HDFS-9500. Fix software version counts for DataNodes during rolling upgrade. Contributed by Erik Krogen.

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/blockmanagement/DatanodeManager.java

@@ -759,17 +759,25 @@ private void decrementVersionCount(String version) {
     }
   }
 
+  /**
+   * Will return true for all Datanodes which have a non-null software
+   * version and are considered alive (by {@link DatanodeDescriptor#isAlive()}),
+   * indicating the node has not yet been removed. Use {@code isAlive}
+   * rather than {@link DatanodeManager#isDatanodeDead(DatanodeDescriptor)}
+   * to ensure that the version is decremented even if the datanode
+   * hasn't issued a heartbeat recently.
+   *
+   * @param node The datanode in question
+   * @return True iff its version count should be decremented
+   */
   private boolean shouldCountVersion(DatanodeDescriptor node) {
-    return node.getSoftwareVersion() != null && node.isAlive() &&
+    return node.getSoftwareVersion() != null && node.isAlive();
-      !isDatanodeDead(node);
   }
 
   private void countSoftwareVersions() {
     synchronized(this) {
       datanodesSoftwareVersions.clear();
       for(DatanodeDescriptor dn: datanodeMap.values()) {
-        // Check isAlive too because right after removeDatanode(),
-        // isDatanodeDead() is still true 
         if (shouldCountVersion(dn)) {
           Integer num = datanodesSoftwareVersions.get(dn.getSoftwareVersion());
           num = num == null ? 1 : num+1;

hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/server/blockmanagement/TestDatanodeManager.java

@@ -85,6 +85,43 @@ private static InetSocketAddress entry(String host) {
     return HostFileManager.parseEntry("dummy", "dummy", host);
   }
 
+  /**
+   * This test checks that if a node is re-registered with a new software
+   * version after the heartbeat expiry interval but before the HeartbeatManager
+   * has a chance to detect this and remove it, the node's version will still
+   * be correctly decremented.
+   */
+  @Test
+  public void testNumVersionsCorrectAfterReregister()
+      throws IOException, InterruptedException {
+    //Create the DatanodeManager which will be tested
+    FSNamesystem fsn = Mockito.mock(FSNamesystem.class);
+    Mockito.when(fsn.hasWriteLock()).thenReturn(true);
+    Configuration conf = new Configuration();
+    conf.setLong(DFSConfigKeys.DFS_HEARTBEAT_INTERVAL_KEY, 0);
+    conf.setLong(DFSConfigKeys.DFS_NAMENODE_HEARTBEAT_RECHECK_INTERVAL_KEY, 10);
+    DatanodeManager dm = mockDatanodeManager(fsn, conf);
+
+    String storageID = "someStorageID1";
+    String ip = "someIP" + storageID;
+
+    // Register then reregister the same node but with a different version
+    for (int i = 0; i <= 1; i++) {
+      dm.registerDatanode(new DatanodeRegistration(
+          new DatanodeID(ip, "", storageID, 9000, 0, 0, 0),
+          null, null, "version" + i));
+      if (i == 0) {
+        Thread.sleep(25);
+      }
+    }
+
+    //Verify DatanodeManager has the correct count
+    Map<String, Integer> mapToCheck = dm.getDatanodesSoftwareVersions();
+    assertNull("should be no more version0 nodes", mapToCheck.get("version0"));
+    assertEquals("should be one version1 node",
+        mapToCheck.get("version1").intValue(), 1);
+  }
+
   /**
    * This test sends a random sequence of node registrations and node removals
    * to the DatanodeManager (of nodes with different IDs and versions), and