Java Code Examples for org.apache.hadoop.hdfs.server.common.Storage.StorageDirectory#getCurrentDir()

/**
 * Rename the existing current dir to previous.tmp, and create a new empty
 * current dir.
 */
public static void renameCurToTmp(StorageDirectory sd) throws IOException {
  File curDir = sd.getCurrentDir();
  File prevDir = sd.getPreviousDir();
  final File tmpDir = sd.getPreviousTmp();

  Preconditions.checkState(curDir.exists(),
      "Current directory must exist for preupgrade.");
  Preconditions.checkState(!prevDir.exists(),
      "Previous directory must not exist for preupgrade.");
  Preconditions.checkState(!tmpDir.exists(),
      "Previous.tmp directory must not exist for preupgrade."
          + "Consider restarting for recovery.");

  // rename current to tmp
  NNStorage.rename(curDir, tmpDir);

  if (!curDir.mkdir()) {
    throw new IOException("Cannot create directory " + curDir);
  }
}
 

/**
 * Perform any steps that must succeed across all storage dirs/JournalManagers
 * involved in an upgrade before proceeding onto the actual upgrade stage. If
 * a call to any JM's or local storage dir's doPreUpgrade method fails, then
 * doUpgrade will not be called for any JM. The existing current dir is
 * renamed to previous.tmp, and then a new, empty current dir is created.
 *
 * @param conf configuration for creating {@link EditLogFileOutputStream}
 * @param sd the storage directory to perform the pre-upgrade procedure.
 * @throws IOException in the event of error
 */
static void doPreUpgrade(Configuration conf, StorageDirectory sd)
    throws IOException {
  LOG.info("Starting upgrade of storage directory " + sd.getRoot());

  // rename current to tmp
  renameCurToTmp(sd);

  final File curDir = sd.getCurrentDir();
  final File tmpDir = sd.getPreviousTmp();
  List<String> fileNameList = IOUtils.listDirectory(tmpDir, new FilenameFilter() {
    @Override
    public boolean accept(File dir, String name) {
      return dir.equals(tmpDir)
          && name.startsWith(NNStorage.NameNodeFile.EDITS.getName());
    }
  });

  for (String s : fileNameList) {
    File prevFile = new File(tmpDir, s);
    File newFile = new File(curDir, prevFile.getName());
    Files.createLink(newFile.toPath(), prevFile.toPath());
  }
}
 

/**
 * Format a namespace slice storage. 
 * @param sd the namespace storage
 * @param nsInfo the name space info
 * @throws IOException Signals that an I/O exception has occurred.
 */
private void format(StorageDirectory nsSdir, NamespaceInfo nsInfo) throws IOException {
  LOG.info("Formatting namespace " + namespaceID + " directory "
      + nsSdir.getCurrentDir());
  nsSdir.clearDirectory(); // create directory
  File rbwDir = new File(nsSdir.getCurrentDir(), STORAGE_DIR_RBW);
  File finalizedDir = new File(nsSdir.getCurrentDir(), STORAGE_DIR_FINALIZED);
  LOG.info("Creating Directories : " + rbwDir + ", " + finalizedDir);
  if (!rbwDir.mkdirs() || !finalizedDir.mkdirs()) {
    throw new IOException("Cannot create directories : " + rbwDir + ", "
        + finalizedDir);
  }
  this.layoutVersion = FSConstants.LAYOUT_VERSION;
  this.cTime = nsInfo.getCTime();
  this.namespaceID = nsInfo.getNamespaceID();
  this.storageType = NodeType.DATA_NODE;
  nsSdir.write();
}
 

/**
 * Perform rollback of the storage dir to the previous state. The existing
 * current dir is removed, and the previous dir is renamed to current.
 * 
 * @param sd the storage directory to roll back.
 * @throws IOException in the event of error
 */
static void doRollBack(StorageDirectory sd)
    throws IOException {
  File prevDir = sd.getPreviousDir();
  if (!prevDir.exists()) {
    return;
  }

  File tmpDir = sd.getRemovedTmp();
  Preconditions.checkState(!tmpDir.exists(),
      "removed.tmp directory must not exist for rollback."
          + "Consider restarting for recovery.");
  // rename current to tmp
  File curDir = sd.getCurrentDir();
  Preconditions.checkState(curDir.exists(),
      "Current directory must exist for rollback.");

