[SBN Read] Slow clients when Observer reads are enabled but there are no Observers on the cluster. Conntributed by Chen Liang

hadoop-hdfs-project/hadoop-hdfs-client/src/main/java/org/apache/hadoop/hdfs/server/namenode/ha/ObserverReadProxyProvider.java

@@ -81,6 +81,12 @@ public class ObserverReadProxyProvider<T>
   /** Client-side context for syncing with the NameNode server side. */
   private final AlignmentContext alignmentContext;
 
+  /** Configuration key for {@link #observerProbeRetryPeriodMs}. */
+  static final String OBSERVER_PROBE_RETRY_PERIOD_KEY =
+      "dfs.client.failover.observer.probe.retry.period";
+  /** Observer probe retry period default to 10 min. */
+  static final long OBSERVER_PROBE_RETRY_PERIOD_DEFAULT = 60 * 10 * 1000;
+
   /** The inner proxy provider used for active/standby failover. */
   private final AbstractNNFailoverProxyProvider<T> failoverProxy;
   /** List of all NameNode proxies. */
@@ -139,6 +145,21 @@ public class ObserverReadProxyProvider<T>
   /** The last proxy that has been used. Only used for testing. */
   private volatile ProxyInfo<T> lastProxy = null;
 
+  /**
+   * In case there is no Observer node, for every read call, client will try
+   * to loop through all Standby nodes and fail eventually. Since there is no
+   * guarantee on when Observer node will be enabled. This can be very
+   * inefficient.
+   * The following value specify the period on how often to retry all Standby.
+   */
+  private long observerProbeRetryPeriodMs;
+
+  /**
+   * The previous time where zero observer were found. If there was observer,
+   * or it is initialization, this is set to 0.
+   */
+  private long lastObserverProbeTime;
+
   /**
    * By default ObserverReadProxyProvider uses
    * {@link ConfiguredFailoverProxyProvider} for failover.
@@ -157,6 +178,7 @@ public class ObserverReadProxyProvider<T>
     this.failoverProxy = failoverProxy;
     this.alignmentContext = new ClientGSIContext();
     factory.setAlignmentContext(alignmentContext);
+    this.lastObserverProbeTime = 0;
 
     // Don't bother configuring the number of retries and such on the retry
     // policy since it is mainly only used for determining whether or not an
@@ -187,6 +209,9 @@ public class ObserverReadProxyProvider<T>
         // The host of the URI is the nameservice ID
         AUTO_MSYNC_PERIOD_KEY_PREFIX + "." + uri.getHost(),
         AUTO_MSYNC_PERIOD_DEFAULT, TimeUnit.MILLISECONDS);
+    observerProbeRetryPeriodMs = conf.getTimeDuration(
+        OBSERVER_PROBE_RETRY_PERIOD_KEY,
+        OBSERVER_PROBE_RETRY_PERIOD_DEFAULT, TimeUnit.MILLISECONDS);
 
     // TODO : make this configurable or remove this variable
     if (wrappedProxy instanceof ClientProtocol) {
@@ -322,6 +347,27 @@ public class ObserverReadProxyProvider<T>
     lastMsyncTimeMs = Time.monotonicNow();
   }
 
+  /**
+   * Check if client need to find an Observer proxy.
+   * If current proxy is Active then we should stick to it and postpone probing
+   * for Observers for a period of time. When this time expires the client will
+   * try to find an Observer again.
+   * *
+   * @return true if we did not reach the threshold
+   * to start looking for Observer, or false otherwise.
+   */
+  private boolean shouldFindObserver() {
+    // lastObserverProbeTime > 0 means we tried, but did not find any
+    // Observers yet
+    // If lastObserverProbeTime <= 0, previous check found observer, so
+    // we should not skip observer read.
+    if (lastObserverProbeTime > 0) {
+      return Time.monotonicNow() - lastObserverProbeTime
+          >= observerProbeRetryPeriodMs;
+    }
+    return true;
+  }
+
   /**
    * This will call {@link ClientProtocol#msync()} on the active NameNode
    * (via the {@link #failoverProxy}) to update the state of this client, only
@@ -369,7 +415,7 @@ public class ObserverReadProxyProvider<T>
       lastProxy = null;
       Object retVal;
 
-      if (observerReadEnabled && isRead(method)) {
+      if (observerReadEnabled && shouldFindObserver() && isRead(method)) {
         if (!msynced) {
           // An msync() must first be performed to ensure that this client is
           // up-to-date with the active's state. This will only be done once.
@@ -446,11 +492,13 @@ public class ObserverReadProxyProvider<T>
                   + "{} standby, {} active, and {} unreachable. Falling back "
                   + "to active.", failedObserverCount, method.getName(),
               standbyCount, activeCount, unreachableCount);
+          lastObserverProbeTime = 0;
         } else {
           if (LOG.isDebugEnabled()) {
             LOG.debug("Read falling back to active without observer read "
                 + "fail, is there no observer node running?");
           }
+          lastObserverProbeTime = Time.monotonicNow();
         }
       }
 

hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/server/namenode/ha/TestObserverNode.java

@@ -19,6 +19,7 @@ package org.apache.hadoop.hdfs.server.namenode.ha;
 
 import static org.apache.hadoop.hdfs.DFSConfigKeys.DFS_NAMENODE_STATE_CONTEXT_ENABLED_KEY;
 import static org.apache.hadoop.hdfs.server.namenode.NameNodeAdapter.getServiceState;
+import static org.apache.hadoop.hdfs.server.namenode.ha.ObserverReadProxyProvider.*;
 import static org.junit.Assert.assertEquals;
 import static org.junit.Assert.assertTrue;
 import static org.junit.Assert.fail;
@@ -33,7 +34,9 @@ import java.net.URI;
 import java.util.ArrayList;
 import java.util.List;
 
