HADOOP-8968. Add a flag to completely disable the worker version check. Contributed by Alejandro Abdelnur

git-svn-id: https://svn.apache.org/repos/asf/hadoop/common/branches/branch-1@1402741 13f79535-47bb-0310-9956-ffa450edef68

CHANGES.txt

@@ -107,6 +107,9 @@ Release 1.2.0 - unreleased
     HADOOP-8567. Port conf servlet to dump running configuration to branch 1.x.
     (Jing Zhao via suresh)
 
+    HADOOP-8968. Add a flag to completely disable the worker version check.
+    (tucu via eli)
+
   OPTIMIZATIONS
 
     HDFS-2533. Backport: Remove needless synchronization on some FSDataSet

src/core/core-default.xml

@@ -577,6 +577,20 @@
   </description>
 </property>
 
+<property>
+  <name>hadoop.skip.worker.version.check</name>
+  <value>false</value>
+  <description>
+    By default datanodes refuse to connect to namenodes if their build
+    revision (svn revision) do not match, and tasktrackers refuse to
+    connect to jobtrackers if their build version (version, revision,
+    user, and source checksum) do not match. This option changes the
+    behavior of hadoop workers to skip doing a version check at all.
+    This option supersedes the 'hadoop.relaxed.worker.version.check'
+    option.
+  </description>
+</property>
+
 <property>
   <name>hadoop.jetty.logs.serve.aliases</name>
   <value>true</value>

src/core/org/apache/hadoop/fs/CommonConfigurationKeys.java

@@ -74,6 +74,11 @@ public class CommonConfigurationKeys {
       "hadoop.relaxed.worker.version.check";
   public static final boolean HADOOP_RELAXED_VERSION_CHECK_DEFAULT = false;
 
+  /** See src/core/core-default.xml */
+  public static final String HADOOP_SKIP_VERSION_CHECK_KEY =
+      "hadoop.skip.worker.version.check";
+  public static final boolean HADOOP_SKIP_VERSION_CHECK_DEFAULT = false;
+
   /** Enable/Disable aliases serving from jetty */
   public static final String HADOOP_JETTY_LOGS_SERVE_ALIASES =
     "hadoop.jetty.logs.serve.aliases";

src/hdfs/org/apache/hadoop/hdfs/server/datanode/DataNode.java

@@ -77,11 +77,9 @@ import org.apache.hadoop.hdfs.security.token.block.BlockTokenIdentifier;
 import org.apache.hadoop.hdfs.security.token.block.BlockTokenSecretManager;
 import org.apache.hadoop.hdfs.security.token.block.BlockTokenSecretManager.AccessMode;
 import org.apache.hadoop.hdfs.security.token.block.ExportedBlockKeys;
-import org.apache.hadoop.hdfs.server.common.GenerationStamp;
 import org.apache.hadoop.hdfs.server.common.HdfsConstants;
 import org.apache.hadoop.hdfs.server.common.HdfsConstants.StartupOption;
 import org.apache.hadoop.hdfs.server.common.IncorrectVersionException;
-import org.apache.hadoop.hdfs.server.common.Storage;
 import org.apache.hadoop.hdfs.server.datanode.FSDataset.VolumeInfo;
 import org.apache.hadoop.hdfs.server.datanode.SecureDataNodeStarter.SecureResources;
 import org.apache.hadoop.hdfs.server.datanode.metrics.DataNodeInstrumentation;
@@ -243,6 +241,11 @@ public class DataNode extends Configured
   private boolean connectToDnViaHostname;
   private boolean relaxedVersionCheck;
 
