001/**
002 *
003 * Licensed to the Apache Software Foundation (ASF) under one
004 * or more contributor license agreements.  See the NOTICE file
005 * distributed with this work for additional information
006 * regarding copyright ownership.  The ASF licenses this file
007 * to you under the Apache License, Version 2.0 (the
008 * "License"); you may not use this file except in compliance
009 * with the License.  You may obtain a copy of the License at
010 *
011 *     http://www.apache.org/licenses/LICENSE-2.0
012 *
013 * Unless required by applicable law or agreed to in writing, software
014 * distributed under the License is distributed on an "AS IS" BASIS,
015 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
016 * See the License for the specific language governing permissions and
017 * limitations under the License.
018 */
019package org.apache.hadoop.hbase.util;
020
021import org.apache.hadoop.hbase.client.RegionInfo;
022import org.apache.hbase.thirdparty.com.google.common.annotations.VisibleForTesting;
023import org.apache.hbase.thirdparty.com.google.common.base.Throwables;
024import org.apache.hbase.thirdparty.com.google.common.collect.Iterators;
025import org.apache.hbase.thirdparty.com.google.common.primitives.Ints;
026
027import edu.umd.cs.findbugs.annotations.CheckForNull;
028
029import java.io.ByteArrayInputStream;
030import java.io.DataInputStream;
031import java.io.EOFException;
032import java.io.FileNotFoundException;
033import java.io.IOException;
034import java.io.InputStream;
035import java.io.InterruptedIOException;
036import java.lang.reflect.InvocationTargetException;
037import java.lang.reflect.Method;
038import java.net.InetSocketAddress;
039import java.util.ArrayList;
040import java.util.Arrays;
041import java.util.Collections;
042import java.util.HashMap;
043import java.util.Iterator;
044import java.util.LinkedList;
045import java.util.List;
046import java.util.Locale;
047import java.util.Map;
048import java.util.Vector;
049import java.util.concurrent.ArrayBlockingQueue;
050import java.util.concurrent.ConcurrentHashMap;
051import java.util.concurrent.ExecutionException;
052import java.util.concurrent.ExecutorService;
053import java.util.concurrent.Future;
054import java.util.concurrent.FutureTask;
055import java.util.concurrent.ThreadPoolExecutor;
056import java.util.concurrent.TimeUnit;
057import java.util.regex.Pattern;
058
059import org.apache.hadoop.conf.Configuration;
060import org.apache.hadoop.fs.BlockLocation;
061import org.apache.hadoop.fs.FSDataInputStream;
062import org.apache.hadoop.fs.FSDataOutputStream;
063import org.apache.hadoop.fs.FileStatus;
064import org.apache.hadoop.fs.FileSystem;
065import org.apache.hadoop.fs.Path;
066import org.apache.hadoop.fs.PathFilter;
067import org.apache.hadoop.fs.permission.FsAction;
068import org.apache.hadoop.fs.permission.FsPermission;
069import org.apache.hadoop.hbase.ClusterId;
070import org.apache.hadoop.hbase.HColumnDescriptor;
071import org.apache.hadoop.hbase.HConstants;
072import org.apache.hadoop.hbase.HDFSBlocksDistribution;
073import org.apache.hadoop.hbase.HRegionInfo;
074import org.apache.hadoop.hbase.TableName;
075import org.apache.yetus.audience.InterfaceAudience;
076import org.slf4j.Logger;
077import org.slf4j.LoggerFactory;
078import org.apache.hadoop.hbase.exceptions.DeserializationException;
079import org.apache.hadoop.hbase.fs.HFileSystem;
080import org.apache.hadoop.hbase.io.HFileLink;
081import org.apache.hadoop.hbase.master.HMaster;
082import org.apache.hadoop.hbase.regionserver.HRegion;
083import org.apache.hadoop.hbase.regionserver.StoreFileInfo;
084import org.apache.hadoop.hbase.security.AccessDeniedException;
085import org.apache.hadoop.hbase.shaded.protobuf.ProtobufUtil;
086import org.apache.hadoop.hbase.shaded.protobuf.generated.FSProtos;
087import org.apache.hadoop.hbase.util.HBaseFsck.ErrorReporter;
088import org.apache.hadoop.hdfs.DFSClient;
089import org.apache.hadoop.hdfs.DFSHedgedReadMetrics;
090import org.apache.hadoop.hdfs.DistributedFileSystem;
091import org.apache.hadoop.hdfs.protocol.HdfsConstants;
092import org.apache.hadoop.io.IOUtils;
093import org.apache.hadoop.ipc.RemoteException;
094import org.apache.hadoop.security.UserGroupInformation;
095import org.apache.hadoop.util.Progressable;
096import org.apache.hadoop.util.ReflectionUtils;
097import org.apache.hadoop.util.StringUtils;
098
099/**
100 * Utility methods for interacting with the underlying file system.
