Examples of org.apache.hadoop.hdfs.server.namenode.AvatarNode$InvalidatesCleaner

Package org.apache.hadoop.hdfs.server.namenode

Examples of org.apache.hadoop.hdfs.server.namenode.AvatarNode$InvalidatesCleaner

org.apache.hadoop.hdfs.server.namenode.AvatarNode
This is an implementation of the AvatarNode, a hot standby for the NameNode. This is really cool, believe me! The AvatarNode has two avatars.. the Standby avatar and the Active avatar. In the Standby avatar, the AvatarNode is consuming transaction logs generated by the primary (via a transaction log stored in a shared device). Typically, the primary Namenode is writing transactions to a NFS filesystem and the Standby is reading the log from the same NFS filesystem. The Standby is also making periodic checkpoints to the primary namenode. A manual command can switch the AvatarNode from the Standby avatar to the Active avatar. In the Active avatar, the AvatarNode performs precisely the same functionality as a real usual Namenode. The switching from Standby avatar to the Active avatar is fast and can typically occur within seconds. Typically, an adminstrator will run require two shared mount points for transaction logs. It has to be set in fs.name.dir.shared0 and fs.name.dir.shared1 (similarly for edits). Then the adminstrator starts the AvatarNode on two different machines as follows: bin/hadoop org.apache.hadoop.hdfs.server.namenode.AvatarNode -zero -active bin/hadoop org.apache.hadoop.hdfs.server.namenode.AvatarNode -one -standby The first AvatarNode uses fs.name.dir.shared0 while the second AvatarNode uses fs.name.dir.shared1 to write its transaction logs. Also, at startup, the first instance is the primary Namenode and the second instance is the Standby After a while, the adminstrator decides to change the avatar of the second instance to Active. In this case, he/she has to first ensure that the first instance is really really dead. This code does not handle the split-brain scenario where there are two active namenodes in one cluster.

  protected void doTestAcrossRestarts(boolean enableQJM) throws Exception {
    setUp("testAcrossRestarts", enableQJM);
    // 3 initial checkpoint
    createEdits(20); //23
    cluster.restartAvatarNodes(); //25 + 2 restart finalizes the segment + checkpoint
    AvatarNode primary = cluster.getPrimaryAvatar(0).avatar;
    AvatarNode standby = cluster.getStandbyAvatar(0).avatar;
    
    assertEquals(27, getCurrentTxId(primary));
    // give time to the standby to start-up
    Thread.sleep(2000);
    assertEquals(getCurrentTxId(primary), getCurrentTxId(standby));
    createEdits(20);
    if (!enableQJM) {
      standby.quiesceStandby(getCurrentTxId(primary)-1);
      assertEquals(47, getCurrentTxId(primary));
      assertEquals(getCurrentTxId(primary), getCurrentTxId(standby));
    }
  }

View Full Code Here

    TestAvatarTxIdsHandler h = new TestAvatarTxIdsHandler();
    InjectionHandler.set(h);
    setUp("testCheckpointAndRestart", false);
    // 3 initial checkpoint
    createEdits(20); //23
    AvatarNode primary = cluster.getPrimaryAvatar(0).avatar;
    AvatarNode standby = cluster.getStandbyAvatar(0).avatar;
    h.doCheckpoint(); //25 checkpoint adds 2 transactions
    createEdits(20);
    standby.quiesceStandby(getCurrentTxId(primary)-1);
    assertEquals(45, getCurrentTxId(primary));
    assertEquals(getCurrentTxId(primary), getCurrentTxId(standby));


    cluster.restartAvatarNodes(); //49 (restart + initial checkpoint)
    primary = cluster.getPrimaryAvatar(0).avatar;
    standby = cluster.getStandbyAvatar(0).avatar;
    createEdits(20);
    standby.quiesceStandby(getCurrentTxId(primary)-1);
     assertEquals(69, getCurrentTxId(primary));
     assertEquals(getCurrentTxId(primary), getCurrentTxId(standby));
    InjectionHandler.clear();
  }

View Full Code Here

  public void testStandbySaveNamespace() throws Exception {
    TestAvatarCheckpointingHandler h = new TestAvatarCheckpointingHandler();
    InjectionHandler.set(h);
    setUp(3600, "testStandbySaveNamespace", true);


    AvatarNode primary = cluster.getPrimaryAvatar(0).avatar;
    AvatarNode standby = cluster.getStandbyAvatar(0).avatar;
    createEdits(40);


    // trigger checkpoint on the standby
    AvatarShell shell = new AvatarShell(conf);


    // trigger regular checkpoint
    runAndAssertCommand(shell, 0, new String[] { "-one", "-saveNamespace" });
    assertCheckpointDone(h);


    // trigger regular checkpoint with uncompressed option
    // this will fail, as we do not support uncompressed
    runAndAssertCommand(shell, -1, new String[] { "-one", "-saveNamespace",
        "uncompressed" });