+  /**
+   * Whether the DN completely skips version check with the NN.
+   */
+  private boolean noVersionCheck;
+
   /**
    * Testing hook that allows tests to delay the sending of blockReceived RPCs
    * to the namenode. This can help find bugs in append.
@@ -364,6 +367,9 @@ public class DataNode extends Configured
     this.relaxedVersionCheck = conf.getBoolean(
         CommonConfigurationKeys.HADOOP_RELAXED_VERSION_CHECK_KEY,
         CommonConfigurationKeys.HADOOP_RELAXED_VERSION_CHECK_DEFAULT);
+    noVersionCheck = conf.getBoolean(
+        CommonConfigurationKeys.HADOOP_SKIP_VERSION_CHECK_KEY,
+        CommonConfigurationKeys.HADOOP_SKIP_VERSION_CHECK_DEFAULT);
 
     InetSocketAddress socAddr = DataNode.getStreamingAddr(conf);
     int tmpPort = socAddr.getPort();
@@ -600,22 +606,31 @@ public class DataNode extends Configured
       nsInfo.getVersion().equals(VersionInfo.getVersion());
     boolean revisionMatch =
       nsInfo.getRevision().equals(VersionInfo.getRevision());
-
     if (revisionMatch && !versionMatch) {
       throw new AssertionError("Invalid build. The revisions match" +
           " but the NN version is " + nsInfo.getVersion() +
           " and the DN version is " + VersionInfo.getVersion());
     }
-    if (relaxedVersionCheck) {
-      if (versionMatch && !revisionMatch) {
-        LOG.info("Permitting datanode revision " + VersionInfo.getRevision() +
-            " to connect to namenode revision " + nsInfo.getRevision() +
-            " because " + CommonConfigurationKeys.HADOOP_RELAXED_VERSION_CHECK_KEY +
-            " is enabled");
-      }
-      return versionMatch;
+    if (noVersionCheck) {
+      LOG.info("Permitting datanode version '" + VersionInfo.getVersion() +
+          "' and revision '" + VersionInfo.getRevision() +
+          "' to connect to namenode version '" + nsInfo.getVersion() +
+          "' and revision '" + nsInfo.getRevision() + "' because " +
+          CommonConfigurationKeys.HADOOP_SKIP_VERSION_CHECK_KEY +
+          " is enabled");
+      return true;
     } else {
-      return revisionMatch;
+      if (relaxedVersionCheck) {
+        if (versionMatch && !revisionMatch) {
+          LOG.info("Permitting datanode revision " + VersionInfo.getRevision() +
+              " to connect to namenode revision " + nsInfo.getRevision() +
+              " because " + CommonConfigurationKeys.HADOOP_RELAXED_VERSION_CHECK_KEY +
+              " is enabled");
+        }
+        return versionMatch;
+      } else {
+        return revisionMatch;
+      }
     }
   }
 
@@ -633,12 +648,15 @@ public class DataNode extends Configured
       }
     }
     if (!isPermittedVersion(nsInfo)) {
-      String errorMsg = "Incompatible versions: namenode version " +
-        nsInfo.getVersion() + " revision " + nsInfo.getRevision() +
-        " datanode version " + VersionInfo.getVersion() + " revision " +
-        VersionInfo.getRevision() + " and " +
-        CommonConfigurationKeys.HADOOP_RELAXED_VERSION_CHECK_KEY +
-        " is " + (relaxedVersionCheck ? "enabled" : "not enabled");
+      String errorMsg = "Shutting down. Incompatible version or revision." +
+          "DataNode version '" + VersionInfo.getVersion() +
+          "' and revision '" + VersionInfo.getRevision() +
+          "' and NameNode version '" + nsInfo.getVersion() +
+          "' and revision '" + nsInfo.getRevision() +
+          " and " + CommonConfigurationKeys.HADOOP_RELAXED_VERSION_CHECK_KEY +
+          " is " + (relaxedVersionCheck ? "enabled" : "not enabled") +
+          " and " + CommonConfigurationKeys.HADOOP_SKIP_VERSION_CHECK_KEY +
+          " is " + (noVersionCheck ? "enabled" : "not enabled");
       LOG.fatal(errorMsg);
       notifyNamenode(DatanodeProtocol.NOTIFY, errorMsg);  
       throw new IOException( errorMsg );

src/mapred/org/apache/hadoop/mapred/TaskTracker.java

@@ -418,6 +418,11 @@ public class TaskTracker implements MRConstants, TaskUmbilicalProtocol,
    */
   private boolean relaxedVersionCheck;
 
