Package org.apache.hadoop.hdfs.server.datanode

Source Code of org.apache.hadoop.hdfs.server.datanode.DataStorage

/**
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements.  See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership.  The ASF licenses this file
* to you under the Apache License, Version 2.0 (the
* "License"); you may not use this file except in compliance
* with the License.  You may obtain a copy of the License at
*
*     http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/

package org.apache.hadoop.hdfs.server.datanode;

import java.io.File;
import java.io.FileInputStream;
import java.io.FileOutputStream;
import java.io.IOException;
import java.io.RandomAccessFile;
import java.nio.channels.FileLock;
import java.util.ArrayList;
import java.util.Collection;
import java.util.Collections;
import java.util.HashMap;
import java.util.Iterator;
import java.util.Map;
import java.util.Properties;
import java.util.regex.Matcher;
import java.util.regex.Pattern;

import org.apache.hadoop.classification.InterfaceAudience;
import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.fs.FileSystem;
import org.apache.hadoop.fs.LocalFileSystem;
import org.apache.hadoop.fs.Path;
import org.apache.hadoop.fs.HardLink;
import org.apache.hadoop.fs.FileUtil;
import org.apache.hadoop.fs.permission.FsPermission;
import org.apache.hadoop.hdfs.DFSConfigKeys;
import org.apache.hadoop.hdfs.HdfsConfiguration;
import org.apache.hadoop.hdfs.protocol.HdfsConstants;
import org.apache.hadoop.hdfs.protocol.LayoutVersion;
import org.apache.hadoop.hdfs.protocol.LayoutVersion.Feature;
import org.apache.hadoop.hdfs.server.common.GenerationStamp;
import org.apache.hadoop.hdfs.server.common.InconsistentFSStateException;
import org.apache.hadoop.hdfs.server.common.Storage;
import org.apache.hadoop.hdfs.server.common.StorageInfo;
import org.apache.hadoop.hdfs.server.common.HdfsServerConstants.NodeType;
import org.apache.hadoop.hdfs.server.common.HdfsServerConstants.StartupOption;
import org.apache.hadoop.hdfs.server.protocol.NamespaceInfo;
import org.apache.hadoop.io.IOUtils;
import org.apache.hadoop.util.Daemon;
import org.apache.hadoop.util.DiskChecker;

/**
* Data storage information file.
* <p>
* @see Storage
*/
@InterfaceAudience.Private
public class DataStorage extends Storage {
  // Constants
  final static String BLOCK_SUBDIR_PREFIX = "subdir";
  final static String BLOCK_FILE_PREFIX = "blk_";
  final static String COPY_FILE_PREFIX = "dncp_";
  final static String STORAGE_DIR_DETACHED = "detach";
  public final static String STORAGE_DIR_RBW = "rbw";
  public final static String STORAGE_DIR_FINALIZED = "finalized";
  public final static String STORAGE_DIR_TMP = "tmp";

  private static final Pattern PRE_GENSTAMP_META_FILE_PATTERN =
    Pattern.compile("(.*blk_[-]*\\d+)\\.meta$");
 
  /** Access to this variable is guarded by "this" */
  private String storageID;

  // flag to ensure initialzing storage occurs only once
  private boolean initilized = false;
 
  // BlockPoolStorage is map of <Block pool Id, BlockPoolStorage>
  private Map<String, BlockPoolSliceStorage> bpStorageMap
      = Collections.synchronizedMap(new HashMap<String, BlockPoolSliceStorage>());


  DataStorage() {
    super(NodeType.DATA_NODE);
    storageID = "";
  }
 
  public StorageInfo getBPStorage(String bpid) {
    return bpStorageMap.get(bpid);
  }
 
  public DataStorage(StorageInfo storageInfo, String strgID) {
    super(NodeType.DATA_NODE, storageInfo);
    this.storageID = strgID;
  }

  synchronized String getStorageID() {
    return storageID;
  }
 
  synchronized void setStorageID(String newStorageID) {
    this.storageID = newStorageID;
  }
 
  synchronized void createStorageID(int datanodePort) {
    if (storageID != null && !storageID.isEmpty()) {
      return;
    }
    storageID = DataNode.createNewStorageId(datanodePort);
  }
 