    // trigger regular checkpoint with force option
    runAndAssertCommand(shell, 0, new String[] { "-one", "-saveNamespace",
        "force" });
    assertCheckpointDone(h);


    // trigger regular checkpoint with force, uncompressed option
    // this will fail, as we do not support uncompressed
    runAndAssertCommand(shell, -1, new String[] { "-one", "-saveNamespace",
        "force", "uncompressed" });


    // ///////////////////


    // some incorrect option
    runAndAssertCommand(shell, -1, new String[] { "-one", "-saveNamespace",
        "someOption" });


    standby.quiesceStandby(DFSAvatarTestUtil.getCurrentTxId(primary) - 1);
    // edits + SLS + 2 * 2(for each successful checkpoint)
    assertEquals(45, DFSAvatarTestUtil.getCurrentTxId(primary));
  }

View Full Code Here

    InjectionHandler.set(h);
    setUp("testInodeIdWithCheckPoint", false);
    long expectedLastINodeId = INodeId.ROOT_INODE_ID;


    DFSTestUtil.createFile(fs, new Path("/testtwo/fileone"), 1024, (short) 1, 0);
    AvatarNode primaryAvatar = cluster.getPrimaryAvatar(0).avatar;
    AvatarNode standbyAvatar = cluster.getStandbyAvatar(0).avatar;
    FSNamesystem primaryNS = primaryAvatar.namesystem;
    FSNamesystem standbyNS = standbyAvatar.namesystem;
    expectedLastINodeId += 2;


    DFSAvatarTestUtil.assertTxnIdSync(primaryAvatar, standbyAvatar);

View Full Code Here

    TestAvatarCheckpointingHandler h = new TestAvatarCheckpointingHandler(null,
        null, true);
    InjectionHandler.set(h);
    setUp("testFailSuccFailQuiesce", enableQJM);
    createEdits(20);
    AvatarNode primary = cluster.getPrimaryAvatar(0).avatar;
    AvatarNode standby = cluster.getStandbyAvatar(0).avatar;
    
    try {
      h.failNextCheckpoint = true;
      h.doCheckpoint();
      fail("Should get IOException here");
    } catch (Exception e) { 
      // checkpoint fails during finalizations (see the checkpointing handler)
      assertTrue(e instanceof FinalizeCheckpointException);
      assertTrue(AvatarSetupUtil.isIngestAlive(standby));
      LOG.warn("Expected: Checkpoint failed", e);
    }
    
    // current txid should be 20 + SLS + ENS + SLS + initial
    assertEquals(25, getCurrentTxId(primary));
    
    h.failNextCheckpoint = false;
    
    // checkpoint should succeed
    h.doCheckpoint();
    
    // another roll adds 2 transactions
    assertEquals(27, getCurrentTxId(primary));
    
    h.failNextCheckpoint = true;    
    try {
      h.doCheckpoint();
      fail("Should get IOException here");
    } catch (Exception e) { 
      LOG.warn("Expected: Checkpoint failed", e);
    }
    
    // another roll adds 2 transactions
    assertEquals(29, getCurrentTxId(primary));
    
    if (!enableQJM) {
      createEdits(20);
      standby.quiesceStandby(getCurrentTxId(primary)-1);
      assertEquals(49, getCurrentTxId(primary));
      assertEquals(getCurrentTxId(primary), getCurrentTxId(standby));
    }
  }

View Full Code Here

    TestAvatarCheckpointingHandler h = new TestAvatarCheckpointingHandler(null,
        null, true);
    InjectionHandler.set(h);
    setUp("testFailCheckpointOnceAndRestartStandby", enableQJM);
    createEdits(20);
    AvatarNode primary = cluster.getPrimaryAvatar(0).avatar;
    AvatarNode standby = cluster.getStandbyAvatar(0).avatar;


    // Wait for first checkpoint.
    while (standby.getStandby().getLastCheckpointTime() == 0) {
      LOG.info("Waiting for standby to do checkpoint");
      Thread.sleep(1000);
    }
    
    try {
      h.failNextCheckpoint = true;
      h.doCheckpoint();
      fail("Should get IOException here");
    } catch (IOException e) { 
      // checkpoint fails during finalization (see the checkpointing handler)
      assertTrue(e instanceof FinalizeCheckpointException);
      assertTrue(AvatarSetupUtil.isIngestAlive(standby));
      LOG.info("Expected: Checkpoint failed", e);
    }
    
    // current txid should be 20 + SLS + ENS + SLS + initial ckpt
    assertEquals(25, getCurrentTxId(primary));
    
    cluster.killStandby(0);
    h.failNextCheckpoint = false;
    
    cluster.restartStandby(0); // ads one checkpoint
    standby = cluster.getStandbyAvatar(0).avatar;
    while (standby.getStandby().getLastCheckpointTime() == 0) {
      LOG.info("Waiting for standby to do checkpoint");
      Thread.sleep(1000);
    }
    
    LOG.info("Start another checkpointing...");
    // checkpoint should succeed
    h.doCheckpoint();
    
