Java Code Examples for org.apache.hadoop.hdfs.MiniDFSCluster#getDataNode()

/**
 * Test the scenario where the NN fails over after issuing a block
 * synchronization request, but before it is committed. The
 * DN running the recovery should then fail to commit the synchronization
 * and a later retry will succeed.
 */
@Test(timeout=30000)
public void testFailoverRightBeforeCommitSynchronization() throws Exception {
  final Configuration conf = new Configuration();
  // Disable permissions so that another user can recover the lease.
  conf.setBoolean(DFSConfigKeys.DFS_PERMISSIONS_ENABLED_KEY, false);
  conf.setInt(DFSConfigKeys.DFS_BLOCK_SIZE_KEY, BLOCK_SIZE);
  
  FSDataOutputStream stm = null;
  final MiniDFSCluster cluster = new MiniDFSCluster.Builder(conf)
    .nnTopology(MiniDFSNNTopology.simpleHATopology())
    .numDataNodes(3)
    .build();
  try {
    cluster.waitActive();
    cluster.transitionToActive(0);
    Thread.sleep(500);

    LOG.info("Starting with NN 0 active");
    FileSystem fs = HATestUtil.configureFailoverFs(cluster, conf);
    stm = fs.create(TEST_PATH);
    
    // write a half block
    AppendTestUtil.write(stm, 0, BLOCK_SIZE / 2);
    stm.hflush();
    
    // Look into the block manager on the active node for the block
    // under construction.
    
    NameNode nn0 = cluster.getNameNode(0);
    ExtendedBlock blk = DFSTestUtil.getFirstBlock(fs, TEST_PATH);
    DatanodeDescriptor expectedPrimary =
        DFSTestUtil.getExpectedPrimaryNode(nn0, blk);
    LOG.info("Expecting block recovery to be triggered on DN " +
        expectedPrimary);
    
    // Find the corresponding DN daemon, and spy on its connection to the
    // active.
    DataNode primaryDN = cluster.getDataNode(expectedPrimary.getIpcPort());
    DatanodeProtocolClientSideTranslatorPB nnSpy =
        DataNodeTestUtils.spyOnBposToNN(primaryDN, nn0);
    
    // Delay the commitBlockSynchronization call
    DelayAnswer delayer = new DelayAnswer(LOG);
    Mockito.doAnswer(delayer).when(nnSpy).commitBlockSynchronization(
        Mockito.eq(blk),
        Mockito.anyInt(), // new genstamp
        Mockito.anyLong(), // new length
        Mockito.eq(true), // close file
        Mockito.eq(false), // delete block
        (DatanodeID[]) Mockito.anyObject(), // new targets
        (String[]) Mockito.anyObject()); // new target storages

    DistributedFileSystem fsOtherUser = createFsAsOtherUser(cluster, conf);
    assertFalse(fsOtherUser.recoverLease(TEST_PATH));
    
    LOG.info("Waiting for commitBlockSynchronization call from primary");
    delayer.waitForCall();

    LOG.info("Failing over to NN 1");
    
    cluster.transitionToStandby(0);
    cluster.transitionToActive(1);
    
    // Let the commitBlockSynchronization call go through, and check that
    // it failed with the correct exception.
    delayer.proceed();
    delayer.waitForResult();
    Throwable t = delayer.getThrown();
    if (t == null) {
      fail("commitBlockSynchronization call did not fail on standby");
    }
    GenericTestUtils.assertExceptionContains(
        "Operation category WRITE is not supported",
        t);
    
    // Now, if we try again to recover the block, it should succeed on the new
    // active.
    loopRecoverLease(fsOtherUser, TEST_PATH);
    
    AppendTestUtil.check(fs, TEST_PATH, BLOCK_SIZE/2);
  } finally {
    IOUtils.closeStream(stm);
    cluster.shutdown();
  }
}
 