+import java.util.concurrent.TimeUnit;
 import org.apache.hadoop.conf.Configuration;
+import org.apache.hadoop.fs.FileSystem;
 import org.apache.hadoop.fs.FileUtil;
 import org.apache.hadoop.fs.Path;
 import org.apache.hadoop.fs.permission.FsPermission;
@@ -53,6 +56,7 @@ import org.apache.hadoop.hdfs.server.namenode.NameNodeAdapter;
 import org.apache.hadoop.hdfs.server.namenode.TestFsck;
 import org.apache.hadoop.hdfs.tools.GetGroups;
 import org.apache.hadoop.ipc.ObserverRetryOnActiveException;
+import org.apache.hadoop.util.Time;
 import org.junit.After;
 import org.junit.AfterClass;
 import org.junit.Before;
@@ -80,6 +84,10 @@ public class TestObserverNode {
   public static void startUpCluster() throws Exception {
     conf = new Configuration();
     conf.setBoolean(DFS_NAMENODE_STATE_CONTEXT_ENABLED_KEY, true);
+    // Set observer probe retry period to 0. Required by the tests that restart
+    // Observer and immediately try to read from it.
+    conf.setTimeDuration(
+        OBSERVER_PROBE_RETRY_PERIOD_KEY, 0, TimeUnit.MILLISECONDS);
     qjmhaCluster = HATestUtil.setUpObserverCluster(conf, 1, 0, true);
     dfsCluster = qjmhaCluster.getDfsCluster();
   }
@@ -413,6 +421,54 @@ public class TestObserverNode {
     assertSentTo(0);
   }
 
+
+  /**
+   * Test that if client connects to Active it does not try to find Observer
+   * on next calls during some period of time.
+   */
+  @Test
+  public void testStickyActive() throws Exception {
+    Path testFile = new Path(testPath, "testStickyActive");
+    Configuration newConf = new Configuration(conf);
+    // Observer probe retry period set to 5 sec
+    newConf.setLong(OBSERVER_PROBE_RETRY_PERIOD_KEY, 5000);
+    // Disable cache, so that a new client actually gets created with new conf.
+    newConf.setBoolean("fs.hdfs.impl.disable.cache", true);
+    DistributedFileSystem newFs = (DistributedFileSystem) FileSystem.get(newConf);
+    newFs.create(testFile, (short)1).close();
+    assertSentTo(0);
+    dfsCluster.rollEditLogAndTail(0);
+    // No Observers present, should still go to Active
+    dfsCluster.transitionToStandby(2);
+    assertEquals("NN[2] should be standby", HAServiceState.STANDBY,
+        getServiceState(dfsCluster.getNameNode(2)));
+    newFs.open(testFile).close();
+    assertSentTo(0);
+    // Restore Observer
+    int newObserver = 1;
+    dfsCluster.transitionToObserver(newObserver);
+    assertEquals("NN[" + newObserver + "] should be observer",
+        HAServiceState.OBSERVER,
+        getServiceState(dfsCluster.getNameNode(newObserver)));
+    long startTime = Time.monotonicNow();
+    try {
+      while(Time.monotonicNow() - startTime <= 5000) {
+        newFs.open(testFile).close();
+        // Client should still talk to Active
+        assertSentTo(0);
+        Thread.sleep(200);
+      }
+    } catch(AssertionError ae) {
+      if(Time.monotonicNow() - startTime <= 5000) {
+        throw ae;
+      }
+      assertSentTo(newObserver);
+    } finally {
+      dfsCluster.transitionToStandby(1);
+      dfsCluster.transitionToObserver(2);
+    }
+  }
+
   private void assertSentTo(int nnIdx) throws IOException {
     assertTrue("Request was not sent to the expected namenode " + nnIdx,
         HATestUtil.isSentToAnyOfNameNodes(dfs, dfsCluster, nnIdx));

hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/server/namenode/ha/TestObserverReadProxyProvider.java

@@ -24,6 +24,7 @@ import java.net.URI;
 import java.util.HashMap;
 import java.util.List;
 import java.util.Map;
+import java.util.concurrent.TimeUnit;
 import java.util.concurrent.atomic.AtomicBoolean;
 import org.apache.hadoop.conf.Configuration;
 import org.apache.hadoop.fs.permission.FsAction;
@@ -43,6 +44,7 @@ import org.mockito.stubbing.Answer;
 
 import static org.apache.hadoop.ha.HAServiceProtocol.HAServiceState;
 
+import static org.apache.hadoop.hdfs.server.namenode.ha.ObserverReadProxyProvider.*;
 import static org.junit.Assert.assertArrayEquals;
 import static org.junit.Assert.assertEquals;
 import static org.junit.Assert.fail;
@@ -72,6 +74,10 @@ public class TestObserverReadProxyProvider {
     nnURI = URI.create("hdfs://" + ns);
     conf = new Configuration();
     conf.set(HdfsClientConfigKeys.DFS_NAMESERVICES, ns);
+    // Set observer probe retry period to 0. Required by the tests that
+    // transition observer back and forth
+    conf.setTimeDuration(
+        OBSERVER_PROBE_RETRY_PERIOD_KEY, 0, TimeUnit.MILLISECONDS);
   }
 
   private void setupProxyProvider(int namenodeCount) throws Exception {