  NNStorage.rename(curDir, tmpDir);
  // rename previous to current
  NNStorage.rename(prevDir, curDir);

  // delete tmp dir
  NNStorage.deleteDir(tmpDir);
  LOG.info("Rollback of " + sd.getRoot() + " is complete.");
}
 

/**
 * Rename the existing current dir to previous.tmp, and create a new empty
 * current dir.
 */
public static void renameCurToTmp(StorageDirectory sd) throws IOException {
  File curDir = sd.getCurrentDir();
  File prevDir = sd.getPreviousDir();
  final File tmpDir = sd.getPreviousTmp();

  Preconditions.checkState(curDir.exists(),
      "Current directory must exist for preupgrade.");
  Preconditions.checkState(!prevDir.exists(),
      "Previous directory must not exist for preupgrade.");
  Preconditions.checkState(!tmpDir.exists(),
      "Previous.tmp directory must not exist for preupgrade."
          + "Consider restarting for recovery.");

  // rename current to tmp
  NNStorage.rename(curDir, tmpDir);

  if (!curDir.mkdir()) {
    throw new IOException("Cannot create directory " + curDir);
  }
}
 

private void assertExistsInStorageDirs(MiniDFSCluster cluster,
    NameNodeDirType dirType,
    String filename) {
  NNStorage storage = cluster.getNamesystem().getFSImage().getStorage();
  for (StorageDirectory sd : storage.dirIterable(dirType)) {
    File f = new File(sd.getCurrentDir(), filename);
    assertTrue("Expect that " + f + " exists", f.exists());
  }
}
 

private long verifyEditLogs(FSNamesystem namesystem, FSImage fsimage, 
                            String logFileName, long startTxId)
  throws IOException {
  
  long numEdits = -1;
  
  // Verify that we can read in all the transactions that we have written.
  // If there were any corruptions, it is likely that the reading in
  // of these transactions will throw an exception.
  for (StorageDirectory sd :
    fsimage.getStorage().dirIterable(NameNodeDirType.EDITS)) {

    File editFile = new File(sd.getCurrentDir(), logFileName);
      
    System.out.println("Verifying file: " + editFile);
    FSEditLogLoader loader = new FSEditLogLoader(namesystem, startTxId);
    long numEditsThisLog = loader.loadFSEdits(
        new EditLogFileInputStream(editFile), startTxId);
    
    System.out.println("Number of edits: " + numEditsThisLog);
    assertTrue(numEdits == -1 || numEditsThisLog == numEdits);
    numEdits = numEditsThisLog;
  }

  assertTrue(numEdits != -1);
  return numEdits;
}
 

/**
 * Ensure that the current/ directory exists in all storage
 * directories
 */
void ensureCurrentDirExists() throws IOException {
  for (Iterator<StorageDirectory> it
         = storage.dirIterator(); it.hasNext();) {
    StorageDirectory sd = it.next();
    File curDir = sd.getCurrentDir();
    if (!curDir.exists() && !curDir.mkdirs()) {
      throw new IOException("Could not create directory " + curDir);
    }
  }
}
 

public static void logStorageContents(Log LOG, NNStorage storage) {
  LOG.info("current storages and corresponding sizes:");
  for (StorageDirectory sd : storage.dirIterable(null)) {
    File curDir = sd.getCurrentDir();
    LOG.info("In directory " + curDir);
    File[] files = curDir.listFiles();
    Arrays.sort(files);
    for (File f : files) {
      LOG.info("  file " + f.getAbsolutePath() + "; len = " + f.length());  
    }
  }
}
 

/**
 * @return the latest edits log, finalized or otherwise, from the given
 * storage directory.
 */
public static EditLogFile findLatestEditsLog(StorageDirectory sd)
throws IOException {
  File currentDir = sd.getCurrentDir();
  List<EditLogFile> foundEditLogs 
    = Lists.newArrayList(FileJournalManager.matchEditLogs(currentDir));
  return Collections.max(foundEditLogs, EditLogFile.COMPARE_BY_START_TXID);
}
 

private void assertExistsInStorageDirs(MiniDFSCluster cluster,
    NameNodeDirType dirType,
    String filename) {
  NNStorage storage = cluster.getNamesystem().getFSImage().getStorage();
  for (StorageDirectory sd : storage.dirIterable(dirType)) {
    File f = new File(sd.getCurrentDir(), filename);
    assertTrue("Expect that " + f + " exists", f.exists());
  }
}
 