  /**
   * Analyze storage directories.
   * Recover from previous transitions if required.
   * Perform fs state transition if necessary depending on the namespace info.
   * Read storage info.
   * <br>
   * This method should be synchronized between multiple DN threads.  Only the
   * first DN thread does DN level storage dir recoverTransitionRead.
   *
   * @param nsInfo namespace information
   * @param dataDirs array of data storage directories
   * @param startOpt startup option
   * @throws IOException
   */
  synchronized void recoverTransitionRead(DataNode datanode,
      NamespaceInfo nsInfo, Collection<File> dataDirs, StartupOption startOpt)
      throws IOException {
    if (initilized) {
      // DN storage has been initialized, no need to do anything
      return;
    }
    assert HdfsConstants.LAYOUT_VERSION == nsInfo.getLayoutVersion() :
      "Data-node version " + HdfsConstants.LAYOUT_VERSION +
      " and name-node layout version " + nsInfo.getLayoutVersion() +
      " must be the same.";
   
    // 1. For each data directory calculate its state and
    // check whether all is consistent before transitioning.
    // Format and recover.
    this.storageDirs = new ArrayList<StorageDirectory>(dataDirs.size());
    ArrayList<StorageState> dataDirStates = new ArrayList<StorageState>(dataDirs.size());
    for(Iterator<File> it = dataDirs.iterator(); it.hasNext();) {
      File dataDir = it.next();
      StorageDirectory sd = new StorageDirectory(dataDir);
      StorageState curState;
      try {
        curState = sd.analyzeStorage(startOpt, this);
        // sd is locked but not opened
        switch(curState) {
        case NORMAL:
          break;
        case NON_EXISTENT:
          // ignore this storage
          LOG.info("Storage directory " + dataDir + " does not exist.");
          it.remove();
          continue;
        case NOT_FORMATTED: // format
          LOG.info("Storage directory " + dataDir + " is not formatted.");
          LOG.info("Formatting ...");
          format(sd, nsInfo);
          break;
        default// recovery part is common
          sd.doRecover(curState);
        }
      } catch (IOException ioe) {
        sd.unlock();
        LOG.warn("Ignoring storage directory " + dataDir
            + " due to an exception", ioe);
        //continue with other good dirs
        continue;
      }
      // add to the storage list
      addStorageDir(sd);
      dataDirStates.add(curState);
    }

    if (dataDirs.size() == 0 || dataDirStates.size() == 0// none of the data dirs exist
      throw new IOException(
          "All specified directories are not accessible or do not exist.");

    // 2. Do transitions
    // Each storage directory is treated individually.
    // During startup some of them can upgrade or rollback
    // while others could be uptodate for the regular startup.
    for(int idx = 0; idx < getNumStorageDirs(); idx++) {
      doTransition(datanode, getStorageDir(idx), nsInfo, startOpt);
      assert this.getLayoutVersion() == nsInfo.getLayoutVersion() :
        "Data-node and name-node layout versions must be the same.";
    }
   
    // make sure we have storage id set - if not - generate new one
    createStorageID(datanode.getPort());
   
    // 3. Update all storages. Some of them might have just been formatted.
    this.writeAll();
   
    // 4. mark DN storage is initilized
    this.initilized = true;
  }

  /**
   * recoverTransitionRead for a specific block pool
   *
   * @param datanode DataNode
   * @param bpID Block pool Id
   * @param nsInfo Namespace info of namenode corresponding to the block pool
   * @param dataDirs Storage directories
   * @param startOpt startup option
   * @throws IOException on error
   */
  void recoverTransitionRead(DataNode datanode, String bpID, NamespaceInfo nsInfo,
      Collection<File> dataDirs, StartupOption startOpt) throws IOException {
    // First ensure datanode level format/snapshot/rollback is completed
    recoverTransitionRead(datanode, nsInfo, dataDirs, startOpt);
   
