Java Code Examples for org.apache.hadoop.hdfs.DistributedFileSystem#create()

/**
 * Test for save namespace should succeed when parent directory renamed with
 * open lease and destination directory exist. 
 * This test is a regression for HDFS-2827
 */
@Test
public void testSaveNamespaceWithRenamedLease() throws Exception {
  MiniDFSCluster cluster = new MiniDFSCluster.Builder(new Configuration())
      .numDataNodes(1).build();
  cluster.waitActive();
  DistributedFileSystem fs = (DistributedFileSystem) cluster.getFileSystem();
  OutputStream out = null;
  try {
    fs.mkdirs(new Path("/test-target"));
    out = fs.create(new Path("/test-source/foo")); // don't close
    fs.rename(new Path("/test-source/"), new Path("/test-target/"));

    fs.setSafeMode(SafeModeAction.SAFEMODE_ENTER);
    cluster.getNameNodeRpc().saveNamespace();
    fs.setSafeMode(SafeModeAction.SAFEMODE_LEAVE);
  } finally {
    IOUtils.cleanup(LOG, out, fs);
    if (cluster != null) {
      cluster.shutdown();
    }
  }
}
 

@Test
public void testLoadingFomHDFS() throws Exception {
  HTU.startMiniDFSCluster(3);
  try {
    MiniDFSCluster cluster = HTU.getDFSCluster();
    DistributedFileSystem fs = cluster.getFileSystem();
    // Writing file
    Path path = new Path(fs.getHomeDirectory(), DEFAULT_RULES_FILE_NAME);
    FSDataOutputStream stream = fs.create(path);
    stream.write("server1 10".getBytes());
    stream.flush();
    stream.close();

    Configuration configuration = HTU.getConfiguration();

    // start costFunction
    configuration.set(
      HeterogeneousRegionCountCostFunction.HBASE_MASTER_BALANCER_HETEROGENEOUS_RULES_FILE,
      path.toString());
    this.costFunction = new HeterogeneousRegionCountCostFunction(configuration);
    this.costFunction.loadRules();
    Assert.assertEquals(1, this.costFunction.getNumberOfRulesLoaded());
  } finally {
    HTU.shutdownMiniCluster();
  }
}
 

@Test
public void testScheduleSameBlock() throws IOException {
  final Configuration conf = new HdfsConfiguration();
  final MiniDFSCluster cluster = new MiniDFSCluster.Builder(conf)
      .numDataNodes(4).build();
  try {
    cluster.waitActive();
    final DistributedFileSystem dfs = cluster.getFileSystem();
    final String file = "/testScheduleSameBlock/file";
    
    {
      final FSDataOutputStream out = dfs.create(new Path(file));
      out.writeChars("testScheduleSameBlock");
      out.close();
    }

    final Mover mover = newMover(conf);
    mover.init();
    final Mover.Processor processor = mover.new Processor();

    final LocatedBlock lb = dfs.getClient().getLocatedBlocks(file, 0).get(0);
    final List<MLocation> locations = MLocation.toLocations(lb);
    final MLocation ml = locations.get(0);
    final DBlock db = mover.newDBlock(lb.getBlock().getLocalBlock(), locations);

    final List<StorageType> storageTypes = new ArrayList<StorageType>(
        Arrays.asList(StorageType.DEFAULT, StorageType.DEFAULT));
    Assert.assertTrue(processor.scheduleMoveReplica(db, ml, storageTypes));
    Assert.assertFalse(processor.scheduleMoveReplica(db, ml, storageTypes));
  } finally {
    cluster.shutdown();
  }
}
 

@Test
public void testMoverFailedRetry() throws Exception {
  // HDFS-8147
  final Configuration conf = new HdfsConfiguration();
  conf.set(DFSConfigKeys.DFS_MOVER_RETRY_MAX_ATTEMPTS_KEY, "2");
  final MiniDFSCluster cluster = new MiniDFSCluster.Builder(conf)
      .numDataNodes(3)
      .storageTypes(
          new StorageType[][] {{StorageType.DISK, StorageType.ARCHIVE},
              {StorageType.DISK, StorageType.ARCHIVE},
              {StorageType.DISK, StorageType.ARCHIVE}}).build();
  try {
    cluster.waitActive();
    final DistributedFileSystem dfs = cluster.getFileSystem();
    final String file = "/testMoverFailedRetry";
    // write to DISK
    final FSDataOutputStream out = dfs.create(new Path(file), (short) 2);
    out.writeChars("testMoverFailedRetry");
    out.close();

