YARN-10084. Allow inheritance of max app lifetime / default app lifet…

…ime. Contributed by Eric Payne.

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

@@ -47,6 +47,7 @@
 import org.apache.hadoop.yarn.api.records.ResourceInformation;
 import org.apache.hadoop.yarn.conf.YarnConfiguration;
 import org.apache.hadoop.yarn.exceptions.YarnException;
+import org.apache.hadoop.yarn.exceptions.YarnRuntimeException;
 import org.apache.hadoop.yarn.factories.RecordFactory;
 import org.apache.hadoop.yarn.factory.providers.RecordFactoryProvider;
 import org.apache.hadoop.yarn.security.AccessRequest;
@@ -116,6 +117,15 @@ public abstract class AbstractCSQueue implements CSQueue {
 
   QueueResourceQuotas queueResourceQuotas;
 
+  // -1 indicates lifetime is disabled
+  private volatile long maxApplicationLifetime = -1;
+
+  private volatile long defaultApplicationLifetime = -1;
+
+  // Indicates if this queue's default lifetime was set by a config property,
+  // either at this level or anywhere in the queue's hierarchy.
+  private volatile boolean defaultAppLifetimeWasSpecifiedInConfig = false;
+
   protected enum CapacityConfigType {
     NONE, PERCENTAGE, ABSOLUTE_RESOURCE
   };
@@ -425,6 +435,19 @@ protected void setupQueueConfigs(Resource clusterResource,
           configuration.getMultiNodesSortingAlgorithmPolicy(getQueuePath()));
 
       this.userWeights = getUserWeightsFromHierarchy(configuration);
+
+      maxApplicationLifetime = getInheritedMaxAppLifetime(this, configuration);
+      defaultApplicationLifetime =
+          getInheritedDefaultAppLifetime(this, configuration,
+              maxApplicationLifetime);
+      if (maxApplicationLifetime > 0 &&
+          defaultApplicationLifetime > maxApplicationLifetime) {
+        throw new YarnRuntimeException(
+            "Default lifetime " + defaultApplicationLifetime
+                + " can't exceed maximum lifetime " + maxApplicationLifetime);
+      }
+      defaultApplicationLifetime = defaultApplicationLifetime > 0
+          ? defaultApplicationLifetime : maxApplicationLifetime;
     } finally {
       writeLock.unlock();
     }
@@ -860,6 +883,68 @@ private boolean isQueueHierarchyPreemptionDisabled(CSQueue q,
                                         parentQ.getPreemptionDisabled());
   }
 
+  private long getInheritedMaxAppLifetime(CSQueue q,
+      CapacitySchedulerConfiguration conf) {
+    CSQueue parentQ = q.getParent();
+    long maxAppLifetime = conf.getMaximumLifetimePerQueue(q.getQueuePath());
+
+    // If q is the root queue, then get max app lifetime from conf.
+    if (parentQ == null) {
+      return maxAppLifetime;
+    }
+
+    // If this is not the root queue, get this queue's max app lifetime
+    // from the conf. The parent's max app lifetime will be used if it's
+    // not set for this queue.
+    // A value of 0 will override the parent's value and means no max lifetime.
+    // A negative value means that the parent's max should be used.
+    long parentsMaxAppLifetime = getParent().getMaximumApplicationLifetime();
+    return (maxAppLifetime >= 0) ? maxAppLifetime : parentsMaxAppLifetime;
+  }
+
+  private long getInheritedDefaultAppLifetime(CSQueue q,
+      CapacitySchedulerConfiguration conf, long myMaxAppLifetime) {
+    CSQueue parentQ = q.getParent();
+    long defaultAppLifetime = conf.getDefaultLifetimePerQueue(getQueuePath());
+    defaultAppLifetimeWasSpecifiedInConfig =
+        (defaultAppLifetime >= 0
+        || (parentQ != null &&
+            parentQ.getDefaultAppLifetimeWasSpecifiedInConfig()));
+
+    // If q is the root queue, then get default app lifetime from conf.
+    if (parentQ == null) {
+      return defaultAppLifetime;
+    }
+
+    // If this is not the root queue, get the parent's default app lifetime. The
+    // parent's default app lifetime will be used if not set for this queue.
+    long parentsDefaultAppLifetime =
+        getParent().getDefaultApplicationLifetime();
+
+    // Negative value indicates default lifetime was not set at this level.
+    // If default lifetime was not set at this level, calculate it based on
+    // parent's default lifetime or current queue's max lifetime.
+    if (defaultAppLifetime < 0) {
+      // If default lifetime was not set at this level but was set somewhere in
+      // the parent's hierarchy, set default lifetime to parent queue's default
+      // only if parent queue's lifetime is less than current queueu's max
+      // lifetime. Otherwise, use current queue's max lifetime value for its
+      // default lifetime.
+      if (defaultAppLifetimeWasSpecifiedInConfig) {
+        if (parentsDefaultAppLifetime <= myMaxAppLifetime) {
+          defaultAppLifetime = parentsDefaultAppLifetime;
+        } else {
+          defaultAppLifetime = myMaxAppLifetime;
+        }
+      } else {
+        // Default app lifetime value was not set anywhere in this queue's
+        // hierarchy. Use current queue's max lifetime as its default.
+        defaultAppLifetime = myMaxAppLifetime;
+      }
+    } // else if >= 0, default lifetime was set at this level. Just use it.
+    return defaultAppLifetime;
+  }
+
   /**
    * The specified queue is intra-queue preemptable if
    * 1) system-wide intra-queue preemption is turned on
@@ -1317,4 +1402,16 @@ public String getMultiNodeSortingPolicyName() {
   public void setMultiNodeSortingPolicyName(String policyName) {
     this.multiNodeSortingPolicyName = policyName;
   }
+
+  public long getMaximumApplicationLifetime() {
+    return maxApplicationLifetime;
+  }
+
+  public long getDefaultApplicationLifetime() {
+    return defaultApplicationLifetime;
+  }
+
+  public boolean getDefaultAppLifetimeWasSpecifiedInConfig() {
+    return defaultAppLifetimeWasSpecifiedInConfig;
+  }
 }