    // Create list of storage directories for the block pool
    Collection<File> bpDataDirs = new ArrayList<File>();
    for(Iterator<File> it = dataDirs.iterator(); it.hasNext();) {
      File dnRoot = it.next();
      File bpRoot = BlockPoolSliceStorage.getBpRoot(bpID, new File(dnRoot,
          STORAGE_DIR_CURRENT));
      bpDataDirs.add(bpRoot);
    }
    // mkdir for the list of BlockPoolStorage
    makeBlockPoolDataDir(bpDataDirs, null);
    BlockPoolSliceStorage bpStorage = new BlockPoolSliceStorage(
        nsInfo.getNamespaceID(), bpID, nsInfo.getCTime(), nsInfo.getClusterID());
   
    bpStorage.recoverTransitionRead(datanode, nsInfo, bpDataDirs, startOpt);
    addBlockPoolStorage(bpID, bpStorage);
  }

  /**
   * Create physical directory for block pools on the data node
   *
   * @param dataDirs
   *          List of data directories
   * @param conf
   *          Configuration instance to use.
   * @throws IOException on errors
   */
  static void makeBlockPoolDataDir(Collection<File> dataDirs,
      Configuration conf) throws IOException {
    if (conf == null)
      conf = new HdfsConfiguration();

    LocalFileSystem localFS = FileSystem.getLocal(conf);
    FsPermission permission = new FsPermission(conf.get(
        DFSConfigKeys.DFS_DATANODE_DATA_DIR_PERMISSION_KEY,
        DFSConfigKeys.DFS_DATANODE_DATA_DIR_PERMISSION_DEFAULT));
    for (File data : dataDirs) {
      try {
        DiskChecker.checkDir(localFS, new Path(data.toURI()), permission);
      } catch ( IOException e ) {
        LOG.warn("Invalid directory in: " + data.getCanonicalPath() + ": "
            + e.getMessage());
      }
    }
  }

  void format(StorageDirectory sd, NamespaceInfo nsInfo) throws IOException {
    sd.clearDirectory(); // create directory
    this.layoutVersion = HdfsConstants.LAYOUT_VERSION;
    this.clusterID = nsInfo.getClusterID();
    this.namespaceID = nsInfo.getNamespaceID();
    this.cTime = 0;
    // store storageID as it currently is
    writeProperties(sd);
  }

  /*
   * Set ClusterID, StorageID, StorageType, CTime into
   * DataStorage VERSION file
  */
  @Override
  protected void setPropertiesFromFields(Properties props,
                           StorageDirectory sd
                           ) throws IOException {
    props.setProperty("storageType", storageType.toString());
    props.setProperty("clusterID", clusterID);
    props.setProperty("cTime", String.valueOf(cTime));
    props.setProperty("layoutVersion", String.valueOf(layoutVersion));
    props.setProperty("storageID", getStorageID());
    // Set NamespaceID in version before federation
    if (!LayoutVersion.supports(Feature.FEDERATION, layoutVersion)) {
      props.setProperty("namespaceID", String.valueOf(namespaceID));
    }
  }

  /*
   * Read ClusterID, StorageID, StorageType, CTime from
   * DataStorage VERSION file and verify them.
   */
  @Override
  protected void setFieldsFromProperties(Properties props, StorageDirectory sd)
      throws IOException {
    setLayoutVersion(props, sd);
    setcTime(props, sd);
    setStorageType(props, sd);
    setClusterId(props, layoutVersion, sd);
   
    // Read NamespaceID in version before federation
    if (!LayoutVersion.supports(Feature.FEDERATION, layoutVersion)) {
      setNamespaceID(props, sd);
    }
   