    // Delete block file so, block move will fail with FileNotFoundException
    LocatedBlock lb = dfs.getClient().getLocatedBlocks(file, 0).get(0);
    cluster.corruptBlockOnDataNodesByDeletingBlockFile(lb.getBlock());
    // move to ARCHIVE
    dfs.setStoragePolicy(new Path(file), "COLD");
    int rc = ToolRunner.run(conf, new Mover.Cli(),
        new String[] {"-p", file.toString()});
    Assert.assertEquals("Movement should fail after some retry",
        ExitStatus.IO_EXCEPTION.getExitCode(), rc);
  } finally {
    cluster.shutdown();
  }
}
 

/** This creates a slow writer and check to see
  * if pipeline heartbeats work fine
  */
 public void testStuckDataNode() throws Exception {
   final int DATANODE_NUM = 3;
   Configuration conf = new Configuration();
   final int timeout = 8000;
   conf.setInt("dfs.socket.timeout",timeout);

   final Path p = new Path("/pipelineHeartbeat/foo");
   System.out.println("p=" + p);

   MiniDFSCluster cluster = new MiniDFSCluster(conf, DATANODE_NUM, true, null);
   DistributedFileSystem fs = (DistributedFileSystem)cluster.getFileSystem();

DataNodeMetrics metrics = cluster.getDataNodes().get(0).myMetrics;
MetricsTimeVaryingLong spyBytesWritten = spy(metrics.bytesWritten);
DelayAnswer delayAnswer = new DelayAnswer(); 
doAnswer(delayAnswer).when(spyBytesWritten).inc(anyInt());
metrics.bytesWritten = spyBytesWritten;

try {
   	// create a new file.
   	FSDataOutputStream stm = fs.create(p);
   	stm.write(1);
   	stm.sync();
   	stm.write(2);
   	stm.close();

   	// verify that entire file is good
   	FSDataInputStream in = fs.open(p);
   	assertEquals(1, in.read());
   	assertEquals(2, in.read());
   	in.close();
   } finally {
     fs.close();
     cluster.shutdown();
   }
 }
 

@Test
public void testAbandon() throws Exception {
  NameNode nn = cluster.getNameNode();
  DistributedFileSystem fs = (DistributedFileSystem) cluster.getFileSystem();
  String fileName = "/testAbandon";
  fs.create(new Path(fileName));
  LocatedBlock lbk = nn.addBlock(fileName, fs.getClient().getClientName());
  INodeFileUnderConstruction cons = (INodeFileUnderConstruction) nn.namesystem.dir
      .getInode(fileName);
  cons.setTargets(null);
  nn.abandonBlock(lbk.getBlock(), fileName, fs.getClient().getClientName());
  assertEquals(0, nn.getBlockLocations(fileName, 0, Long.MAX_VALUE)
      .locatedBlockCount());
}
 

@BeforeClass
public static void setUpBeforeClass() throws Exception {
  try {
    MiniDFSCluster cluster = UTIL.startMiniDFSCluster(3); // Need 3 DNs for RS-3-2 policy
    DistributedFileSystem fs = cluster.getFileSystem();

    Method enableAllECPolicies =
      DFSTestUtil.class.getMethod("enableAllECPolicies", DistributedFileSystem.class);
    enableAllECPolicies.invoke(null, fs);

    DFSClient client = fs.getClient();
    Method setErasureCodingPolicy =
      DFSClient.class.getMethod("setErasureCodingPolicy", String.class, String.class);
    setErasureCodingPolicy.invoke(client, "/", "RS-3-2-1024k"); // try a built-in policy

    try (FSDataOutputStream out = fs.create(new Path("/canary"))) {
      // If this comes back as having hflush then some test setup assumption is wrong.
      // Fail the test so that a developer has to look and triage
      assertFalse("Did not enable EC!", out.hasCapability(StreamCapabilities.HFLUSH));
    }
  } catch (NoSuchMethodException e) {
    // We're not testing anything interesting if EC is not available, so skip the rest of the test
    Assume.assumeNoException("Using an older version of hadoop; EC not available.", e);
  }

  UTIL.getConfiguration().setBoolean(CommonFSUtils.UNSAFE_STREAM_CAPABILITY_ENFORCE, true);

}
 

@Test
public void testReceivePacketMetrics() throws Exception {
  Configuration conf = new HdfsConfiguration();
  final int interval = 1;
  conf.set(DFSConfigKeys.DFS_METRICS_PERCENTILES_INTERVALS_KEY, "" + interval);
  MiniDFSCluster cluster = new MiniDFSCluster.Builder(conf).build();
  try {
    cluster.waitActive();
    DistributedFileSystem fs = cluster.getFileSystem();