+  /**
+   * Whether the TT completely skips version check with the JT.
+   */
+  private boolean skipVersionCheck;
+
   /*
    * A list of commitTaskActions for whom commit response has been received 
    */
@@ -1459,6 +1464,9 @@ public class TaskTracker implements MRConstants, TaskUmbilicalProtocol,
     relaxedVersionCheck = conf.getBoolean(
         CommonConfigurationKeys.HADOOP_RELAXED_VERSION_CHECK_KEY,
         CommonConfigurationKeys.HADOOP_RELAXED_VERSION_CHECK_DEFAULT);
+    skipVersionCheck = conf.getBoolean(
+        CommonConfigurationKeys.HADOOP_SKIP_VERSION_CHECK_KEY,
+        CommonConfigurationKeys.HADOOP_SKIP_VERSION_CHECK_DEFAULT);
     FILE_CACHE_SIZE = conf.getInt("mapred.tasktracker.file.cache.size", 2000);
     maxMapSlots = conf.getInt(
                   "mapred.tasktracker.map.tasks.maximum", 2);
@@ -1623,7 +1631,7 @@ public class TaskTracker implements MRConstants, TaskUmbilicalProtocol,
   private long getHeartbeatInterval(int numFinishedTasks) {
     return (heartbeatInterval / (numFinishedTasks * oobHeartbeatDamper + 1));
   }
-  
+
   /**
    * @return true if this tasktracker is permitted to connect to
    *    the given jobtracker version
@@ -1637,16 +1645,26 @@ public class TaskTracker implements MRConstants, TaskUmbilicalProtocol,
           " but the JT version is " + jtVersion +
           " and the TT version is " + VersionInfo.getVersion());
     }
-    if (relaxedVersionCheck) {
-      if (!buildVersionMatch && versionMatch) {
-        LOG.info("Permitting tasktracker revision " + VersionInfo.getRevision() +
-            " to connect to jobtracker " + jtBuildVersion + " because " +
-            CommonConfigurationKeys.HADOOP_RELAXED_VERSION_CHECK_KEY +
-            " is enabled");
-      }
-      return versionMatch;
+    if (skipVersionCheck) {
+      LOG.info("Permitting tasktracker version '" + VersionInfo.getVersion() +
+          "' and build '" + VersionInfo.getBuildVersion() +
+          "' to connect to jobtracker version '" + jtVersion +
+          "' and build '" + jtBuildVersion + "' because " +
+          CommonConfigurationKeys.HADOOP_SKIP_VERSION_CHECK_KEY +
+          " is enabled");
+      return true;
     } else {
-      return buildVersionMatch;
+      if (relaxedVersionCheck) {
+        if (!buildVersionMatch && versionMatch) {
+          LOG.info("Permitting tasktracker build " + VersionInfo.getBuildVersion() +
+              " to connect to jobtracker build " + jtBuildVersion + " because " +
+              CommonConfigurationKeys.HADOOP_RELAXED_VERSION_CHECK_KEY +
+              " is enabled");
+        }
+        return versionMatch;
+      } else {
+        return buildVersionMatch;
+      }
     }
   }
 