    // valid storage id, storage id may be empty
    String ssid = props.getProperty("storageID");
    if (ssid == null) {
      throw new InconsistentFSStateException(sd.getRoot(), "file "
          + STORAGE_FILE_VERSION + " is invalid.");
    }
    String sid = getStorageID();
    if (!(sid.equals("") || ssid.equals("") || sid.equals(ssid))) {
      throw new InconsistentFSStateException(sd.getRoot(),
          "has incompatible storage Id.");
    }
   
    if (sid.equals("")) { // update id only if it was empty
      setStorageID(ssid);
    }
  }

  @Override
  public boolean isPreUpgradableLayout(StorageDirectory sd) throws IOException {
    File oldF = new File(sd.getRoot(), "storage");
    if (!oldF.exists())
      return false;
    // check the layout version inside the storage file
    // Lock and Read old storage file
    RandomAccessFile oldFile = new RandomAccessFile(oldF, "rws");
    FileLock oldLock = oldFile.getChannel().tryLock();
    try {
      oldFile.seek(0);
      int oldVersion = oldFile.readInt();
      if (oldVersion < LAST_PRE_UPGRADE_LAYOUT_VERSION)
        return false;
    } finally {
      oldLock.release();
      oldFile.close();
    }
    return true;
  }
 
  /**
   * Analize which and whether a transition of the fs state is required
   * and perform it if necessary.
   *
   * Rollback if previousLV >= LAYOUT_VERSION && prevCTime <= namenode.cTime
   * Upgrade if this.LV > LAYOUT_VERSION || this.cTime < namenode.cTime
   * Regular startup if this.LV = LAYOUT_VERSION && this.cTime = namenode.cTime
   *
   * @param datanode Datanode to which this storage belongs to
   * @param sd  storage directory
   * @param nsInfo  namespace info
   * @param startOpt  startup option
   * @throws IOException
   */
  private void doTransition( DataNode datanode,
                             StorageDirectory sd,
                             NamespaceInfo nsInfo,
                             StartupOption startOpt
                             ) throws IOException {
    if (startOpt == StartupOption.ROLLBACK) {
      doRollback(sd, nsInfo); // rollback if applicable
    }
    readProperties(sd);
    checkVersionUpgradable(this.layoutVersion);
    assert this.layoutVersion >= HdfsConstants.LAYOUT_VERSION :
      "Future version is not allowed";
   
    boolean federationSupported =
      LayoutVersion.supports(Feature.FEDERATION, layoutVersion);
    // For pre-federation version - validate the namespaceID
    if (!federationSupported &&
        getNamespaceID() != nsInfo.getNamespaceID()) {
      throw new IOException("Incompatible namespaceIDs in "
          + sd.getRoot().getCanonicalPath() + ": namenode namespaceID = "
          + nsInfo.getNamespaceID() + "; datanode namespaceID = "
          + getNamespaceID());
    }
   
    // For version that supports federation, validate clusterID
    if (federationSupported
        && !getClusterID().equals(nsInfo.getClusterID())) {
      throw new IOException("Incompatible clusterIDs in "
          + sd.getRoot().getCanonicalPath() + ": namenode clusterID = "
          + nsInfo.getClusterID() + "; datanode clusterID = " + getClusterID());
    }
   
    // regular start up
    if (this.layoutVersion == HdfsConstants.LAYOUT_VERSION
        && this.cTime == nsInfo.getCTime())
      return; // regular startup
    // verify necessity of a distributed upgrade
    UpgradeManagerDatanode um =
      datanode.getUpgradeManagerDatanode(nsInfo.getBlockPoolID());
    verifyDistributedUpgradeProgress(um, nsInfo);
   
    // do upgrade
    if (this.layoutVersion > HdfsConstants.LAYOUT_VERSION
        || this.cTime < nsInfo.getCTime()) {
      doUpgrade(sd, nsInfo)// upgrade
      return;
    }
   