    Path testFile = new Path("/testFlushNanosMetric.txt");
    FSDataOutputStream fout = fs.create(testFile);
    fout.write(new byte[1]);
    fout.hsync();
    fout.close();
    List<DataNode> datanodes = cluster.getDataNodes();
    DataNode datanode = datanodes.get(0);
    MetricsRecordBuilder dnMetrics = getMetrics(datanode.getMetrics().name());
    // Expect two flushes, 1 for the flush that occurs after writing, 
    // 1 that occurs on closing the data and metadata files.
    assertCounter("FlushNanosNumOps", 2L, dnMetrics);
    // Expect two syncs, one from the hsync, one on close.
    assertCounter("FsyncNanosNumOps", 2L, dnMetrics);
    // Wait for at least 1 rollover
    Thread.sleep((interval + 1) * 1000);
    // Check the receivePacket percentiles that should be non-zero
    String sec = interval + "s";
    assertQuantileGauges("FlushNanos" + sec, dnMetrics);
    assertQuantileGauges("FsyncNanos" + sec, dnMetrics);
  } finally {
    if (cluster != null) {cluster.shutdown();}
  }
}
 

private void writeFile(final DistributedFileSystem dfs,
    Path dir, String fileName) throws IOException {
  Path filePath = new Path(dir.toString() + Path.SEPARATOR + fileName);
  final FSDataOutputStream out = dfs.create(filePath);
  out.writeChars("teststring");
  out.close();
}
 

@Test
public void testMoverFailedRetry() throws Exception {
  // HDFS-8147
  final Configuration conf = new HdfsConfiguration();
  conf.set(DFSConfigKeys.DFS_MOVER_RETRY_MAX_ATTEMPTS_KEY, "2");
  final MiniDFSCluster cluster = new MiniDFSCluster.Builder(conf)
      .numDataNodes(3)
      .storageTypes(
          new StorageType[][] {{StorageType.DISK, StorageType.ARCHIVE},
              {StorageType.DISK, StorageType.ARCHIVE},
              {StorageType.DISK, StorageType.ARCHIVE}}).build();
  try {
    cluster.waitActive();
    final DistributedFileSystem dfs = cluster.getFileSystem();
    final String file = "/testMoverFailedRetry";
    // write to DISK
    final FSDataOutputStream out = dfs.create(new Path(file), (short) 2);
    out.writeChars("testMoverFailedRetry");
    out.close();

    // Delete block file so, block move will fail with FileNotFoundException
    LocatedBlock lb = dfs.getClient().getLocatedBlocks(file, 0).get(0);
    cluster.corruptBlockOnDataNodesByDeletingBlockFile(lb.getBlock());
    // move to ARCHIVE
    dfs.setStoragePolicy(new Path(file), "COLD");
    int rc = ToolRunner.run(conf, new Mover.Cli(),
        new String[] {"-p", file.toString()});
    Assert.assertEquals("Movement should fail after some retry",
        ExitStatus.IO_EXCEPTION.getExitCode(), rc);
  } finally {
    cluster.shutdown();
  }
}
 

@Test
public void testTransferRbw() throws Exception {
  final HdfsConfiguration conf = new HdfsConfiguration();
  final MiniDFSCluster cluster = new MiniDFSCluster.Builder(conf
      ).numDataNodes(REPLICATION).build();
  try {
    cluster.waitActive();
    final DistributedFileSystem fs = cluster.getFileSystem();

    //create a file, write some data and leave it open. 
    final Path p = new Path("/foo");
    final int size = (1 << 16) + RAN.nextInt(1 << 16);
    LOG.info("size = " + size);
    final FSDataOutputStream out = fs.create(p, REPLICATION);
    final byte[] bytes = new byte[1024];
    for(int remaining = size; remaining > 0; ) {
      RAN.nextBytes(bytes);
      final int len = bytes.length < remaining? bytes.length: remaining;
      out.write(bytes, 0, len);
      out.hflush();
      remaining -= len;
    }

    //get the RBW
    final ReplicaBeingWritten oldrbw;
    final DataNode newnode;
    final DatanodeInfo newnodeinfo;
    final String bpid = cluster.getNamesystem().getBlockPoolId();
    {
      final DataNode oldnode = cluster.getDataNodes().get(0);
      oldrbw = getRbw(oldnode, bpid);
      LOG.info("oldrbw = " + oldrbw);
      
      //add a datanode
      cluster.startDataNodes(conf, 1, true, null, null);
      newnode = cluster.getDataNodes().get(REPLICATION);
      
      final DatanodeInfo oldnodeinfo;
      {
        final DatanodeInfo[] datatnodeinfos = cluster.getNameNodeRpc(
            ).getDatanodeReport(DatanodeReportType.LIVE);
        Assert.assertEquals(2, datatnodeinfos.length);
        int i = 0;
        for(DatanodeRegistration dnReg = newnode.getDNRegistrationForBP(bpid);
            i < datatnodeinfos.length && !datatnodeinfos[i].equals(dnReg); i++);
        Assert.assertTrue(i < datatnodeinfos.length);
        newnodeinfo = datatnodeinfos[i];
        oldnodeinfo = datatnodeinfos[1 - i];
      }
      