/**
 * @return the latest edits log, finalized or otherwise, from the given
 * storage directory.
 */
public static EditLogFile findLatestEditsLog(StorageDirectory sd)
throws IOException {
  File currentDir = sd.getCurrentDir();
  List<EditLogFile> foundEditLogs 
    = Lists.newArrayList(FileJournalManager.matchEditLogs(currentDir));
  return Collections.max(foundEditLogs, EditLogFile.COMPARE_BY_START_TXID);
}
 

/**
 * Ensure that the current/ directory exists in all storage
 * directories
 */
void ensureCurrentDirExists() throws IOException {
  for (Iterator<StorageDirectory> it
         = storage.dirIterator(); it.hasNext();) {
    StorageDirectory sd = it.next();
    File curDir = sd.getCurrentDir();
    if (!curDir.exists() && !curDir.mkdirs()) {
      throw new IOException("Could not create directory " + curDir);
    }
  }
}
 

public static void logStorageContents(Log LOG, NNStorage storage) {
  LOG.info("current storages and corresponding sizes:");
  for (StorageDirectory sd : storage.dirIterable(null)) {
    File curDir = sd.getCurrentDir();
    LOG.info("In directory " + curDir);
    File[] files = curDir.listFiles();
    Arrays.sort(files);
    for (File f : files) {
      LOG.info("  file " + f.getAbsolutePath() + "; len = " + f.length());  
    }
  }
}
 

/**
 * Test case where the NN is configured with a name-only and an edits-only
 * dir, with storage-restore turned on. In this case, if the name-only dir
 * disappears and comes back, a new checkpoint after it has been restored
 * should function correctly.
 * @throws Exception
 */
@Test
public void testCheckpointWithSeparateDirsAfterNameFails() throws Exception {
  MiniDFSCluster cluster = null;
  SecondaryNameNode secondary = null;
  File currentDir = null;
  
  Configuration conf = new HdfsConfiguration();

  File base_dir = new File(MiniDFSCluster.getBaseDirectory());
  conf.setBoolean(DFSConfigKeys.DFS_NAMENODE_NAME_DIR_RESTORE_KEY, true);
  conf.set(DFSConfigKeys.DFS_NAMENODE_NAME_DIR_KEY,
      MiniDFSCluster.getBaseDirectory() + "/name-only");
  conf.set(DFSConfigKeys.DFS_NAMENODE_EDITS_DIR_KEY,
      MiniDFSCluster.getBaseDirectory() + "/edits-only");
  conf.set(DFSConfigKeys.DFS_NAMENODE_CHECKPOINT_DIR_KEY,
      fileAsURI(new File(base_dir, "namesecondary1")).toString());

  try {
    cluster = new MiniDFSCluster.Builder(conf).numDataNodes(0).format(true)
        .manageNameDfsDirs(false).build();

    secondary = startSecondaryNameNode(conf);

    // Checkpoint once
    secondary.doCheckpoint();

    // Now primary NN experiences failure of its only name dir -- fake by
    // setting its current dir to a-x permissions
    NamenodeProtocols nn = cluster.getNameNodeRpc();
    NNStorage storage = cluster.getNameNode().getFSImage().getStorage();
    StorageDirectory sd0 = storage.getStorageDir(0);
    assertEquals(NameNodeDirType.IMAGE, sd0.getStorageDirType());
    currentDir = sd0.getCurrentDir();
    assertEquals(0, FileUtil.chmod(currentDir.getAbsolutePath(), "000"));

    // Try to upload checkpoint -- this should fail since there are no
    // valid storage dirs
    try {
      secondary.doCheckpoint();
      fail("Did not fail to checkpoint when there are no valid storage dirs");
    } catch (IOException ioe) {
      GenericTestUtils.assertExceptionContains(
          "No targets in destination storage", ioe);
    }
    
    // Restore the good dir
    assertEquals(0, FileUtil.chmod(currentDir.getAbsolutePath(), "755"));
    nn.restoreFailedStorage("true");
    nn.rollEditLog();

