HDFS-15323. StandbyNode fails transition to active due to insufficient transaction tailing. Contributed by Konstantin V Shvachko.

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/qjournal/client/QuorumJournalManager.java

@@ -73,9 +73,9 @@ public class QuorumJournalManager implements JournalManager {
   static final Logger LOG = LoggerFactory.getLogger(QuorumJournalManager.class);
 
   // This config is not publicly exposed
-  static final String QJM_RPC_MAX_TXNS_KEY =
+  public static final String QJM_RPC_MAX_TXNS_KEY =
       "dfs.ha.tail-edits.qjm.rpc.max-txns";
-  static final int QJM_RPC_MAX_TXNS_DEFAULT = 5000;
+  public static final int QJM_RPC_MAX_TXNS_DEFAULT = 5000;
 
   // Maximum number of transactions to fetch at a time when using the
   // RPC edit fetch mechanism

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

@@ -298,13 +298,23 @@ public void catchupDuringFailover() throws IOException {
     SecurityUtil.doAsLoginUser(new PrivilegedExceptionAction<Void>() {
       @Override
       public Void run() throws Exception {
+        long editsTailed = 0;
+        // Fully tail the journal to the end
+        do {
+          long startTime = Time.monotonicNow();
           try {
-          // It is already under the full name system lock and the checkpointer
+            NameNode.getNameNodeMetrics().addEditLogTailInterval(
-          // thread is already stopped. No need to acqure any other lock.
+                startTime - lastLoadTimeMs);
-          doTailEdits();
+            // It is already under the name system lock and the checkpointer
+            // thread is already stopped. No need to acquire any other lock.
+            editsTailed = doTailEdits();
           } catch (InterruptedException e) {
             throw new IOException(e);
+          } finally {
+            NameNode.getNameNodeMetrics().addEditLogTailTime(
+                Time.monotonicNow() - startTime);
           }
+        } while(editsTailed > 0);
         return null;
       }
     });

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

@@ -42,6 +42,7 @@
 import org.apache.hadoop.hdfs.server.namenode.NameNode;
 import org.apache.hadoop.test.GenericTestUtils;
 import static org.apache.hadoop.hdfs.server.namenode.NameNodeAdapter.getFileInfo;
+import static org.apache.hadoop.hdfs.qjournal.client.QuorumJournalManager.QJM_RPC_MAX_TXNS_KEY;
 
 import org.junit.After;
 import org.junit.Before;
@@ -72,6 +73,8 @@ public void startUp() throws IOException {
     conf.setBoolean(DFSConfigKeys.DFS_HA_TAILEDITS_INPROGRESS_KEY, true);
     conf.setInt(DFSConfigKeys.DFS_QJOURNAL_SELECT_INPUT_STREAMS_TIMEOUT_KEY,
         500);
+    // Set very samll limit of transactions per a journal rpc call
+    conf.setInt(QJM_RPC_MAX_TXNS_KEY, 3);
     HAUtil.setAllowStandbyReads(conf, true);
     qjmhaCluster = new MiniQJMHACluster.Builder(conf).build();
     cluster = qjmhaCluster.getDfsCluster();
@@ -300,6 +303,31 @@ public void testNewStartInProgressTail() throws Exception {
     waitForFileInfo(nn1, "/test", "/test2", "/test3");
   }
 
+  /**
+   * Test that Standby Node tails multiple segments while catching up
+   * during the transition to Active.
+   */
+  @Test
+  public void testUndertailingWhileFailover() throws Exception {
+    cluster.transitionToActive(0);
+    cluster.waitActive(0);
+
+    String p = "/testFailoverWhileTailingWithoutCache/";
+    mkdirs(nn0, p + 0, p + 1, p + 2, p + 3, p + 4);
+    nn0.getRpcServer().rollEditLog(); // create segment 1
+
+    mkdirs(nn0, p + 5, p + 6, p + 7, p + 8, p + 9);
+    nn0.getRpcServer().rollEditLog(); // create segment 2
+
+    mkdirs(nn0, p + 10, p + 11, p + 12, p + 13, p + 14);
+    nn0.getRpcServer().rollEditLog(); // create segment 3
+
+    cluster.transitionToStandby(0);
+    cluster.transitionToActive(1);
+    cluster.waitActive(1);
+    waitForFileInfo(nn1, p + 0, p + 1, p + 14);
+  }
+
   @Test
   public void testNonUniformConfig() throws Exception {
     // Test case where some NNs (in this case the active NN) in the cluster