      //transfer RBW
      final ExtendedBlock b = new ExtendedBlock(bpid, oldrbw.getBlockId(), oldrbw.getBytesAcked(),
          oldrbw.getGenerationStamp());
      final BlockOpResponseProto s = DFSTestUtil.transferRbw(
          b, DFSClientAdapter.getDFSClient(fs), oldnodeinfo, newnodeinfo);
      Assert.assertEquals(Status.SUCCESS, s.getStatus());
    }

    //check new rbw
    final ReplicaBeingWritten newrbw = getRbw(newnode, bpid);
    LOG.info("newrbw = " + newrbw);
    Assert.assertEquals(oldrbw.getBlockId(), newrbw.getBlockId());
    Assert.assertEquals(oldrbw.getGenerationStamp(), newrbw.getGenerationStamp());
    Assert.assertEquals(oldrbw.getVisibleLength(), newrbw.getVisibleLength());

    LOG.info("DONE");
  } finally {
    cluster.shutdown();
  }
}
 

private void testPersistHelper(Configuration conf) throws IOException {
  MiniDFSCluster cluster = null;
  try {
    cluster = new MiniDFSCluster.Builder(conf).build();
    cluster.waitActive();
    FSNamesystem fsn = cluster.getNamesystem();
    DistributedFileSystem fs = cluster.getFileSystem();

    final Path dir = new Path("/abc/def");
    final Path file1 = new Path(dir, "f1");
    final Path file2 = new Path(dir, "f2");

    // create an empty file f1
    fs.create(file1).close();

    // create an under-construction file f2
    FSDataOutputStream out = fs.create(file2);
    out.writeBytes("hello");
    ((DFSOutputStream) out.getWrappedStream()).hsync(EnumSet
        .of(SyncFlag.UPDATE_LENGTH));

    // checkpoint
    fs.setSafeMode(SafeModeAction.SAFEMODE_ENTER);
    fs.saveNamespace();
    fs.setSafeMode(SafeModeAction.SAFEMODE_LEAVE);

    cluster.restartNameNode();
    cluster.waitActive();
    fs = cluster.getFileSystem();

    assertTrue(fs.isDirectory(dir));
    assertTrue(fs.exists(file1));
    assertTrue(fs.exists(file2));

    // check internals of file2
    INodeFile file2Node = fsn.dir.getINode4Write(file2.toString()).asFile();
    assertEquals("hello".length(), file2Node.computeFileSize());
    assertTrue(file2Node.isUnderConstruction());
    BlockInfoContiguous[] blks = file2Node.getBlocks();
    assertEquals(1, blks.length);
    assertEquals(BlockUCState.UNDER_CONSTRUCTION, blks[0].getBlockUCState());
    // check lease manager
    Lease lease = fsn.leaseManager.getLeaseByPath(file2.toString());
    Assert.assertNotNull(lease);
  } finally {
    if (cluster != null) {
      cluster.shutdown();
    }
  }
}
 

@BeforeClass
public static void createOriginalFSImage() throws IOException {
  MiniDFSCluster cluster = null;
  try {
    Configuration conf = new Configuration();
    conf.setLong(
        DFSConfigKeys.DFS_NAMENODE_DELEGATION_TOKEN_MAX_LIFETIME_KEY, 10000);
    conf.setLong(
        DFSConfigKeys.DFS_NAMENODE_DELEGATION_TOKEN_RENEW_INTERVAL_KEY, 5000);
    conf.setBoolean(
        DFSConfigKeys.DFS_NAMENODE_DELEGATION_TOKEN_ALWAYS_USE_KEY, true);
    conf.set(CommonConfigurationKeysPublic.HADOOP_SECURITY_AUTH_TO_LOCAL,
        "RULE:[2:$1@$0](JobTracker@.*FOO.COM)s/@.*//" + "DEFAULT");
    cluster = new MiniDFSCluster.Builder(conf).numDataNodes(1).build();
    cluster.waitActive();
    DistributedFileSystem hdfs = cluster.getFileSystem();

    // Create a reasonable namespace
    for (int i = 0; i < NUM_DIRS; i++) {
      Path dir = new Path("/dir" + i);
      hdfs.mkdirs(dir);
      writtenFiles.put(dir.toString(), pathToFileEntry(hdfs, dir.toString()));
      for (int j = 0; j < FILES_PER_DIR; j++) {
        Path file = new Path(dir, "file" + j);
        FSDataOutputStream o = hdfs.create(file);
        o.write(23);
        o.close();

        writtenFiles.put(file.toString(),
            pathToFileEntry(hdfs, file.toString()));
      }
    }