    // Checkpoint again -- this should upload to the restored name dir
    secondary.doCheckpoint();
    
    assertNNHasCheckpoints(cluster, ImmutableList.of(8));
    assertParallelFilesInvariant(cluster, ImmutableList.of(secondary));
  } finally {
    if (currentDir != null) {
      FileUtil.chmod(currentDir.getAbsolutePath(), "755");
    }
    cleanup(secondary);
    secondary = null;
    cleanup(cluster);
    cluster = null;
  }
}
 

/**
 * Test that, if a storage directory is failed when a checkpoint occurs,
 * the non-failed storage directory receives the checkpoint.
 */
@Test
public void testCheckpointWithFailedStorageDir() throws Exception {
  MiniDFSCluster cluster = null;
  SecondaryNameNode secondary = null;
  File currentDir = null;
  
  Configuration conf = new HdfsConfiguration();

  try {
    cluster = new MiniDFSCluster.Builder(conf).numDataNodes(0)
        .format(true).build();

    secondary = startSecondaryNameNode(conf);

    // Checkpoint once
    secondary.doCheckpoint();

    // Now primary NN experiences failure of a volume -- fake by
    // setting its current dir to a-x permissions
    NamenodeProtocols nn = cluster.getNameNodeRpc();
    NNStorage storage = cluster.getNameNode().getFSImage().getStorage();
    StorageDirectory sd0 = storage.getStorageDir(0);
    StorageDirectory sd1 = storage.getStorageDir(1);
    
    currentDir = sd0.getCurrentDir();
    FileUtil.setExecutable(currentDir, false);

    // Upload checkpoint when NN has a bad storage dir. This should
    // succeed and create the checkpoint in the good dir.
    secondary.doCheckpoint();
    
    GenericTestUtils.assertExists(
        new File(sd1.getCurrentDir(), NNStorage.getImageFileName(2)));
    
    // Restore the good dir
    FileUtil.setExecutable(currentDir, true);
    nn.restoreFailedStorage("true");
    nn.rollEditLog();

    // Checkpoint again -- this should upload to both dirs
    secondary.doCheckpoint();
    
    assertNNHasCheckpoints(cluster, ImmutableList.of(8));
    assertParallelFilesInvariant(cluster, ImmutableList.of(secondary));
  } finally {
    if (currentDir != null) {
      FileUtil.setExecutable(currentDir, true);
    }
    cleanup(secondary);
    secondary = null;
    cleanup(cluster);
    cluster = null;
  }
}
 

@Test(timeout=20000)
public void testCancelSaveNamespace() throws Exception {
  Configuration conf = getConf();
  NameNode.initMetrics(conf, NamenodeRole.NAMENODE);
  DFSTestUtil.formatNameNode(conf);
  FSNamesystem fsn = FSNamesystem.loadFromDisk(conf);

  // Replace the FSImage with a spy
  final FSImage image = fsn.getFSImage();
  NNStorage storage = image.getStorage();
  storage.close(); // unlock any directories that FSNamesystem's initialization may have locked
  storage.setStorageDirectories(
      FSNamesystem.getNamespaceDirs(conf), 
      FSNamesystem.getNamespaceEditsDirs(conf));

  FSNamesystem spyFsn = spy(fsn);
  final FSNamesystem finalFsn = spyFsn;
  DelayAnswer delayer = new GenericTestUtils.DelayAnswer(LOG);
  BlockIdManager bid = spy(spyFsn.getBlockIdManager());
  Whitebox.setInternalState(finalFsn, "blockIdManager", bid);
  doAnswer(delayer).when(bid).getGenerationStampV2();

  ExecutorService pool = Executors.newFixedThreadPool(2);
  
  try {
    doAnEdit(fsn, 1);
    final Canceler canceler = new Canceler();
    
    // Save namespace
    fsn.setSafeMode(SafeModeAction.SAFEMODE_ENTER);
    try {
      Future<Void> saverFuture = pool.submit(new Callable<Void>() {
        @Override
        public Void call() throws Exception {
          image.saveNamespace(finalFsn, NameNodeFile.IMAGE, canceler);
          return null;
        }
      });

      // Wait until saveNamespace calls getGenerationStamp
      delayer.waitForCall();
      // then cancel the saveNamespace
      Future<Void> cancelFuture = pool.submit(new Callable<Void>() {
        @Override
        public Void call() throws Exception {
          canceler.cancel("cancelled");
          return null;
        }
      });
      // give the cancel call time to run
      Thread.sleep(500);
      