101 */
102@InterfaceAudience.Private
103public abstract class FSUtils extends CommonFSUtils {
104  private static final Logger LOG = LoggerFactory.getLogger(FSUtils.class);
105
106  private static final String THREAD_POOLSIZE = "hbase.client.localityCheck.threadPoolSize";
107  private static final int DEFAULT_THREAD_POOLSIZE = 2;
108
109  /** Set to true on Windows platforms */
110  @VisibleForTesting // currently only used in testing. TODO refactor into a test class
111  public static final boolean WINDOWS = System.getProperty("os.name").startsWith("Windows");
112
113  protected FSUtils() {
114    super();
115  }
116
117  /**
118   * @return True is <code>fs</code> is instance of DistributedFileSystem
119   * @throws IOException
120   */
121  public static boolean isDistributedFileSystem(final FileSystem fs) throws IOException {
122    FileSystem fileSystem = fs;
123    // If passed an instance of HFileSystem, it fails instanceof DistributedFileSystem.
124    // Check its backing fs for dfs-ness.
125    if (fs instanceof HFileSystem) {
126      fileSystem = ((HFileSystem)fs).getBackingFs();
127    }
128    return fileSystem instanceof DistributedFileSystem;
129  }
130
131  /**
132   * Compare path component of the Path URI; e.g. if hdfs://a/b/c and /a/b/c, it will compare the
133   * '/a/b/c' part. If you passed in 'hdfs://a/b/c and b/c, it would return true.  Does not consider
134   * schema; i.e. if schemas different but path or subpath matches, the two will equate.
135   * @param pathToSearch Path we will be trying to match.
136   * @param pathTail
137   * @return True if <code>pathTail</code> is tail on the path of <code>pathToSearch</code>
138   */
139  public static boolean isMatchingTail(final Path pathToSearch, final Path pathTail) {
140    if (pathToSearch.depth() != pathTail.depth()) return false;
141    Path tailPath = pathTail;
142    String tailName;
143    Path toSearch = pathToSearch;
144    String toSearchName;
145    boolean result = false;
146    do {
147      tailName = tailPath.getName();
148      if (tailName == null || tailName.length() <= 0) {
149        result = true;
150        break;
151      }
152      toSearchName = toSearch.getName();
153      if (toSearchName == null || toSearchName.length() <= 0) break;
154      // Move up a parent on each path for next go around.  Path doesn't let us go off the end.
155      tailPath = tailPath.getParent();
156      toSearch = toSearch.getParent();
157    } while(tailName.equals(toSearchName));
158    return result;
159  }
160
161  public static FSUtils getInstance(FileSystem fs, Configuration conf) {
162    String scheme = fs.getUri().getScheme();
163    if (scheme == null) {
164      LOG.warn("Could not find scheme for uri " +
165          fs.getUri() + ", default to hdfs");
166      scheme = "hdfs";
167    }
168    Class<?> fsUtilsClass = conf.getClass("hbase.fsutil." +
169        scheme + ".impl", FSHDFSUtils.class); // Default to HDFS impl
170    FSUtils fsUtils = (FSUtils)ReflectionUtils.newInstance(fsUtilsClass, conf);
171    return fsUtils;
172  }
173
174  /**
175   * Delete the region directory if exists.
176   * @param conf
177   * @param hri
178   * @return True if deleted the region directory.
179   * @throws IOException
180   */
181  public static boolean deleteRegionDir(final Configuration conf, final HRegionInfo hri)
182  throws IOException {
183    Path rootDir = getRootDir(conf);
184    FileSystem fs = rootDir.getFileSystem(conf);
185    return deleteDirectory(fs,
186      new Path(getTableDir(rootDir, hri.getTable()), hri.getEncodedName()));
187  }
188
189 /**
190   * Create the specified file on the filesystem. By default, this will:
191   * <ol>
192   * <li>overwrite the file if it exists</li>
193   * <li>apply the umask in the configuration (if it is enabled)</li>
194   * <li>use the fs configured buffer size (or 4096 if not set)</li>
195   * <li>use the configured column family replication or default replication if
196   * {@link HColumnDescriptor#DEFAULT_DFS_REPLICATION}</li>
197   * <li>use the default block size</li>
198   * <li>not track progress</li>
199   * </ol>
200   * @param conf configurations
201   * @param fs {@link FileSystem} on which to write the file
202   * @param path {@link Path} to the file to write
203   * @param perm permissions
204   * @param favoredNodes
205   * @return output stream to the created file
206   * @throws IOException if the file cannot be created
207   */
208  public static FSDataOutputStream create(Configuration conf, FileSystem fs, Path path,
209      FsPermission perm, InetSocketAddress[] favoredNodes) throws IOException {
210    if (fs instanceof HFileSystem) {
211      FileSystem backingFs = ((HFileSystem)fs).getBackingFs();
212      if (backingFs instanceof DistributedFileSystem) {
213        // Try to use the favoredNodes version via reflection to allow backwards-
214        // compatibility.