/**
 * The following test first creates a file.
 * It verifies the block information from a datanode.
 * Then, it updates the block with new information and verifies again.
 * @param useDnHostname whether DNs should connect to other DNs by hostname
 */
private void checkBlockMetaDataInfo(boolean useDnHostname) throws Exception {
  MiniDFSCluster cluster = null;

  conf.setBoolean(DFSConfigKeys.DFS_DATANODE_USE_DN_HOSTNAME, useDnHostname);
  if (useDnHostname) {
    // Since the mini cluster only listens on the loopback we have to
    // ensure the hostname used to access DNs maps to the loopback. We
    // do this by telling the DN to advertise localhost as its hostname
    // instead of the default hostname.
    conf.set(DFSConfigKeys.DFS_DATANODE_HOST_NAME_KEY, "localhost");
  }

  try {
    cluster = new MiniDFSCluster.Builder(conf)
      .numDataNodes(3)
      .checkDataNodeHostConfig(true)
      .build();
    cluster.waitActive();

    //create a file
    DistributedFileSystem dfs = cluster.getFileSystem();
    String filestr = "/foo";
    Path filepath = new Path(filestr);
    DFSTestUtil.createFile(dfs, filepath, 1024L, (short)3, 0L);
    assertTrue(dfs.exists(filepath));

    //get block info
    LocatedBlock locatedblock = getLastLocatedBlock(
        DFSClientAdapter.getDFSClient(dfs).getNamenode(), filestr);
    DatanodeInfo[] datanodeinfo = locatedblock.getLocations();
    assertTrue(datanodeinfo.length > 0);

    //connect to a data node
    DataNode datanode = cluster.getDataNode(datanodeinfo[0].getIpcPort());
    InterDatanodeProtocol idp = DataNodeTestUtils.createInterDatanodeProtocolProxy(
        datanode, datanodeinfo[0], conf, useDnHostname);
    
    // Stop the block scanners.
    datanode.getBlockScanner().removeAllVolumeScanners();

    //verify BlockMetaDataInfo
    ExtendedBlock b = locatedblock.getBlock();
    InterDatanodeProtocol.LOG.info("b=" + b + ", " + b.getClass());
    checkMetaInfo(b, datanode);
    long recoveryId = b.getGenerationStamp() + 1;
    idp.initReplicaRecovery(
        new RecoveringBlock(b, locatedblock.getLocations(), recoveryId));

    //verify updateBlock
    ExtendedBlock newblock = new ExtendedBlock(b.getBlockPoolId(),
        b.getBlockId(), b.getNumBytes()/2, b.getGenerationStamp()+1);
    idp.updateReplicaUnderRecovery(b, recoveryId, b.getBlockId(),
        newblock.getNumBytes());
    checkMetaInfo(newblock, datanode);
    
    // Verify correct null response trying to init recovery for a missing block
    ExtendedBlock badBlock = new ExtendedBlock("fake-pool",
        b.getBlockId(), 0, 0);
    assertNull(idp.initReplicaRecovery(
        new RecoveringBlock(badBlock,
            locatedblock.getLocations(), recoveryId)));
  }
  finally {
    if (cluster != null) {cluster.shutdown();}
  }
}
 

/** 
 * Test  for
 * {@link FsDatasetImpl#updateReplicaUnderRecovery(ExtendedBlock, long, long)} 
 * */
@Test
public void testUpdateReplicaUnderRecovery() throws IOException {
  MiniDFSCluster cluster = null;

  try {
    cluster = new MiniDFSCluster.Builder(conf).numDataNodes(3).build();
    cluster.waitActive();
    String bpid = cluster.getNamesystem().getBlockPoolId();

    //create a file
    DistributedFileSystem dfs = cluster.getFileSystem();
    String filestr = "/foo";
    Path filepath = new Path(filestr);
    DFSTestUtil.createFile(dfs, filepath, 1024L, (short)3, 0L);

    //get block info
    final LocatedBlock locatedblock = getLastLocatedBlock(
        DFSClientAdapter.getDFSClient(dfs).getNamenode(), filestr);
    final DatanodeInfo[] datanodeinfo = locatedblock.getLocations();
    Assert.assertTrue(datanodeinfo.length > 0);