    // Create an empty directory
    Path emptydir = new Path("/emptydir");
    hdfs.mkdirs(emptydir);
    writtenFiles.put(emptydir.toString(), hdfs.getFileStatus(emptydir));

    //Create a directory whose name should be escaped in XML
    Path invalidXMLDir = new Path("/dirContainingInvalidXMLChar\u0000here");
    hdfs.mkdirs(invalidXMLDir);

    // Get delegation tokens so we log the delegation token op
    Token<?>[] delegationTokens = hdfs
        .addDelegationTokens(TEST_RENEWER, null);
    for (Token<?> t : delegationTokens) {
      LOG.debug("got token " + t);
    }

    final Path snapshot = new Path("/snapshot");
    hdfs.mkdirs(snapshot);
    hdfs.allowSnapshot(snapshot);
    hdfs.mkdirs(new Path("/snapshot/1"));
    hdfs.delete(snapshot, true);

    // Set XAttrs so the fsimage contains XAttr ops
    final Path xattr = new Path("/xattr");
    hdfs.mkdirs(xattr);
    hdfs.setXAttr(xattr, "user.a1", new byte[]{ 0x31, 0x32, 0x33 });
    hdfs.setXAttr(xattr, "user.a2", new byte[]{ 0x37, 0x38, 0x39 });
    // OIV should be able to handle empty value XAttrs
    hdfs.setXAttr(xattr, "user.a3", null);
    writtenFiles.put(xattr.toString(), hdfs.getFileStatus(xattr));

    // Write results to the fsimage file
    hdfs.setSafeMode(SafeModeAction.SAFEMODE_ENTER, false);
    hdfs.saveNamespace();

    // Determine location of fsimage file
    originalFsimage = FSImageTestUtil.findLatestImageFile(FSImageTestUtil
        .getFSImage(cluster.getNameNode()).getStorage().getStorageDir(0));
    if (originalFsimage == null) {
      throw new RuntimeException("Didn't generate or can't find fsimage");
    }
    LOG.debug("original FS image file is " + originalFsimage);
  } finally {
    if (cluster != null)
      cluster.shutdown();
  }
}
 

/** Test NN crash and client crash/stuck immediately after block allocation */
@Test(timeout = 100000)
public void testOpenFileWhenNNAndClientCrashAfterAddBlock() throws Exception {
  cluster.getConfiguration(0).set(
      DFSConfigKeys.DFS_NAMENODE_SAFEMODE_THRESHOLD_PCT_KEY, "1.0f");
  String testData = "testData";
  // to make sure we write the full block before creating dummy block at NN.
  cluster.getConfiguration(0).setInt("io.bytes.per.checksum",
      testData.length());
  cluster.restartNameNode(0);
  try {
    cluster.waitActive();
    cluster.transitionToActive(0);
    cluster.transitionToStandby(1);
    DistributedFileSystem dfs = cluster.getFileSystem(0);
    String pathString = "/tmp1.txt";
    Path filePath = new Path(pathString);
    FSDataOutputStream create = dfs.create(filePath,
        FsPermission.getDefault(), true, 1024, (short) 3, testData.length(),
        null);
    create.write(testData.getBytes());
    create.hflush();
    long fileId = ((DFSOutputStream)create.
        getWrappedStream()).getFileId();
    FileStatus fileStatus = dfs.getFileStatus(filePath);
    DFSClient client = DFSClientAdapter.getClient(dfs);
    // add one dummy block at NN, but not write to DataNode
    ExtendedBlock previousBlock =
        DFSClientAdapter.getPreviousBlock(client, fileId);
    DFSClientAdapter.getNamenode(client).addBlock(
        pathString,
        client.getClientName(),
        new ExtendedBlock(previousBlock),
        new DatanodeInfo[0],
        DFSClientAdapter.getFileId((DFSOutputStream) create
            .getWrappedStream()), null);
    cluster.restartNameNode(0, true);
    cluster.restartDataNode(0);
    cluster.transitionToActive(0);
    // let the block reports be processed.
    Thread.sleep(2000);
    FSDataInputStream is = dfs.open(filePath);
    is.close();
    dfs.recoverLease(filePath);// initiate recovery
    assertTrue("Recovery also should be success", dfs.recoverLease(filePath));
  } finally {
    cluster.shutdown();
  }
}
 

/**
 * Run file operations to create edits for all op codes
 * to be tested.
 */
private void runOperations() throws IOException {