215        short replication = Short.parseShort(conf.get(HColumnDescriptor.DFS_REPLICATION,
216          String.valueOf(HColumnDescriptor.DEFAULT_DFS_REPLICATION)));
217        try {
218          return (FSDataOutputStream) (DistributedFileSystem.class.getDeclaredMethod("create",
219            Path.class, FsPermission.class, boolean.class, int.class, short.class, long.class,
220            Progressable.class, InetSocketAddress[].class).invoke(backingFs, path, perm, true,
221            getDefaultBufferSize(backingFs),
222            replication > 0 ? replication : getDefaultReplication(backingFs, path),
223            getDefaultBlockSize(backingFs, path), null, favoredNodes));
224        } catch (InvocationTargetException ite) {
225          // Function was properly called, but threw it's own exception.
226          throw new IOException(ite.getCause());
227        } catch (NoSuchMethodException e) {
228          LOG.debug("DFS Client does not support most favored nodes create; using default create");
229          if (LOG.isTraceEnabled()) LOG.trace("Ignoring; use default create", e);
230        } catch (IllegalArgumentException e) {
231          LOG.debug("Ignoring (most likely Reflection related exception) " + e);
232        } catch (SecurityException e) {
233          LOG.debug("Ignoring (most likely Reflection related exception) " + e);
234        } catch (IllegalAccessException e) {
235          LOG.debug("Ignoring (most likely Reflection related exception) " + e);
236        }
237      }
238    }
239    return create(fs, path, perm, true);
240  }
241
242  /**
243   * Checks to see if the specified file system is available
244   *
245   * @param fs filesystem
246   * @throws IOException e
247   */
248  public static void checkFileSystemAvailable(final FileSystem fs)
249  throws IOException {
250    if (!(fs instanceof DistributedFileSystem)) {
251      return;
252    }
253    IOException exception = null;
254    DistributedFileSystem dfs = (DistributedFileSystem) fs;
255    try {
256      if (dfs.exists(new Path("/"))) {
257        return;
258      }
259    } catch (IOException e) {
260      exception = e instanceof RemoteException ?
261              ((RemoteException)e).unwrapRemoteException() : e;
262    }
263    try {
264      fs.close();
265    } catch (Exception e) {
266      LOG.error("file system close failed: ", e);
267    }
268    IOException io = new IOException("File system is not available");
269    io.initCause(exception);
270    throw io;
271  }
272
273  /**
274   * We use reflection because {@link DistributedFileSystem#setSafeMode(
275   * HdfsConstants.SafeModeAction action, boolean isChecked)} is not in hadoop 1.1
276   *
277   * @param dfs
278   * @return whether we're in safe mode
279   * @throws IOException
280   */
281  private static boolean isInSafeMode(DistributedFileSystem dfs) throws IOException {
282    boolean inSafeMode = false;
283    try {
284      Method m = DistributedFileSystem.class.getMethod("setSafeMode", new Class<?> []{
285          org.apache.hadoop.hdfs.protocol.HdfsConstants.SafeModeAction.class, boolean.class});
286      inSafeMode = (Boolean) m.invoke(dfs,
287        org.apache.hadoop.hdfs.protocol.HdfsConstants.SafeModeAction.SAFEMODE_GET, true);
288    } catch (Exception e) {
289      if (e instanceof IOException) throw (IOException) e;
290
291      // Check whether dfs is on safemode.
292      inSafeMode = dfs.setSafeMode(
293        org.apache.hadoop.hdfs.protocol.HdfsConstants.SafeModeAction.SAFEMODE_GET);
294    }
295    return inSafeMode;
296  }
297
298  /**
299   * Check whether dfs is in safemode.
300   * @param conf
301   * @throws IOException
302   */
303  public static void checkDfsSafeMode(final Configuration conf)
304  throws IOException {
305    boolean isInSafeMode = false;
306    FileSystem fs = FileSystem.get(conf);
307    if (fs instanceof DistributedFileSystem) {
308      DistributedFileSystem dfs = (DistributedFileSystem)fs;
309      isInSafeMode = isInSafeMode(dfs);
310    }
311    if (isInSafeMode) {
312      throw new IOException("File system is in safemode, it can't be written now");
313    }
314  }
315
316  /**
317   * Verifies current version of file system
318   *
319   * @param fs filesystem object
320   * @param rootdir root hbase directory
321   * @return null if no version file exists, version string otherwise.