    //get DataNode and FSDataset objects
    final DataNode datanode = cluster.getDataNode(datanodeinfo[0].getIpcPort());
    Assert.assertTrue(datanode != null);

    //initReplicaRecovery
    final ExtendedBlock b = locatedblock.getBlock();
    final long recoveryid = b.getGenerationStamp() + 1;
    final long newlength = b.getNumBytes() - 1;
    final FsDatasetSpi<?> fsdataset = DataNodeTestUtils.getFSDataset(datanode);
    final ReplicaRecoveryInfo rri = fsdataset.initReplicaRecovery(
        new RecoveringBlock(b, null, recoveryid));

    //check replica
    final ReplicaInfo replica = FsDatasetTestUtil.fetchReplicaInfo(
        fsdataset, bpid, b.getBlockId());
    Assert.assertEquals(ReplicaState.RUR, replica.getState());

    //check meta data before update
    FsDatasetImpl.checkReplicaFiles(replica);

    //case "THIS IS NOT SUPPOSED TO HAPPEN"
    //with (block length) != (stored replica's on disk length). 
    {
      //create a block with same id and gs but different length.
      final ExtendedBlock tmp = new ExtendedBlock(b.getBlockPoolId(), rri
          .getBlockId(), rri.getNumBytes() - 1, rri.getGenerationStamp());
      try {
        //update should fail
        fsdataset.updateReplicaUnderRecovery(tmp, recoveryid,
            tmp.getBlockId(), newlength);
        Assert.fail();
      } catch(IOException ioe) {
        System.out.println("GOOD: getting " + ioe);
      }
    }

    //update
    final String storageID = fsdataset.updateReplicaUnderRecovery(
        new ExtendedBlock(b.getBlockPoolId(), rri), recoveryid,
        rri.getBlockId(), newlength);
    assertTrue(storageID != null);

  } finally {
    if (cluster != null) cluster.shutdown();
  }
}
 

/**
 * Test the scenario where the NN fails over after issuing a block
 * synchronization request, but before it is committed. The
 * DN running the recovery should then fail to commit the synchronization
 * and a later retry will succeed.
 */
@Test(timeout=30000)
public void testFailoverRightBeforeCommitSynchronization() throws Exception {
  final Configuration conf = new Configuration();
  // Disable permissions so that another user can recover the lease.
  conf.setBoolean(DFSConfigKeys.DFS_PERMISSIONS_ENABLED_KEY, false);
  conf.setInt(DFSConfigKeys.DFS_BLOCK_SIZE_KEY, BLOCK_SIZE);
  
  FSDataOutputStream stm = null;
  final MiniDFSCluster cluster = new MiniDFSCluster.Builder(conf)
    .nnTopology(MiniDFSNNTopology.simpleHATopology())
    .numDataNodes(3)
    .build();
  try {
    cluster.waitActive();
    cluster.transitionToActive(0);
    Thread.sleep(500);

    LOG.info("Starting with NN 0 active");
    FileSystem fs = HATestUtil.configureFailoverFs(cluster, conf);
    stm = fs.create(TEST_PATH);
    
    // write a half block
    AppendTestUtil.write(stm, 0, BLOCK_SIZE / 2);
    stm.hflush();
    
    // Look into the block manager on the active node for the block
    // under construction.
    