    // layoutVersion == LAYOUT_VERSION && this.cTime > nsInfo.cTime
    // must shutdown
    throw new IOException("Datanode state: LV = " + this.getLayoutVersion()
                          + " CTime = " + this.getCTime()
                          + " is newer than the namespace state: LV = "
                          + nsInfo.getLayoutVersion()
                          + " CTime = " + nsInfo.getCTime());
  }

  /**
   * Upgrade -- Move current storage into a backup directory,
   * and hardlink all its blocks into the new current directory.
   *
   * Upgrade from pre-0.22 to 0.22 or later release e.g. 0.19/0.20/ => 0.22/0.23
   * <ul>
   * <li> If <SD>/previous exists then delete it </li>
   * <li> Rename <SD>/current to <SD>/previous.tmp </li>
   * <li>Create new <SD>/current/<bpid>/current directory<li>
   * <ul>
   * <li> Hard links for block files are created from <SD>/previous.tmp
   * to <SD>/current/<bpid>/current </li>
   * <li> Saves new version file in <SD>/current/<bpid>/current directory </li>
   * </ul>
   * <li> Rename <SD>/previous.tmp to <SD>/previous </li>
   * </ul>
   *
   * There should be only ONE namenode in the cluster for first
   * time upgrade to 0.22
   * @param sd  storage directory
   * @throws IOException on error
   */
  void doUpgrade(StorageDirectory sd, NamespaceInfo nsInfo) throws IOException {
    if (LayoutVersion.supports(Feature.FEDERATION, layoutVersion)) {
      clusterID = nsInfo.getClusterID();
      layoutVersion = nsInfo.getLayoutVersion();
      writeProperties(sd);
      return;
    }
   
    LOG.info("Upgrading storage directory " + sd.getRoot()
             + ".\n   old LV = " + this.getLayoutVersion()
             + "; old CTime = " + this.getCTime()
             + ".\n   new LV = " + nsInfo.getLayoutVersion()
             + "; new CTime = " + nsInfo.getCTime());
   
    File curDir = sd.getCurrentDir();
    File prevDir = sd.getPreviousDir();
    File bbwDir = new File(sd.getRoot(), Storage.STORAGE_1_BBW);

    assert curDir.exists() : "Data node current directory must exist.";
    // Cleanup directory "detach"
    cleanupDetachDir(new File(curDir, STORAGE_DIR_DETACHED));
   
    // 1. delete <SD>/previous dir before upgrading
    if (prevDir.exists())
      deleteDir(prevDir);
    // get previous.tmp directory, <SD>/previous.tmp
    File tmpDir = sd.getPreviousTmp();
    assert !tmpDir.exists() :
      "Data node previous.tmp directory must not exist.";
   
    // 2. Rename <SD>/current to <SD>/previous.tmp
    rename(curDir, tmpDir);
   
    // 3. Format BP and hard link blocks from previous directory
    File curBpDir = BlockPoolSliceStorage.getBpRoot(nsInfo.getBlockPoolID(), curDir);
    BlockPoolSliceStorage bpStorage = new BlockPoolSliceStorage(nsInfo.getNamespaceID(),
        nsInfo.getBlockPoolID(), nsInfo.getCTime(), nsInfo.getClusterID());
    bpStorage.format(curDir, nsInfo);
    linkAllBlocks(tmpDir, bbwDir, new File(curBpDir, STORAGE_DIR_CURRENT));
   
    // 4. Write version file under <SD>/current
    layoutVersion = HdfsConstants.LAYOUT_VERSION;
    clusterID = nsInfo.getClusterID();
    writeProperties(sd);
   
    // 5. Rename <SD>/previous.tmp to <SD>/previous
    rename(tmpDir, prevDir);
    LOG.info("Upgrade of " + sd.getRoot()+ " is complete.");
    addBlockPoolStorage(nsInfo.getBlockPoolID(), bpStorage);
  }