    LOG.info("Second checkpointing succeeded.");
    // roll adds two transactions
    assertEquals(29, getCurrentTxId(primary));
    
    if (!enableQJM) {
      createEdits(20);
      standby.quiesceStandby(getCurrentTxId(primary)-1);
      assertEquals(49, getCurrentTxId(primary));
      assertEquals(getCurrentTxId(primary), getCurrentTxId(standby));
    }
  }

View Full Code Here

    TestAvatarCheckpointingHandler h = new TestAvatarCheckpointingHandler(
        stopOnEvent, waitUntilEvent, scf);
    InjectionHandler.set(h);
    setUp(3, "testQuiesceInterruption", false, enableQJM); //simulate interruption, no ckpt failure   
    
    AvatarNode primary = cluster.getPrimaryAvatar(0).avatar;
    AvatarNode standby = cluster.getStandbyAvatar(0).avatar;
    createEdits(40);


    while (!h.receivedEvents.contains(stopOnEvent)) {
      LOG.info("Waiting for event : " + stopOnEvent);
      Thread.sleep(1000);
    }
    
    if (!enableQJM) {
      standby.quiesceStandby(getCurrentTxId(primary)-1);
      // only assert this for FileJournalManager.
      // edits + SLS + ELS + SLS (checkpoint fails, but roll happened)
      assertEquals(43, getCurrentTxId(primary));
      
      // if quiesce happened before roll, the standby will be behind by 1 transaction
      // which will be reclaimed by opening the log after
      long extraTransaction = rollAfterQuiesce ? 1 : 0;
      assertEquals(getCurrentTxId(primary), getCurrentTxId(standby) + extraTransaction);
    } else {
      standby.quiesceStandby(FSEditLogLoader.TXID_IGNORE);
    }
    
    // make sure the checkpoint indeed failed
    assertTrue(h.receivedEvents
        .contains(InjectionEvent.STANDBY_EXIT_CHECKPOINT_EXCEPTION));

View Full Code Here

  public static void tearDownAfterClass() throws Exception {
    MiniAvatarCluster.shutDownZooKeeper();
  }


  private long getSessionId(int index) throws Exception {
    AvatarNode primaryAvatar = cluster.getPrimaryAvatar(index).avatar;
    String address = primaryAvatar.getStartupConf().get(
        NameNode.DFS_NAMENODE_RPC_ADDRESS_KEY);
    return zkClient.getPrimarySsId(address, false);
  }

View Full Code Here

        StartupOption.SERVICE.getName(), nnInfo.nameserviceId,
        AvatarConstants.StartupOption.STANDBY.getName() };
    String[] federationArgs = (!standby) ? federationPrimaryArgs
        : federationStandbyArgs;
    String[] args = (federation) ? federationArgs : normalArgs;
    AvatarNode primary1 = MiniAvatarCluster.instantiateAvatarNode(
        args,
        MiniAvatarCluster.getServerConf(instance, nnInfo));
    extraNodes.add(primary1);
    if (singleStartup) {
      if (!standby) {
        assertEquals(primary1.getSessionId(), getSessionId(index));
      }
      return;
    }
    try {      
      AvatarNode second = MiniAvatarCluster.instantiateAvatarNode(args,
          MiniAvatarCluster.getServerConf(instance, nnInfo));
      extraNodes.add(second);
      fail("Did not throw exception");
    } catch (Exception e) {
      LOG.info("Expected exception : ", e);

View Full Code Here

    TestAvatarCheckpointingHandler h = new TestAvatarCheckpointingHandler(null,
        null, true);
    InjectionHandler.set(h);
    setUp("testFailCheckpointMultiAndCrash", true);
    createEdits(20);
    AvatarNode primary = cluster.getPrimaryAvatar(0).avatar;
    AvatarNode standby = cluster.getStandbyAvatar(0).avatar;
    
    try {
      h.failNextCheckpoint = true;
      h.doCheckpoint();
      fail("Should get IOException here");

View Full Code Here

0 1 2 3 4 5 6 7

TOP

Related Classes of org.apache.hadoop.hdfs.server.namenode.AvatarNode$InvalidatesCleaner

org.apache.hadoop.conf.Configuration

org.apache.hadoop.fs.FileSystem

org.apache.hadoop.fs.Path

org.apache.hadoop.hdfs.AvatarFailoverSnapshot

org.apache.hadoop.hdfs.AvatarZooKeeperClient

org.apache.hadoop.hdfs.FailoverTestUtil

org.apache.hadoop.hdfs.FileStatusExtended

org.apache.hadoop.hdfs.MiniAvatarCluster

org.apache.hadoop.hdfs.protocol.AvatarConstants.InstanceId

org.apache.hadoop.hdfs.protocol.AvatarConstants.StartupOption

All source code are property of their respective owners. Java is a trademark of Sun Microsystems, Inc and owned by ORACLE Inc. Contact coftware#gmail.com.