322   * @throws IOException e
323   * @throws org.apache.hadoop.hbase.exceptions.DeserializationException
324   */
325  public static String getVersion(FileSystem fs, Path rootdir)
326  throws IOException, DeserializationException {
327    Path versionFile = new Path(rootdir, HConstants.VERSION_FILE_NAME);
328    FileStatus[] status = null;
329    try {
330      // hadoop 2.0 throws FNFE if directory does not exist.
331      // hadoop 1.0 returns null if directory does not exist.
332      status = fs.listStatus(versionFile);
333    } catch (FileNotFoundException fnfe) {
334      return null;
335    }
336    if (status == null || status.length == 0) return null;
337    String version = null;
338    byte [] content = new byte [(int)status[0].getLen()];
339    FSDataInputStream s = fs.open(versionFile);
340    try {
341      IOUtils.readFully(s, content, 0, content.length);
342      if (ProtobufUtil.isPBMagicPrefix(content)) {
343        version = parseVersionFrom(content);
344      } else {
345        // Presume it pre-pb format.
346        InputStream is = new ByteArrayInputStream(content);
347        DataInputStream dis = new DataInputStream(is);
348        try {
349          version = dis.readUTF();
350        } finally {
351          dis.close();
352        }
353      }
354    } catch (EOFException eof) {
355      LOG.warn("Version file was empty, odd, will try to set it.");
356    } finally {
357      s.close();
358    }
359    return version;
360  }
361
362  /**
363   * Parse the content of the ${HBASE_ROOTDIR}/hbase.version file.
364   * @param bytes The byte content of the hbase.version file.
365   * @return The version found in the file as a String.
366   * @throws DeserializationException
367   */
368  static String parseVersionFrom(final byte [] bytes)
369  throws DeserializationException {
370    ProtobufUtil.expectPBMagicPrefix(bytes);
371    int pblen = ProtobufUtil.lengthOfPBMagic();
372    FSProtos.HBaseVersionFileContent.Builder builder =
373      FSProtos.HBaseVersionFileContent.newBuilder();
374    try {
375      ProtobufUtil.mergeFrom(builder, bytes, pblen, bytes.length - pblen);
376      return builder.getVersion();
377    } catch (IOException e) {
378      // Convert
379      throw new DeserializationException(e);
380    }
381  }
382
383  /**
384   * Create the content to write into the ${HBASE_ROOTDIR}/hbase.version file.
385   * @param version Version to persist
386   * @return Serialized protobuf with <code>version</code> content and a bit of pb magic for a prefix.
387   */
388  static byte [] toVersionByteArray(final String version) {
389    FSProtos.HBaseVersionFileContent.Builder builder =
390      FSProtos.HBaseVersionFileContent.newBuilder();
391    return ProtobufUtil.prependPBMagic(builder.setVersion(version).build().toByteArray());
392  }
393
394  /**
395   * Verifies current version of file system
396   *
397   * @param fs file system
398   * @param rootdir root directory of HBase installation
399   * @param message if true, issues a message on System.out
400   *
401   * @throws IOException e
402   * @throws DeserializationException
403   */
404  public static void checkVersion(FileSystem fs, Path rootdir, boolean message)
405  throws IOException, DeserializationException {
406    checkVersion(fs, rootdir, message, 0, HConstants.DEFAULT_VERSION_FILE_WRITE_ATTEMPTS);
407  }
408
409  /**
410   * Verifies current version of file system
411   *
412   * @param fs file system
413   * @param rootdir root directory of HBase installation
414   * @param message if true, issues a message on System.out
415   * @param wait wait interval
416   * @param retries number of times to retry
417   *
418   * @throws IOException e
419   * @throws DeserializationException
420   */
421  public static void checkVersion(FileSystem fs, Path rootdir,
422      boolean message, int wait, int retries)
423  throws IOException, DeserializationException {
424    String version = getVersion(fs, rootdir);
425    if (version == null) {
426      if (!metaRegionExists(fs, rootdir)) {
427        // rootDir is empty (no version file and no root region)
428        // just create new version file (HBASE-1195)
429        setVersion(fs, rootdir, wait, retries);
430        return;
431      }
432    } else if (version.compareTo(HConstants.FILE_SYSTEM_VERSION) == 0) return;
433
434    // version is deprecated require migration
435    // Output on stdout so user sees it in terminal.
436    String msg = "HBase file layout needs to be upgraded."
437      + " You have version " + version
438      + " and I want version " + HConstants.FILE_SYSTEM_VERSION
439      + ". Consult http://hbase.apache.org/book.html for further information about upgrading HBase."