  /**
   * Cleanup the detachDir.
   *
   * If the directory is not empty report an error;
   * Otherwise remove the directory.
   *
   * @param detachDir detach directory
   * @throws IOException if the directory is not empty or it can not be removed
   */
  private void cleanupDetachDir(File detachDir) throws IOException {
    if (!LayoutVersion.supports(Feature.APPEND_RBW_DIR, layoutVersion) &&
        detachDir.exists() && detachDir.isDirectory() ) {
     
        if (FileUtil.list(detachDir).length != 0 ) {
          throw new IOException("Detached directory " + detachDir +
              " is not empty. Please manually move each file under this " +
              "directory to the finalized directory if the finalized " +
              "directory tree does not have the file.");
        } else if (!detachDir.delete()) {
          throw new IOException("Cannot remove directory " + detachDir);
        }
    }
  }
 
  /**
   * Rolling back to a snapshot in previous directory by moving it to current
   * directory.
   * Rollback procedure:
   * <br>
   * If previous directory exists:
   * <ol>
   * <li> Rename current to removed.tmp </li>
   * <li> Rename previous to current </li>
   * <li> Remove removed.tmp </li>
   * </ol>
   *
   * Do nothing, if previous directory does not exist.
   */
  void doRollback( StorageDirectory sd,
                   NamespaceInfo nsInfo
                   ) throws IOException {
    File prevDir = sd.getPreviousDir();
    // regular startup if previous dir does not exist
    if (!prevDir.exists())
      return;
    DataStorage prevInfo = new DataStorage();
    prevInfo.readPreviousVersionProperties(sd);

    // We allow rollback to a state, which is either consistent with
    // the namespace state or can be further upgraded to it.
    if (!(prevInfo.getLayoutVersion() >= HdfsConstants.LAYOUT_VERSION
          && prevInfo.getCTime() <= nsInfo.getCTime()))  // cannot rollback
      throw new InconsistentFSStateException(sd.getRoot(),
          "Cannot rollback to a newer state.\nDatanode previous state: LV = "
              + prevInfo.getLayoutVersion() + " CTime = " + prevInfo.getCTime()
              + " is newer than the namespace state: LV = "
              + nsInfo.getLayoutVersion() + " CTime = " + nsInfo.getCTime());
    LOG.info("Rolling back storage directory " + sd.getRoot()
             + ".\n   target LV = " + nsInfo.getLayoutVersion()
             + "; target CTime = " + nsInfo.getCTime());
    File tmpDir = sd.getRemovedTmp();
    assert !tmpDir.exists() : "removed.tmp directory must not exist.";
    // rename current to tmp
    File curDir = sd.getCurrentDir();
    assert curDir.exists() : "Current directory must exist.";
    rename(curDir, tmpDir);
    // rename previous to current
    rename(prevDir, curDir);
    // delete tmp dir
    deleteDir(tmpDir);
    LOG.info("Rollback of " + sd.getRoot() + " is complete.");
  }
 
  /**
   * Finalize procedure deletes an existing snapshot.
   * <ol>
   * <li>Rename previous to finalized.tmp directory</li>
   * <li>Fully delete the finalized.tmp directory</li>
   * </ol>
   *
   * Do nothing, if previous directory does not exist
   */
  void doFinalize(StorageDirectory sd) throws IOException {
    File prevDir = sd.getPreviousDir();
    if (!prevDir.exists())
      return; // already discarded
   
    final String dataDirPath = sd.getRoot().getCanonicalPath();
    LOG.info("Finalizing upgrade for storage directory "
             + dataDirPath
             + ".\n   cur LV = " + this.getLayoutVersion()
             + "; cur CTime = " + this.getCTime());
    assert sd.getCurrentDir().exists() : "Current directory must exist.";
    final File tmpDir = sd.getFinalizedTmp();//finalized.tmp directory
    final File bbwDir = new File(sd.getRoot(), Storage.STORAGE_1_BBW);
    // 1. rename previous to finalized.tmp
    rename(prevDir, tmpDir);