      // allow saveNamespace to proceed - it should check the cancel flag after
      // this point and throw an exception
      delayer.proceed();

      cancelFuture.get();
      saverFuture.get();
      fail("saveNamespace did not fail even though cancelled!");
    } catch (Throwable t) {
      GenericTestUtils.assertExceptionContains(
          "SaveNamespaceCancelledException", t);
    }
    LOG.info("Successfully cancelled a saveNamespace");


    // Check that we have only the original image and not any
    // cruft left over from half-finished images
    FSImageTestUtil.logStorageContents(LOG, storage);
    for (StorageDirectory sd : storage.dirIterable(null)) {
      File curDir = sd.getCurrentDir();
      GenericTestUtils.assertGlobEquals(curDir, "fsimage_.*",
          NNStorage.getImageFileName(0),
          NNStorage.getImageFileName(0) + MD5FileUtils.MD5_SUFFIX);
    }      
  } finally {
    fsn.close();
  }
}
 

@Test(timeout=20000)
public void testCancelSaveNamespace() throws Exception {
  Configuration conf = getConf();
  NameNode.initMetrics(conf, NamenodeRole.NAMENODE);
  DFSTestUtil.formatNameNode(conf);
  FSNamesystem fsn = FSNamesystem.loadFromDisk(conf);

  // Replace the FSImage with a spy
  final FSImage image = fsn.getFSImage();
  NNStorage storage = image.getStorage();
  storage.close(); // unlock any directories that FSNamesystem's initialization may have locked
  storage.setStorageDirectories(
      FSNamesystem.getNamespaceDirs(conf), 
      FSNamesystem.getNamespaceEditsDirs(conf));

  FSNamesystem spyFsn = spy(fsn);
  final FSNamesystem finalFsn = spyFsn;
  DelayAnswer delayer = new GenericTestUtils.DelayAnswer(LOG);
  BlockIdManager bid = spy(spyFsn.getBlockIdManager());
  Whitebox.setInternalState(finalFsn, "blockIdManager", bid);
  doAnswer(delayer).when(bid).getGenerationStampV2();

  ExecutorService pool = Executors.newFixedThreadPool(2);
  
  try {
    doAnEdit(fsn, 1);
    final Canceler canceler = new Canceler();
    
    // Save namespace
    fsn.setSafeMode(SafeModeAction.SAFEMODE_ENTER);
    try {
      Future<Void> saverFuture = pool.submit(new Callable<Void>() {
        @Override
        public Void call() throws Exception {
          image.saveNamespace(finalFsn, NameNodeFile.IMAGE, canceler);
          return null;
        }
      });

      // Wait until saveNamespace calls getGenerationStamp
      delayer.waitForCall();
      // then cancel the saveNamespace
      Future<Void> cancelFuture = pool.submit(new Callable<Void>() {
        @Override
        public Void call() throws Exception {
          canceler.cancel("cancelled");
          return null;
        }
      });
      // give the cancel call time to run
      Thread.sleep(500);
      
      // allow saveNamespace to proceed - it should check the cancel flag after
      // this point and throw an exception
      delayer.proceed();

      cancelFuture.get();
      saverFuture.get();
      fail("saveNamespace did not fail even though cancelled!");
    } catch (Throwable t) {
      GenericTestUtils.assertExceptionContains(
          "SaveNamespaceCancelledException", t);
    }
    LOG.info("Successfully cancelled a saveNamespace");


    // Check that we have only the original image and not any
    // cruft left over from half-finished images
    FSImageTestUtil.logStorageContents(LOG, storage);
    for (StorageDirectory sd : storage.dirIterable(null)) {
      File curDir = sd.getCurrentDir();
      GenericTestUtils.assertGlobEquals(curDir, "fsimage_.*",
          NNStorage.getImageFileName(0),
          NNStorage.getImageFileName(0) + MD5FileUtils.MD5_SUFFIX);
    }      
  } finally {
    fsn.close();
  }
}
 