440      + " Is your hbase.rootdir valid? If so, you may need to run "
441      + "'hbase hbck -fixVersionFile'.";
442    if (message) {
443      System.out.println("WARNING! " + msg);
444    }
445    throw new FileSystemVersionException(msg);
446  }
447
448  /**
449   * Sets version of file system
450   *
451   * @param fs filesystem object
452   * @param rootdir hbase root
453   * @throws IOException e
454   */
455  public static void setVersion(FileSystem fs, Path rootdir)
456  throws IOException {
457    setVersion(fs, rootdir, HConstants.FILE_SYSTEM_VERSION, 0,
458      HConstants.DEFAULT_VERSION_FILE_WRITE_ATTEMPTS);
459  }
460
461  /**
462   * Sets version of file system
463   *
464   * @param fs filesystem object
465   * @param rootdir hbase root
466   * @param wait time to wait for retry
467   * @param retries number of times to retry before failing
468   * @throws IOException e
469   */
470  public static void setVersion(FileSystem fs, Path rootdir, int wait, int retries)
471  throws IOException {
472    setVersion(fs, rootdir, HConstants.FILE_SYSTEM_VERSION, wait, retries);
473  }
474
475
476  /**
477   * Sets version of file system
478   *
479   * @param fs filesystem object
480   * @param rootdir hbase root directory
481   * @param version version to set
482   * @param wait time to wait for retry
483   * @param retries number of times to retry before throwing an IOException
484   * @throws IOException e
485   */
486  public static void setVersion(FileSystem fs, Path rootdir, String version,
487      int wait, int retries) throws IOException {
488    Path versionFile = new Path(rootdir, HConstants.VERSION_FILE_NAME);
489    Path tempVersionFile = new Path(rootdir, HConstants.HBASE_TEMP_DIRECTORY + Path.SEPARATOR +
490      HConstants.VERSION_FILE_NAME);
491    while (true) {
492      try {
493        // Write the version to a temporary file
494        FSDataOutputStream s = fs.create(tempVersionFile);
495        try {
496          s.write(toVersionByteArray(version));
497          s.close();
498          s = null;
499          // Move the temp version file to its normal location. Returns false
500          // if the rename failed. Throw an IOE in that case.
501          if (!fs.rename(tempVersionFile, versionFile)) {
502            throw new IOException("Unable to move temp version file to " + versionFile);
503          }
504        } finally {
505          // Cleaning up the temporary if the rename failed would be trying
506          // too hard. We'll unconditionally create it again the next time
507          // through anyway, files are overwritten by default by create().
508
509          // Attempt to close the stream on the way out if it is still open.
510          try {
511            if (s != null) s.close();
512          } catch (IOException ignore) { }
513        }
514        LOG.info("Created version file at " + rootdir.toString() + " with version=" + version);
515        return;
516      } catch (IOException e) {
517        if (retries > 0) {
518          LOG.debug("Unable to create version file at " + rootdir.toString() + ", retrying", e);
519          fs.delete(versionFile, false);
520          try {
521            if (wait > 0) {
522              Thread.sleep(wait);
523            }
524          } catch (InterruptedException ie) {
525            throw (InterruptedIOException)new InterruptedIOException().initCause(ie);
526          }
527          retries--;
528        } else {
529          throw e;
530        }
531      }
532    }
533  }
534
535  /**
536   * Checks that a cluster ID file exists in the HBase root directory
537   * @param fs the root directory FileSystem
538   * @param rootdir the HBase root directory in HDFS
539   * @param wait how long to wait between retries
540   * @return <code>true</code> if the file exists, otherwise <code>false</code>
541   * @throws IOException if checking the FileSystem fails
542   */
543  public static boolean checkClusterIdExists(FileSystem fs, Path rootdir,
544      int wait) throws IOException {
545    while (true) {
546      try {
547        Path filePath = new Path(rootdir, HConstants.CLUSTER_ID_FILE_NAME);
548        return fs.exists(filePath);
549      } catch (IOException ioe) {
550        if (wait > 0) {
551          LOG.warn("Unable to check cluster ID file in " + rootdir.toString() +
552              ", retrying in "+wait+"msec: "+StringUtils.stringifyException(ioe));
553          try {
554            Thread.sleep(wait);
555          } catch (InterruptedException e) {
556            throw (InterruptedIOException)new InterruptedIOException().initCause(e);
557          }
558        } else {
559          throw ioe;
560        }
561      }
562    }
563  }
564
565  /**
566   * Returns the value of the unique cluster ID stored for this HBase instance.
567   * @param fs the root directory FileSystem
568   * @param rootdir the path to the HBase root directory
569   * @return the unique cluster identifier
570   * @throws IOException if reading the cluster ID file fails
571   */
572  public static ClusterId getClusterId(FileSystem fs, Path rootdir)
573  throws IOException {
574    Path idPath = new Path(rootdir, HConstants.CLUSTER_ID_FILE_NAME);
575    ClusterId clusterId = null;
576    FileStatus status = fs.exists(idPath)? fs.getFileStatus(idPath):  null;
577    if (status != null) {
578      int len = Ints.checkedCast(status.getLen());
579      byte [] content = new byte[len];
580      FSDataInputStream in = fs.open(idPath);
581      try {
582        in.readFully(content);
583      } catch (EOFException eof) {
584        LOG.warn("Cluster ID file " + idPath.toString() + " was empty");
585      } finally{
586        in.close();
587      }
588      try {
589        clusterId = ClusterId.parseFrom(content);
590      } catch (DeserializationException e) {
591        throw new IOException("content=" + Bytes.toString(content), e);
592      }
593      // If not pb'd, make it so.