    NameNode nn0 = cluster.getNameNode(0);
    ExtendedBlock blk = DFSTestUtil.getFirstBlock(fs, TEST_PATH);
    DatanodeDescriptor expectedPrimary =
        DFSTestUtil.getExpectedPrimaryNode(nn0, blk);
    LOG.info("Expecting block recovery to be triggered on DN " +
        expectedPrimary);
    
    // Find the corresponding DN daemon, and spy on its connection to the
    // active.
    DataNode primaryDN = cluster.getDataNode(expectedPrimary.getIpcPort());
    DatanodeProtocolClientSideTranslatorPB nnSpy =
        DataNodeTestUtils.spyOnBposToNN(primaryDN, nn0);
    
    // Delay the commitBlockSynchronization call
    DelayAnswer delayer = new DelayAnswer(LOG);
    Mockito.doAnswer(delayer).when(nnSpy).commitBlockSynchronization(
        Mockito.eq(blk),
        Mockito.anyInt(), // new genstamp
        Mockito.anyLong(), // new length
        Mockito.eq(true), // close file
        Mockito.eq(false), // delete block
        (DatanodeID[]) Mockito.anyObject(), // new targets
        (String[]) Mockito.anyObject()); // new target storages

    DistributedFileSystem fsOtherUser = createFsAsOtherUser(cluster, conf);
    assertFalse(fsOtherUser.recoverLease(TEST_PATH));
    
    LOG.info("Waiting for commitBlockSynchronization call from primary");
    delayer.waitForCall();

    LOG.info("Failing over to NN 1");
    
    cluster.transitionToStandby(0);
    cluster.transitionToActive(1);
    
    // Let the commitBlockSynchronization call go through, and check that
    // it failed with the correct exception.
    delayer.proceed();
    delayer.waitForResult();
    Throwable t = delayer.getThrown();
    if (t == null) {
      fail("commitBlockSynchronization call did not fail on standby");
    }
    GenericTestUtils.assertExceptionContains(
        "Operation category WRITE is not supported",
        t);
    
    // Now, if we try again to recover the block, it should succeed on the new
    // active.
    loopRecoverLease(fsOtherUser, TEST_PATH);
    
    AppendTestUtil.check(fs, TEST_PATH, BLOCK_SIZE/2);
  } finally {
    IOUtils.closeStream(stm);
    cluster.shutdown();
  }
}
 

/**
 * The following test first creates a file.
 * It verifies the block information from a datanode.
 * Then, it updates the block with new information and verifies again.
 * @param useDnHostname whether DNs should connect to other DNs by hostname
 */
private void checkBlockMetaDataInfo(boolean useDnHostname) throws Exception {
  MiniDFSCluster cluster = null;

  conf.setBoolean(DFSConfigKeys.DFS_DATANODE_USE_DN_HOSTNAME, useDnHostname);
  if (useDnHostname) {
    // Since the mini cluster only listens on the loopback we have to
    // ensure the hostname used to access DNs maps to the loopback. We
    // do this by telling the DN to advertise localhost as its hostname
    // instead of the default hostname.
    conf.set(DFSConfigKeys.DFS_DATANODE_HOST_NAME_KEY, "localhost");
  }

  try {
    cluster = new MiniDFSCluster.Builder(conf)
      .numDataNodes(3)
      .checkDataNodeHostConfig(true)
      .build();
    cluster.waitActive();

    //create a file
    DistributedFileSystem dfs = cluster.getFileSystem();
    String filestr = "/foo";
    Path filepath = new Path(filestr);
    DFSTestUtil.createFile(dfs, filepath, 1024L, (short)3, 0L);
    assertTrue(dfs.exists(filepath));

    //get block info
    LocatedBlock locatedblock = getLastLocatedBlock(
        DFSClientAdapter.getDFSClient(dfs).getNamenode(), filestr);
    DatanodeInfo[] datanodeinfo = locatedblock.getLocations();
    assertTrue(datanodeinfo.length > 0);

    //connect to a data node
    DataNode datanode = cluster.getDataNode(datanodeinfo[0].getIpcPort());
    InterDatanodeProtocol idp = DataNodeTestUtils.createInterDatanodeProtocolProxy(
        datanode, datanodeinfo[0], conf, useDnHostname);
    
    // Stop the block scanners.
    datanode.getBlockScanner().removeAllVolumeScanners();