  LOG.info("Creating edits by performing fs operations");
  // no check, if it's not it throws an exception which is what we want
  DistributedFileSystem dfs =
    (DistributedFileSystem)cluster.getFileSystem();
  // OP_ADD 0, OP_SET_GENSTAMP 10
  Path pathFileCreate = new Path("/file_create");
  FSDataOutputStream s = dfs.create(pathFileCreate);
  // OP_CLOSE 9
  s.close();
  // OP_RENAME 1
  Path pathFileMoved = new Path("/file_moved");
  dfs.rename(pathFileCreate, pathFileMoved);
  // OP_DELETE 2
  dfs.delete(pathFileMoved, false);
  // OP_MKDIR 3
  Path pathDirectoryMkdir = new Path("/directory_mkdir");
  dfs.mkdirs(pathDirectoryMkdir);
  // OP_SET_REPLICATION 4
  s = dfs.create(pathFileCreate);
  s.close();
  dfs.setReplication(pathFileCreate, (short)1);
  // OP_SET_PERMISSIONS 7
  Short permission = 0777;
  dfs.setPermission(pathFileCreate, new FsPermission(permission));
  // OP_SET_OWNER 8
  dfs.setOwner(pathFileCreate, new String("newOwner"), null);
  // OP_CLOSE 9 see above
  // OP_SET_GENSTAMP 10 see above
  // OP_SET_NS_QUOTA 11 obsolete
  // OP_CLEAR_NS_QUOTA 12 obsolete
  // OP_TIMES 13
  long mtime = 1285195527000L; // Wed, 22 Sep 2010 22:45:27 GMT
  long atime = mtime;
  dfs.setTimes(pathFileCreate, mtime, atime);
  // OP_SET_QUOTA 14
  dfs.setQuota(pathDirectoryMkdir, 1000L, FSConstants.QUOTA_DONT_SET);
  // OP_CONCAT_DELETE 16
  Path   pathConcatTarget = new Path("/file_concat_target");
  Path[] pathConcatFiles  = new Path[2];
  pathConcatFiles[0]      = new Path("/file_concat_0");
  pathConcatFiles[1]      = new Path("/file_concat_1");

  long  length      = blockSize * 3; // multiple of blocksize for concat
  short replication = 1;
  long  seed        = 1;

  DFSTestUtil.createFile(dfs, pathConcatTarget, length, replication, seed);
  DFSTestUtil.createFile(dfs, pathConcatFiles[0], length, replication, seed);
  DFSTestUtil.createFile(dfs, pathConcatFiles[1], length, replication, seed);
  dfs.concat(pathConcatTarget, pathConcatFiles, false);

  // sync to disk, otherwise we parse partial edits
  cluster.getNameNode().getFSImage().getEditLog().logSync();
  dfs.close();
}
 

public void write() throws Exception {
  long endTime = System.currentTimeMillis() + rtc.max_time;
  long currentId = 0;
  FSDataOutputStream out = null;
  DistributedFileSystem dfs = (DistributedFileSystem) fs;
  while (System.currentTimeMillis() < endTime
      && currentId < rtc.max_files) {
    if (running_type == RUNNING_TYPE.PREPARE) {
      //The number of files reach the minimum limit, exit
      if (getNumberOfFiles() > rtc.min_file) 
        break;
    }
    Path fileName = new Path(outputPath, file_prefix + currentId);
    try { 
      out = dfs.create(fileName,
                       FsPermission.getDefault(),
                       false,
                       dfs.getConf().getInt("io.file.buffer.size", 4096),
                       (short)replication,
                       dfs.getDefaultBlockSize(),
                       dfs.getConf().getInt("io.bytes.per.checksum", 512),
                       null,
                       rtc.victims);
      long size = 0;
      while (true) {
        rb.nextBytes(buffer);
        tb.getTokens(rtc.buffer_size);
        out.write(buffer, 0, rtc.buffer_size);
        size += rtc.buffer_size;
        if (System.currentTimeMillis() > endTime 
            || size + rtc.buffer_size > max_size) {
          // Roll the file
          out.close();
          out = null;
          currentId++;
          files_processed++;
          processed_size += size;
          write_size += size;
          Path fullName = fs.makeQualified(fileName);
          BlockLocation bl = dfs.getClient().getBlockLocations(
              fullName.toUri().getPath(), 0L, 1L)[0];
          String hosts = "";
          for (String host: bl.getHosts()) {
            hosts += host + " ";
          }
          LOG.info("[close (" + size + "B)] " + hosts + " file " + fullName);
          break;
        }
      }
    } catch (Exception e) {
      LOG.error("Error in writing file:" + fileName, e);
      this.errors.add(e);
    } finally {
      IOUtils.closeStream(out);
    }
  }
}
 