594      if (!ProtobufUtil.isPBMagicPrefix(content)) {
595        String cid = null;
596        in = fs.open(idPath);
597        try {
598          cid = in.readUTF();
599          clusterId = new ClusterId(cid);
600        } catch (EOFException eof) {
601          LOG.warn("Cluster ID file " + idPath.toString() + " was empty");
602        } finally {
603          in.close();
604        }
605        rewriteAsPb(fs, rootdir, idPath, clusterId);
606      }
607      return clusterId;
608    } else {
609      LOG.warn("Cluster ID file does not exist at " + idPath.toString());
610    }
611    return clusterId;
612  }
613
614  /**
615   * @param cid
616   * @throws IOException
617   */
618  private static void rewriteAsPb(final FileSystem fs, final Path rootdir, final Path p,
619      final ClusterId cid)
620  throws IOException {
621    // Rewrite the file as pb.  Move aside the old one first, write new
622    // then delete the moved-aside file.
623    Path movedAsideName = new Path(p + "." + System.currentTimeMillis());
624    if (!fs.rename(p, movedAsideName)) throw new IOException("Failed rename of " + p);
625    setClusterId(fs, rootdir, cid, 100);
626    if (!fs.delete(movedAsideName, false)) {
627      throw new IOException("Failed delete of " + movedAsideName);
628    }
629    LOG.debug("Rewrote the hbase.id file as pb");
630  }
631
632  /**
633   * Writes a new unique identifier for this cluster to the "hbase.id" file
634   * in the HBase root directory
635   * @param fs the root directory FileSystem
636   * @param rootdir the path to the HBase root directory
637   * @param clusterId the unique identifier to store
638   * @param wait how long (in milliseconds) to wait between retries
639   * @throws IOException if writing to the FileSystem fails and no wait value
640   */
641  public static void setClusterId(FileSystem fs, Path rootdir, ClusterId clusterId,
642      int wait) throws IOException {
643    while (true) {
644      try {
645        Path idFile = new Path(rootdir, HConstants.CLUSTER_ID_FILE_NAME);
646        Path tempIdFile = new Path(rootdir, HConstants.HBASE_TEMP_DIRECTORY +
647          Path.SEPARATOR + HConstants.CLUSTER_ID_FILE_NAME);
648        // Write the id file to a temporary location
649        FSDataOutputStream s = fs.create(tempIdFile);
650        try {
651          s.write(clusterId.toByteArray());
652          s.close();
653          s = null;
654          // Move the temporary file to its normal location. Throw an IOE if
655          // the rename failed
656          if (!fs.rename(tempIdFile, idFile)) {
657            throw new IOException("Unable to move temp version file to " + idFile);
658          }
659        } finally {
660          // Attempt to close the stream if still open on the way out
661          try {
662            if (s != null) s.close();
663          } catch (IOException ignore) { }
664        }
665        if (LOG.isDebugEnabled()) {
666          LOG.debug("Created cluster ID file at " + idFile.toString() + " with ID: " + clusterId);
667        }
668        return;
669      } catch (IOException ioe) {
670        if (wait > 0) {
671          LOG.warn("Unable to create cluster ID file in " + rootdir.toString() +
672              ", retrying in " + wait + "msec: " + StringUtils.stringifyException(ioe));
673          try {
674            Thread.sleep(wait);
675          } catch (InterruptedException e) {
676            throw (InterruptedIOException)new InterruptedIOException().initCause(e);
677          }
678        } else {
679          throw ioe;
680        }
681      }
682    }
683  }
684
685  /**
686   * If DFS, check safe mode and if so, wait until we clear it.