    //verify BlockMetaDataInfo
    ExtendedBlock b = locatedblock.getBlock();
    InterDatanodeProtocol.LOG.info("b=" + b + ", " + b.getClass());
    checkMetaInfo(b, datanode);
    long recoveryId = b.getGenerationStamp() + 1;
    idp.initReplicaRecovery(
        new RecoveringBlock(b, locatedblock.getLocations(), recoveryId));

    //verify updateBlock
    ExtendedBlock newblock = new ExtendedBlock(b.getBlockPoolId(),
        b.getBlockId(), b.getNumBytes()/2, b.getGenerationStamp()+1);
    idp.updateReplicaUnderRecovery(b, recoveryId, b.getBlockId(),
        newblock.getNumBytes());
    checkMetaInfo(newblock, datanode);
    
    // Verify correct null response trying to init recovery for a missing block
    ExtendedBlock badBlock = new ExtendedBlock("fake-pool",
        b.getBlockId(), 0, 0);
    assertNull(idp.initReplicaRecovery(
        new RecoveringBlock(badBlock,
            locatedblock.getLocations(), recoveryId)));
  }
  finally {
    if (cluster != null) {cluster.shutdown();}
  }
}
 

/** 
 * Test  for
 * {@link FsDatasetImpl#updateReplicaUnderRecovery(ExtendedBlock, long, long)} 
 * */
@Test
public void testUpdateReplicaUnderRecovery() throws IOException {
  MiniDFSCluster cluster = null;

  try {
    cluster = new MiniDFSCluster.Builder(conf).numDataNodes(3).build();
    cluster.waitActive();
    String bpid = cluster.getNamesystem().getBlockPoolId();

    //create a file
    DistributedFileSystem dfs = cluster.getFileSystem();
    String filestr = "/foo";
    Path filepath = new Path(filestr);
    DFSTestUtil.createFile(dfs, filepath, 1024L, (short)3, 0L);

    //get block info
    final LocatedBlock locatedblock = getLastLocatedBlock(
        DFSClientAdapter.getDFSClient(dfs).getNamenode(), filestr);
    final DatanodeInfo[] datanodeinfo = locatedblock.getLocations();
    Assert.assertTrue(datanodeinfo.length > 0);

    //get DataNode and FSDataset objects
    final DataNode datanode = cluster.getDataNode(datanodeinfo[0].getIpcPort());
    Assert.assertTrue(datanode != null);

    //initReplicaRecovery
    final ExtendedBlock b = locatedblock.getBlock();
    final long recoveryid = b.getGenerationStamp() + 1;
    final long newlength = b.getNumBytes() - 1;
    final FsDatasetSpi<?> fsdataset = DataNodeTestUtils.getFSDataset(datanode);
    final ReplicaRecoveryInfo rri = fsdataset.initReplicaRecovery(
        new RecoveringBlock(b, null, recoveryid));

    //check replica
    final ReplicaInfo replica = FsDatasetTestUtil.fetchReplicaInfo(
        fsdataset, bpid, b.getBlockId());
    Assert.assertEquals(ReplicaState.RUR, replica.getState());

    //check meta data before update
    FsDatasetImpl.checkReplicaFiles(replica);

    //case "THIS IS NOT SUPPOSED TO HAPPEN"
    //with (block length) != (stored replica's on disk length). 
    {
      //create a block with same id and gs but different length.
      final ExtendedBlock tmp = new ExtendedBlock(b.getBlockPoolId(), rri
          .getBlockId(), rri.getNumBytes() - 1, rri.getGenerationStamp());
      try {
        //update should fail
        fsdataset.updateReplicaUnderRecovery(tmp, recoveryid,
            tmp.getBlockId(), newlength);
        Assert.fail();
      } catch(IOException ioe) {
        System.out.println("GOOD: getting " + ioe);
      }
    }

    //update
    final String storageID = fsdataset.updateReplicaUnderRecovery(
        new ExtendedBlock(b.getBlockPoolId(), rri), recoveryid,
        rri.getBlockId(), newlength);
    assertTrue(storageID != null);

  } finally {
    if (cluster != null) cluster.shutdown();
  }
}