/**
 * Test case where the NN is configured with a name-only and an edits-only
 * dir, with storage-restore turned on. In this case, if the name-only dir
 * disappears and comes back, a new checkpoint after it has been restored
 * should function correctly.
 * @throws Exception
 */
@Test
public void testCheckpointWithSeparateDirsAfterNameFails() throws Exception {
  MiniDFSCluster cluster = null;
  SecondaryNameNode secondary = null;
  File currentDir = null;
  
  Configuration conf = new HdfsConfiguration();

  File base_dir = new File(MiniDFSCluster.getBaseDirectory());
  conf.setBoolean(DFSConfigKeys.DFS_NAMENODE_NAME_DIR_RESTORE_KEY, true);
  conf.set(DFSConfigKeys.DFS_NAMENODE_NAME_DIR_KEY,
      MiniDFSCluster.getBaseDirectory() + "/name-only");
  conf.set(DFSConfigKeys.DFS_NAMENODE_EDITS_DIR_KEY,
      MiniDFSCluster.getBaseDirectory() + "/edits-only");
  conf.set(DFSConfigKeys.DFS_NAMENODE_CHECKPOINT_DIR_KEY,
      fileAsURI(new File(base_dir, "namesecondary1")).toString());

  try {
    cluster = new MiniDFSCluster.Builder(conf).numDataNodes(0).format(true)
        .manageNameDfsDirs(false).build();

    secondary = startSecondaryNameNode(conf);

    // Checkpoint once
    secondary.doCheckpoint();

    // Now primary NN experiences failure of its only name dir -- fake by
    // setting its current dir to a-x permissions
    NamenodeProtocols nn = cluster.getNameNodeRpc();
    NNStorage storage = cluster.getNameNode().getFSImage().getStorage();
    StorageDirectory sd0 = storage.getStorageDir(0);
    assertEquals(NameNodeDirType.IMAGE, sd0.getStorageDirType());
    currentDir = sd0.getCurrentDir();
    assertEquals(0, FileUtil.chmod(currentDir.getAbsolutePath(), "000"));

    // Try to upload checkpoint -- this should fail since there are no
    // valid storage dirs
    try {
      secondary.doCheckpoint();
      fail("Did not fail to checkpoint when there are no valid storage dirs");
    } catch (IOException ioe) {
      GenericTestUtils.assertExceptionContains(
          "No targets in destination storage", ioe);
    }
    
    // Restore the good dir
    assertEquals(0, FileUtil.chmod(currentDir.getAbsolutePath(), "755"));
    nn.restoreFailedStorage("true");
    nn.rollEditLog();

    // Checkpoint again -- this should upload to the restored name dir
    secondary.doCheckpoint();
    
    assertNNHasCheckpoints(cluster, ImmutableList.of(8));
    assertParallelFilesInvariant(cluster, ImmutableList.of(secondary));
  } finally {
    if (currentDir != null) {
      FileUtil.chmod(currentDir.getAbsolutePath(), "755");
    }
    cleanup(secondary);
    secondary = null;
    cleanup(cluster);
    cluster = null;
  }
}
 

/**
 * Perform any steps that must succeed across all storage dirs/JournalManagers
 * involved in an upgrade before proceeding onto the actual upgrade stage. If
 * a call to any JM's or local storage dir's doPreUpgrade method fails, then
 * doUpgrade will not be called for any JM. The existing current dir is
 * renamed to previous.tmp, and then a new, empty current dir is created.
 *
 * @param conf configuration for creating {@link EditLogFileOutputStream}
 * @param sd the storage directory to perform the pre-upgrade procedure.
 * @throws IOException in the event of error
 */
static void doPreUpgrade(Configuration conf, StorageDirectory sd)
    throws IOException {
  LOG.info("Starting upgrade of storage directory " + sd.getRoot());

  // rename current to tmp
  renameCurToTmp(sd);

  final File curDir = sd.getCurrentDir();
  final File tmpDir = sd.getPreviousTmp();
  List<String> fileNameList = IOUtils.listDirectory(tmpDir, new FilenameFilter() {
    @Override
    public boolean accept(File dir, String name) {
      return dir.equals(tmpDir)
          && name.startsWith(NNStorage.NameNodeFile.EDITS.getName());
    }
  });

  for (String s : fileNameList) {
    File prevFile = new File(tmpDir, s);
    File newFile = new File(curDir, prevFile.getName());
    Files.createLink(newFile.toPath(), prevFile.toPath());
  }
}