687   * @param conf configuration
688   * @param wait Sleep between retries
689   * @throws IOException e
690   */
691  public static void waitOnSafeMode(final Configuration conf,
692    final long wait)
693  throws IOException {
694    FileSystem fs = FileSystem.get(conf);
695    if (!(fs instanceof DistributedFileSystem)) return;
696    DistributedFileSystem dfs = (DistributedFileSystem)fs;
697    // Make sure dfs is not in safe mode
698    while (isInSafeMode(dfs)) {
699      LOG.info("Waiting for dfs to exit safe mode...");
700      try {
701        Thread.sleep(wait);
702      } catch (InterruptedException e) {
703        throw (InterruptedIOException)new InterruptedIOException().initCause(e);
704      }
705    }
706  }
707
708  /**
709   * Checks if meta region exists
710   *
711   * @param fs file system
712   * @param rootdir root directory of HBase installation
713   * @return true if exists
714   * @throws IOException e
715   */
716  @SuppressWarnings("deprecation")
717  public static boolean metaRegionExists(FileSystem fs, Path rootdir)
718  throws IOException {
719    Path metaRegionDir =
720      HRegion.getRegionDir(rootdir, HRegionInfo.FIRST_META_REGIONINFO);
721    return fs.exists(metaRegionDir);
722  }
723
724  /**
725   * Compute HDFS blocks distribution of a given file, or a portion of the file
726   * @param fs file system
727   * @param status file status of the file
728   * @param start start position of the portion
729   * @param length length of the portion
730   * @return The HDFS blocks distribution
731   */
732  static public HDFSBlocksDistribution computeHDFSBlocksDistribution(
733    final FileSystem fs, FileStatus status, long start, long length)
734    throws IOException {
735    HDFSBlocksDistribution blocksDistribution = new HDFSBlocksDistribution();
736    BlockLocation [] blockLocations =
737      fs.getFileBlockLocations(status, start, length);
738    for(BlockLocation bl : blockLocations) {
739      String [] hosts = bl.getHosts();
740      long len = bl.getLength();
741      blocksDistribution.addHostsAndBlockWeight(hosts, len);
742    }
743
744    return blocksDistribution;
745  }
746
747  /**
748   * Update blocksDistribution with blockLocations
749   * @param blocksDistribution the hdfs blocks distribution
750   * @param blockLocations an array containing block location
751   */
752  static public void addToHDFSBlocksDistribution(
753      HDFSBlocksDistribution blocksDistribution, BlockLocation[] blockLocations)
754      throws IOException {
755    for (BlockLocation bl : blockLocations) {
756      String[] hosts = bl.getHosts();
757      long len = bl.getLength();
758      blocksDistribution.addHostsAndBlockWeight(hosts, len);
759    }
760  }
761
762  // TODO move this method OUT of FSUtils. No dependencies to HMaster
763  /**
764   * Returns the total overall fragmentation percentage. Includes hbase:meta and
765   * -ROOT- as well.
766   *
767   * @param master  The master defining the HBase root and file system.
768   * @return A map for each table and its percentage.
769   * @throws IOException When scanning the directory fails.
770   */
771  public static int getTotalTableFragmentation(final HMaster master)
772  throws IOException {
773    Map<String, Integer> map = getTableFragmentation(master);
774    return map != null && map.size() > 0 ? map.get("-TOTAL-") : -1;
775  }
776
777  /**
778   * Runs through the HBase rootdir and checks how many stores for each table
779   * have more than one file in them. Checks -ROOT- and hbase:meta too. The total
780   * percentage across all tables is stored under the special key "-TOTAL-".
781   *
782   * @param master  The master defining the HBase root and file system.
783   * @return A map for each table and its percentage.
784   *
785   * @throws IOException When scanning the directory fails.
786   */
787  public static Map<String, Integer> getTableFragmentation(
788    final HMaster master)
789  throws IOException {
790    Path path = getRootDir(master.getConfiguration());
791    // since HMaster.getFileSystem() is package private
792    FileSystem fs = path.getFileSystem(master.getConfiguration());
793    return getTableFragmentation(fs, path);
794  }
795
796  /**
797   * Runs through the HBase rootdir and checks how many stores for each table
798   * have more than one file in them. Checks -ROOT- and hbase:meta too. The total
799   * percentage across all tables is stored under the special key "-TOTAL-".
800   *
801   * @param fs  The file system to use.
802   * @param hbaseRootDir  The root directory to scan.
803   * @return A map for each table and its percentage.
804   * @throws IOException When scanning the directory fails.
805   */
806  public static Map<String, Integer> getTableFragmentation(
807    final FileSystem fs, final Path hbaseRootDir)
808  throws IOException {
809    Map<String, Integer> frags = new HashMap<>();
810    int cfCountTotal = 0;
811    int cfFragTotal = 0;
812    PathFilter regionFilter = new RegionDirFilter(fs);
813    PathFilter familyFilter = new FamilyDirFilter(fs);
814    List<Path> tableDirs = getTableDirs(fs, hbaseRootDir);
815    for (Path d : tableDirs) {
816      int cfCount = 0;
817      int cfFrag = 0;
818      FileStatus[] regionDirs = fs.listStatus(d, regionFilter);
819      for (FileStatus regionDir : regionDirs) {
820        Path dd = regionDir.getPath();
821        // else its a region name, now look in region for families
822        FileStatus[] familyDirs = fs.listStatus(dd, familyFilter);
823        for (FileStatus familyDir : familyDirs) {
824          cfCount++;
825          cfCountTotal++;
826          Path family = familyDir.getPath();
827          // now in family make sure only one file
828          FileStatus[] familyStatus = fs.listStatus(family);
829          if (familyStatus.length > 1) {
830            cfFrag++;
831            cfFragTotal++;
832          }
833        }
834      }
835      // compute percentage per table and store in result list
836      frags.put(FSUtils.getTableName(d).getNameAsString(),
837        cfCount == 0? 0: Math.round((float) cfFrag / cfCount * 100));
838    }
839    // set overall percentage for all tables
840    frags.put("-TOTAL-",
841      cfCountTotal == 0? 0: Math.round((float) cfFragTotal / cfCountTotal * 100));
842    return frags;
843  }
844
845  /**
846   * A {@link PathFilter} that returns only regular files.
