedMetrics.add(metricName);
+      }
+    }
+  }
+
   public boolean isTracked(String queueName) {
     return trackedQueues.contains(queueName);
   }
@@ -547,40 +568,13 @@ public abstract class SchedulerMetrics {
     return "counter.queue." + queue + "." + metric.value;
   }
 
-  private void traceQueueIfNotTraced(String queue) {
-    queueLock.lock();
-    try {
-      if (!isTracked(queue)) {
-        trackQueue(queue);
-      }
-    } finally {
-      queueLock.unlock();
-    }
-  }
-
-  void initQueueMetric(String queueName){
-    SortedMap<String, Counter> counterMap = metrics.getCounters();
-
-    for (QueueMetric queueMetric : QueueMetric.values()) {
-      String metricName = getQueueMetricName(queueName, queueMetric);
-      if (!counterMap.containsKey(metricName)) {
-        metrics.counter(metricName);
-        counterMap = metrics.getCounters();
-      }
-    }
-
-    traceQueueIfNotTraced(queueName);
-  }
-
   void updateQueueMetrics(Resource pendingResource, Resource allocatedResource,
       String queueName) {
+    trackQueue(queueName);
+
     SortedMap<String, Counter> counterMap = metrics.getCounters();
     for(QueueMetric metric : QueueMetric.values()) {
       String metricName = getQueueMetricName(queueName, metric);
-      if (!counterMap.containsKey(metricName)) {
-        metrics.counter(metricName);
-        counterMap = metrics.getCounters();
-      }
 
       if (metric == QueueMetric.PENDING_MEMORY) {
         counterMap.get(metricName).inc(pendingResource.getMemorySize());
@@ -592,8 +586,6 @@ public abstract class SchedulerMetrics {
         counterMap.get(metricName).inc(allocatedResource.getVirtualCores());
       }
     }
-
-    traceQueueIfNotTraced(queueName);
   }
 
   void updateQueueMetricsByRelease(Resource releaseResource, String queue) {