@Test
public void testTransferRbw() throws Exception {
  final HdfsConfiguration conf = new HdfsConfiguration();
  final MiniDFSCluster cluster = new MiniDFSCluster.Builder(conf
      ).numDataNodes(REPLICATION).build();
  try {
    cluster.waitActive();
    final DistributedFileSystem fs = cluster.getFileSystem();

    //create a file, write some data and leave it open. 
    final Path p = new Path("/foo");
    final int size = (1 << 16) + RAN.nextInt(1 << 16);
    LOG.info("size = " + size);
    final FSDataOutputStream out = fs.create(p, REPLICATION);
    final byte[] bytes = new byte[1024];
    for(int remaining = size; remaining > 0; ) {
      RAN.nextBytes(bytes);
      final int len = bytes.length < remaining? bytes.length: remaining;
      out.write(bytes, 0, len);
      out.hflush();
      remaining -= len;
    }

    //get the RBW
    final ReplicaBeingWritten oldrbw;
    final DataNode newnode;
    final DatanodeInfo newnodeinfo;
    final String bpid = cluster.getNamesystem().getBlockPoolId();
    {
      final DataNode oldnode = cluster.getDataNodes().get(0);
      oldrbw = getRbw(oldnode, bpid);
      LOG.info("oldrbw = " + oldrbw);
      
      //add a datanode
      cluster.startDataNodes(conf, 1, true, null, null);
      newnode = cluster.getDataNodes().get(REPLICATION);
      
      final DatanodeInfo oldnodeinfo;
      {
        final DatanodeInfo[] datatnodeinfos = cluster.getNameNodeRpc(
            ).getDatanodeReport(DatanodeReportType.LIVE);
        Assert.assertEquals(2, datatnodeinfos.length);
        int i = 0;
        for(DatanodeRegistration dnReg = newnode.getDNRegistrationForBP(bpid);
            i < datatnodeinfos.length && !datatnodeinfos[i].equals(dnReg); i++);
        Assert.assertTrue(i < datatnodeinfos.length);
        newnodeinfo = datatnodeinfos[i];
        oldnodeinfo = datatnodeinfos[1 - i];
      }
      
      //transfer RBW
      final ExtendedBlock b = new ExtendedBlock(bpid, oldrbw.getBlockId(), oldrbw.getBytesAcked(),
          oldrbw.getGenerationStamp());
      final BlockOpResponseProto s = DFSTestUtil.transferRbw(
          b, DFSClientAdapter.getDFSClient(fs), oldnodeinfo, newnodeinfo);
      Assert.assertEquals(Status.SUCCESS, s.getStatus());
    }

    //check new rbw
    final ReplicaBeingWritten newrbw = getRbw(newnode, bpid);
    LOG.info("newrbw = " + newrbw);
    Assert.assertEquals(oldrbw.getBlockId(), newrbw.getBlockId());
    Assert.assertEquals(oldrbw.getGenerationStamp(), newrbw.getGenerationStamp());
    Assert.assertEquals(oldrbw.getVisibleLength(), newrbw.getVisibleLength());

    LOG.info("DONE");
  } finally {
    cluster.shutdown();
  }
}
 

private void testHSyncOperation(boolean testWithAppend) throws IOException {
  Configuration conf = new HdfsConfiguration();
  MiniDFSCluster cluster = new MiniDFSCluster.Builder(conf).build();
  final DistributedFileSystem fs = cluster.getFileSystem();

  final Path p = new Path("/testHSync/foo");
  final int len = 1 << 16;
  FSDataOutputStream out = fs.create(p, FsPermission.getDefault(),
      EnumSet.of(CreateFlag.CREATE, CreateFlag.OVERWRITE, CreateFlag.SYNC_BLOCK),
      4096, (short) 1, len, null);
  if (testWithAppend) {
    // re-open the file with append call
    out.close();
    out = fs.append(p, EnumSet.of(CreateFlag.APPEND, CreateFlag.SYNC_BLOCK),
        4096, null);
  }
  out.hflush();
  // hflush does not sync
  checkSyncMetric(cluster, 0);
  out.hsync();
  // hsync on empty file does nothing
  checkSyncMetric(cluster, 0);
  out.write(1);
  checkSyncMetric(cluster, 0);
  out.hsync();
  checkSyncMetric(cluster, 1);
  // avoiding repeated hsyncs is a potential future optimization
  out.hsync();
  checkSyncMetric(cluster, 2);
  out.hflush();
  // hflush still does not sync
  checkSyncMetric(cluster, 2);
  out.close();
  // close is sync'ing
  checkSyncMetric(cluster, 3);