    // 2. delete finalized.tmp dir in a separate thread
    // Also delete the blocksBeingWritten from HDFS 1.x and earlier, if
    // it exists.
    new Daemon(new Runnable() {
        public void run() {
          try {
            deleteDir(tmpDir);
            if (bbwDir.exists()) {
              deleteDir(bbwDir);
            }
          } catch(IOException ex) {
            LOG.error("Finalize upgrade for " + dataDirPath + " failed.", ex);
          }
          LOG.info("Finalize upgrade for " + dataDirPath + " is complete.");
        }
        public String toString() { return "Finalize " + dataDirPath; }
      }).start();
  }
 
 
  /*
   * Finalize the upgrade for a block pool
   */
  void finalizeUpgrade(String bpID) throws IOException {
    // To handle finalizing a snapshot taken at datanode level while
    // upgrading to federation, if datanode level snapshot previous exists,
    // then finalize it. Else finalize the corresponding BP.
    for (StorageDirectory sd : storageDirs) {
      File prevDir = sd.getPreviousDir();
      if (prevDir.exists()) {
        // data node level storage finalize
        doFinalize(sd);
      } else {
        // block pool storage finalize using specific bpID
        BlockPoolSliceStorage bpStorage = bpStorageMap.get(bpID);
        bpStorage.doFinalize(sd.getCurrentDir());
      }
    }
  }

  /**
   * Hardlink all finalized and RBW blocks in fromDir to toDir
   *
   * @param fromDir      The directory where the 'from' snapshot is stored
   * @param fromBbwDir   In HDFS 1.x, the directory where blocks
   *                     that are under construction are stored.
   * @param toDir        The current data directory
   *
   * @throws IOException If error occurs during hardlink
   */
  private void linkAllBlocks(File fromDir, File fromBbwDir, File toDir)
      throws IOException {
    HardLink hardLink = new HardLink();
    // do the link
    int diskLayoutVersion = this.getLayoutVersion();
    if (LayoutVersion.supports(Feature.APPEND_RBW_DIR, diskLayoutVersion)) {
      // hardlink finalized blocks in tmpDir/finalized
      linkBlocks(new File(fromDir, STORAGE_DIR_FINALIZED),
          new File(toDir, STORAGE_DIR_FINALIZED), diskLayoutVersion, hardLink);
      // hardlink rbw blocks in tmpDir/rbw
      linkBlocks(new File(fromDir, STORAGE_DIR_RBW),
          new File(toDir, STORAGE_DIR_RBW), diskLayoutVersion, hardLink);
    } else { // pre-RBW version
      // hardlink finalized blocks in tmpDir
      linkBlocks(fromDir, new File(toDir, STORAGE_DIR_FINALIZED),
          diskLayoutVersion, hardLink);     
      if (fromBbwDir.exists()) {
        /*
         * We need to put the 'blocksBeingWritten' from HDFS 1.x into the rbw
         * directory.  It's a little messy, because the blocksBeingWriten was
         * NOT underneath the 'current' directory in those releases.  See
         * HDFS-3731 for details.
         */
        linkBlocks(fromBbwDir,
            new File(toDir, STORAGE_DIR_RBW), diskLayoutVersion, hardLink);
      }
    }
    LOG.info( hardLink.linkStats.report() );
  }
 
  static void linkBlocks(File from, File to, int oldLV, HardLink hl)
  throws IOException {
    if (!from.exists()) {
      return;
    }
    if (!from.isDirectory()) {
      if (from.getName().startsWith(COPY_FILE_PREFIX)) {
        FileInputStream in = new FileInputStream(from);
        try {
          FileOutputStream out = new FileOutputStream(to);
          try {
            IOUtils.copyBytes(in, out, 16*1024);
            hl.linkStats.countPhysicalFileCopies++;
          } finally {
            out.close();
          }
        } finally {
          in.close();
        }
      } else {
       