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

@@ -449,4 +449,26 @@ public void validateSubmitApplication(ApplicationId applicationId,
    * @return policy name
    */
   String getMultiNodeSortingPolicyName();
+
+  /**
+   * Get the maximum lifetime in seconds of an application which is submitted to
+   * this queue. Apps can set their own lifetime timeout up to this value.
+   * @return max lifetime in seconds
+   */
+  long getMaximumApplicationLifetime();
+
+  /**
+   * Get the default lifetime in seconds of an application which is submitted to
+   * this queue. If an app doesn't specify its own timeout when submitted, this
+   * value will be used.
+   * @return default app lifetime
+   */
+  long getDefaultApplicationLifetime();
+
+  /**
+   * Get the indicator of whether or not the default application lifetime was
+   * set by a config property or was calculated by the capacity scheduler.
+   * @return indicator whether set or calculated
+   */
+  boolean getDefaultAppLifetimeWasSpecifiedInConfig();
 }

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

@@ -45,7 +45,6 @@
 import org.apache.hadoop.yarn.api.records.QueueState;
 import org.apache.hadoop.yarn.api.records.QueueUserACLInfo;
 import org.apache.hadoop.yarn.api.records.Resource;
-import org.apache.hadoop.yarn.exceptions.YarnRuntimeException;
 import org.apache.hadoop.yarn.factories.RecordFactory;
 import org.apache.hadoop.yarn.factory.providers.RecordFactoryProvider;
 import org.apache.hadoop.yarn.nodelabels.CommonNodeLabelsManager;
@@ -130,10 +129,6 @@ public class LeafQueue extends AbstractCSQueue {
   List<AppPriorityACLGroup> priorityAcls =
       new ArrayList<AppPriorityACLGroup>();
 
-  // -1 indicates lifetime is disabled
-  private volatile long maxApplicationLifetime = -1;
-  private volatile long defaultApplicationLifetime = -1;
-
   @SuppressWarnings({ "unchecked", "rawtypes" })
   public LeafQueue(CapacitySchedulerContext cs,
       String queueName, CSQueue parent, CSQueue old) throws IOException {
@@ -256,19 +251,6 @@ protected void setupQueueConfigs(Resource clusterResource,
       defaultAppPriorityPerQueue = Priority.newInstance(
           conf.getDefaultApplicationPriorityConfPerQueue(getQueuePath()));
 
-      maxApplicationLifetime =
-          conf.getMaximumLifetimePerQueue((getQueuePath()));
-      defaultApplicationLifetime =
-          conf.getDefaultLifetimePerQueue((getQueuePath()));
-      if (maxApplicationLifetime > 0 &&
-          defaultApplicationLifetime > maxApplicationLifetime) {
-        throw new YarnRuntimeException(
-            "Default lifetime" + defaultApplicationLifetime
-                + " can't exceed maximum lifetime " + maxApplicationLifetime);
-      }
-      defaultApplicationLifetime = defaultApplicationLifetime > 0
-          ? defaultApplicationLifetime : maxApplicationLifetime;
-
       // Validate leaf queue's user's weights.
       int queueUL = Math.min(100, conf.getUserLimit(getQueuePath()));
       for (Entry<String, Float> e : getUserWeights().entrySet()) {
@@ -329,9 +311,9 @@ protected void setupQueueConfigs(Resource clusterResource,
               + reservationsContinueLooking + "\n" + "preemptionDisabled = "
               + getPreemptionDisabled() + "\n" + "defaultAppPriorityPerQueue = "
               + defaultAppPriorityPerQueue + "\npriority = " + priority
-              + "\nmaxLifetime = " + maxApplicationLifetime + " seconds"
-              + "\ndefaultLifetime = "
-              + defaultApplicationLifetime + " seconds");
+              + "\nmaxLifetime = " + getMaximumApplicationLifetime()
+              + " seconds" + "\ndefaultLifetime = "
+              + getDefaultApplicationLifetime() + " seconds");
     } finally {
       writeLock.unlock();
     }
@@ -2209,14 +2191,6 @@ static class CachedUserLimit {
     }
   }
 
-  public long getMaximumApplicationLifetime() {
-    return maxApplicationLifetime;
-  }
-
-  public long getDefaultApplicationLifetime() {
-    return defaultApplicationLifetime;
-  }
-
   private void updateQueuePreemptionMetrics(RMContainer rmc) {
     final long usedMillis = rmc.getFinishTime() - rmc.getCreationTime();
     final long usedSeconds = usedMillis / DateUtils.MILLIS_PER_SECOND;