@@ -1689,11 +1707,15 @@ public class TaskTracker implements MRConstants, TaskUmbilicalProtocol,
           String jtBuildVersion = jobClient.getBuildVersion();
           String jtVersion = jobClient.getVIVersion();
           if (!isPermittedVersion(jtBuildVersion, jtVersion)) {
-            String msg = "Shutting down. Incompatible buildVersion." +
-              "\nJobTracker's: " + jtBuildVersion + 
-              "\nTaskTracker's: "+ VersionInfo.getBuildVersion() +
+            String msg = "Shutting down. Incompatible version or build version." +
+              "TaskTracker version '" + VersionInfo.getVersion() +
+              "' and build '" + VersionInfo.getBuildVersion() +
+              "' and JobTracker version '" + jtVersion +
+              "' and build '" + jtBuildVersion +
               " and " + CommonConfigurationKeys.HADOOP_RELAXED_VERSION_CHECK_KEY +
-              " is " + (relaxedVersionCheck ? "enabled" : "not enabled");
+              " is " + (relaxedVersionCheck ? "enabled" : "not enabled") +
+              " and " + CommonConfigurationKeys.HADOOP_SKIP_VERSION_CHECK_KEY +
+              " is " + (skipVersionCheck ? "enabled" : "not enabled");
             LOG.fatal(msg);
             try {
               jobClient.reportTaskTrackerError(taskTrackerName, null, msg);

src/test/org/apache/hadoop/hdfs/server/datanode/TestDataNodeVersionCheck.java

@@ -116,4 +116,42 @@ public class TestDataNodeVersionCheck {
       }
     }
   }
+
+  /**
+   * Test no DN version checking
+   */
+  @Test
+  public void testNoVersionCheck() throws IOException {
+    MiniDFSCluster cluster = null;
+    try {
+      Configuration conf = new Configuration();
+      conf.setBoolean(
+          CommonConfigurationKeys.HADOOP_SKIP_VERSION_CHECK_KEY, true);
+      cluster = new MiniDFSCluster(conf, 1, true, null);
+
+      DataNode dn = cluster.getDataNodes().get(0);
+
+      final NamespaceInfo currInfo = new NamespaceInfo(0, 0, 0);
+      assertTrue(dn.isPermittedVersion(currInfo));
+
+      // Different revisions are permitted
+      NamespaceInfo infoDiffRev = new NamespaceInfo(0, 0, 0) {
+        @Override public String getRevision() { return "bogus"; }
+      };
+      assertTrue("Different revisions should be permitted",
+          dn.isPermittedVersion(infoDiffRev));
+
+      // Different versions are permitted
+      NamespaceInfo infoDiffVersion = new NamespaceInfo(0, 0, 0) {
+        @Override public String getVersion() { return "bogus"; }
+        @Override public String getRevision() { return "bogus"; }
+      };
+      assertTrue("Different versions should be permitted",
+          dn.isPermittedVersion(infoDiffVersion));
+    } finally {
+      if (cluster != null) {
+        cluster.shutdown();
+      }
+    }
+  }
 }

src/test/org/apache/hadoop/mapred/TestTaskTrackerVersionCheck.java

@@ -88,4 +88,29 @@ public class TestTaskTrackerVersionCheck {
       }
     }
   }
+
+  /**
+   * Test no TT version checking
+   */
+  @Test
+  public void testNoVersionCheck() throws IOException {
+    MiniMRCluster mr = null;
+    try {
+      JobConf jtConf = new JobConf();
+      jtConf.setBoolean(
+          CommonConfigurationKeys.HADOOP_SKIP_VERSION_CHECK_KEY, true);
+      mr = new MiniMRCluster(1, "file:///", 1, null, null, jtConf);
+      TaskTracker tt = mr.getTaskTrackerRunner(0).getTaskTracker();
+      String currFullVersion = VersionInfo.getBuildVersion();
+      String currVersion = VersionInfo.getVersion();
+
+      assertTrue(tt.isPermittedVersion(currFullVersion, currVersion));
+      assertTrue(tt.isPermittedVersion(currFullVersion+"x", currVersion+"x"));
+      assertTrue(tt.isPermittedVersion(currFullVersion+"x", currVersion));
+    } finally {
+      if (mr != null) {
+        mr.shutdown();
+      }
+    }
+  }
 }