        //check if we are upgrading from pre-generation stamp version.
        if (oldLV >= PRE_GENERATIONSTAMP_LAYOUT_VERSION) {
          // Link to the new file name.
          to = new File(convertMetatadataFileName(to.getAbsolutePath()));
        }
       
        HardLink.createHardLink(from, to);
        hl.linkStats.countSingleLinks++;
      }
      return;
    }
    // from is a directory
    hl.linkStats.countDirs++;
   
    if (!to.mkdirs())
      throw new IOException("Cannot create directory " + to);
   
    //If upgrading from old stuff, need to munge the filenames.  That has to
    //be done one file at a time, so hardlink them one at a time (slow).
    if (oldLV >= PRE_GENERATIONSTAMP_LAYOUT_VERSION) {
      String[] blockNames = from.list(new java.io.FilenameFilter() {
          public boolean accept(File dir, String name) {
            return name.startsWith(BLOCK_SUBDIR_PREFIX)
              || name.startsWith(BLOCK_FILE_PREFIX)
              || name.startsWith(COPY_FILE_PREFIX);
          }
        });
      if (blockNames.length == 0) {
        hl.linkStats.countEmptyDirs++;
      }
      else for(int i = 0; i < blockNames.length; i++)
        linkBlocks(new File(from, blockNames[i]),
            new File(to, blockNames[i]), oldLV, hl);
    }
    else {
      //If upgrading from a relatively new version, we only need to create
      //links with the same filename.  This can be done in bulk (much faster).
      String[] blockNames = from.list(new java.io.FilenameFilter() {
        public boolean accept(File dir, String name) {
          return name.startsWith(BLOCK_FILE_PREFIX);
        }
      });
      if (blockNames.length > 0) {
        HardLink.createHardLinkMult(from, blockNames, to);
        hl.linkStats.countMultLinks++;
        hl.linkStats.countFilesMultLinks += blockNames.length;
      } else {
        hl.linkStats.countEmptyDirs++;
      }
     
      //now take care of the rest of the files and subdirectories
      String[] otherNames = from.list(new java.io.FilenameFilter() {
          public boolean accept(File dir, String name) {
            return name.startsWith(BLOCK_SUBDIR_PREFIX)
              || name.startsWith(COPY_FILE_PREFIX);
          }
        });
      for(int i = 0; i < otherNames.length; i++)
        linkBlocks(new File(from, otherNames[i]),
            new File(to, otherNames[i]), oldLV, hl);
    }
  }

  private void verifyDistributedUpgradeProgress(UpgradeManagerDatanode um,
                  NamespaceInfo nsInfo
                ) throws IOException {
    assert um != null : "DataNode.upgradeManager is null.";
    um.setUpgradeState(false, getLayoutVersion());
    um.initializeUpgrade(nsInfo);
  }
 
  /**
   * This is invoked on target file names when upgrading from pre generation
   * stamp version (version -13) to correct the metatadata file name.
   * @param oldFileName
   * @return the new metadata file name with the default generation stamp.
   */
  private static String convertMetatadataFileName(String oldFileName) {
    Matcher matcher = PRE_GENSTAMP_META_FILE_PATTERN.matcher(oldFileName);
    if (matcher.matches()) {
      //return the current metadata file name
      return DatanodeUtil.getMetaFileName(matcher.group(1),
          GenerationStamp.GRANDFATHER_GENERATION_STAMP);
    }
    return oldFileName;
  }

  /**
   * Add bpStorage into bpStorageMap
   */
  private void addBlockPoolStorage(String bpID, BlockPoolSliceStorage bpStorage)
      throws IOException {
    if (!this.bpStorageMap.containsKey(bpID)) {
      this.bpStorageMap.put(bpID, bpStorage);
    }
  }

  synchronized void removeBlockPoolStorage(String bpId) {
    bpStorageMap.remove(bpId);
  }
}
TOP

Related Classes of org.apache.hadoop.hdfs.server.datanode.DataStorage

TOP
Copyright © 2018 www.massapi.com. All rights reserved.
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.