  // same with a file created with out SYNC_BLOCK
  out = fs.create(p, FsPermission.getDefault(),
      EnumSet.of(CreateFlag.CREATE, CreateFlag.OVERWRITE),
      4096, (short) 1, len, null);
  out.hsync();
  checkSyncMetric(cluster, 3);
  out.write(1);
  checkSyncMetric(cluster, 3);
  out.hsync();
  checkSyncMetric(cluster, 4);
  // repeated hsyncs
  out.hsync();
  checkSyncMetric(cluster, 5);
  out.close();
  // close does not sync (not opened with SYNC_BLOCK)
  checkSyncMetric(cluster, 5);
  cluster.shutdown();
}
 

private void testPersistHelper(Configuration conf) throws IOException {
  MiniDFSCluster cluster = null;
  try {
    cluster = new MiniDFSCluster.Builder(conf).build();
    cluster.waitActive();
    FSNamesystem fsn = cluster.getNamesystem();
    DistributedFileSystem fs = cluster.getFileSystem();

    final Path dir = new Path("/abc/def");
    final Path file1 = new Path(dir, "f1");
    final Path file2 = new Path(dir, "f2");

    // create an empty file f1
    fs.create(file1).close();

    // create an under-construction file f2
    FSDataOutputStream out = fs.create(file2);
    out.writeBytes("hello");
    ((DFSOutputStream) out.getWrappedStream()).hsync(EnumSet
        .of(SyncFlag.UPDATE_LENGTH));

    // checkpoint
    fs.setSafeMode(SafeModeAction.SAFEMODE_ENTER);
    fs.saveNamespace();
    fs.setSafeMode(SafeModeAction.SAFEMODE_LEAVE);

    cluster.restartNameNode();
    cluster.waitActive();
    fs = cluster.getFileSystem();

    assertTrue(fs.isDirectory(dir));
    assertTrue(fs.exists(file1));
    assertTrue(fs.exists(file2));

    // check internals of file2
    INodeFile file2Node = fsn.dir.getINode4Write(file2.toString()).asFile();
    assertEquals("hello".length(), file2Node.computeFileSize());
    assertTrue(file2Node.isUnderConstruction());
    BlockInfoContiguous[] blks = file2Node.getBlocks();
    assertEquals(1, blks.length);
    assertEquals(BlockUCState.UNDER_CONSTRUCTION, blks[0].getBlockUCState());
    // check lease manager
    Lease lease = fsn.leaseManager.getLeaseByPath(file2.toString());
    Assert.assertNotNull(lease);
  } finally {
    if (cluster != null) {
      cluster.shutdown();
    }
  }
}
 

/** Test NN crash and client crash/stuck immediately after block allocation */
@Test(timeout = 100000)
public void testOpenFileWhenNNAndClientCrashAfterAddBlock() throws Exception {
  cluster.getConfiguration(0).set(
      DFSConfigKeys.DFS_NAMENODE_SAFEMODE_THRESHOLD_PCT_KEY, "1.0f");
  String testData = "testData";
  // to make sure we write the full block before creating dummy block at NN.
  cluster.getConfiguration(0).setInt("io.bytes.per.checksum",
      testData.length());
  cluster.restartNameNode(0);
  try {
    cluster.waitActive();
    cluster.transitionToActive(0);
    cluster.transitionToStandby(1);
    DistributedFileSystem dfs = cluster.getFileSystem(0);
    String pathString = "/tmp1.txt";
    Path filePath = new Path(pathString);
    FSDataOutputStream create = dfs.create(filePath,
        FsPermission.getDefault(), true, 1024, (short) 3, testData.length(),
        null);
    create.write(testData.getBytes());
    create.hflush();
    long fileId = ((DFSOutputStream)create.
        getWrappedStream()).getFileId();
    FileStatus fileStatus = dfs.getFileStatus(filePath);
    DFSClient client = DFSClientAdapter.getClient(dfs);
    // add one dummy block at NN, but not write to DataNode
    ExtendedBlock previousBlock =
        DFSClientAdapter.getPreviousBlock(client, fileId);
    DFSClientAdapter.getNamenode(client).addBlock(
        pathString,
        client.getClientName(),
        new ExtendedBlock(previousBlock),
        new DatanodeInfo[0],
        DFSClientAdapter.getFileId((DFSOutputStream) create
            .getWrappedStream()), null);
    cluster.restartNameNode(0, true);
    cluster.restartDataNode(0);
    cluster.transitionToActive(0);
    // let the block reports be processed.
    Thread.sleep(2000);
    FSDataInputStream is = dfs.open(filePath);
    is.close();
    dfs.recoverLease(filePath);// initiate recovery
    assertTrue("Recovery also should be success", dfs.recoverLease(filePath));
  } finally {
    cluster.shutdown();
  }
}