847   */
848  static class FileFilter extends AbstractFileStatusFilter {
849    private final FileSystem fs;
850
851    public FileFilter(final FileSystem fs) {
852      this.fs = fs;
853    }
854
855    @Override
856    protected boolean accept(Path p, @CheckForNull Boolean isDir) {
857      try {
858        return isFile(fs, isDir, p);
859      } catch (IOException e) {
860        LOG.warn("unable to verify if path=" + p + " is a regular file", e);
861        return false;
862      }
863    }
864  }
865
866  /**
867   * Directory filter that doesn't include any of the directories in the specified blacklist
868   */
869  public static class BlackListDirFilter extends AbstractFileStatusFilter {
870    private final FileSystem fs;
871    private List<String> blacklist;
872
873    /**
874     * Create a filter on the givem filesystem with the specified blacklist
875     * @param fs filesystem to filter
876     * @param directoryNameBlackList list of the names of the directories to filter. If
877     *          <tt>null</tt>, all directories are returned
878     */
879    @SuppressWarnings("unchecked")
880    public BlackListDirFilter(final FileSystem fs, final List<String> directoryNameBlackList) {
881      this.fs = fs;
882      blacklist =
883        (List<String>) (directoryNameBlackList == null ? Collections.emptyList()
884          : directoryNameBlackList);
885    }
886
887    @Override
888    protected boolean accept(Path p, @CheckForNull Boolean isDir) {
889      if (!isValidName(p.getName())) {
890        return false;
891      }
892
893      try {
894        return isDirectory(fs, isDir, p);
895      } catch (IOException e) {
896        LOG.warn("An error occurred while verifying if [" + p.toString()
897            + "] is a valid directory. Returning 'not valid' and continuing.", e);
898        return false;
899      }
900    }
901
902    protected boolean isValidName(final String name) {
903      return !blacklist.contains(name);
904    }
905  }
906
907  /**
908   * A {@link PathFilter} that only allows directories.
909   */
910  public static class DirFilter extends BlackListDirFilter {
911
912    public DirFilter(FileSystem fs) {
913      super(fs, null);
914    }
915  }
916
917  /**
918   * A {@link PathFilter} that returns usertable directories. To get all directories use the
919   * {@link BlackListDirFilter} with a <tt>null</tt> blacklist
920   */
921  public static class UserTableDirFilter extends BlackListDirFilter {
922    public UserTableDirFilter(FileSystem fs) {
923      super(fs, HConstants.HBASE_NON_TABLE_DIRS);
924    }
925
926    @Override
927    protected boolean isValidName(final String name) {
928      if (!super.isValidName(name))
929        return false;
930
931      try {
932        TableName.isLegalTableQualifierName(Bytes.toBytes(name));
933      } catch (IllegalArgumentException e) {
934        LOG.info("INVALID NAME " + name);
935        return false;
936      }
937      return true;
938    }
939  }
940
941  /**
942   * Recover file lease. Used when a file might be suspect
943   * to be had been left open by another process.
944   * @param fs FileSystem handle
945   * @param p Path of file to recover lease
946   * @param conf Configuration handle
947   * @throws IOException
948   */
949  public abstract void recoverFileLease(final FileSystem fs, final Path p,
950      Configuration conf, CancelableProgressable reporter) throws IOException;
951
952  public static List<Path> getTableDirs(final FileSystem fs, final Path rootdir)
953      throws IOException {
954    List<Path> tableDirs = new LinkedList<>();
955
956    for(FileStatus status :
957        fs.globStatus(new Path(rootdir,
958            new Path(HConstants.BASE_NAMESPACE_DIR, "*")))) {
959      tableDirs.addAll(FSUtils.getLocalTableDirs(fs, status.getPath()));
960    }
961    return tableDirs;
962  }
963
964  /**
965   * @param fs
966   * @param rootdir
967   * @return All the table directories under <code>rootdir</code>. Ignore non table hbase folders such as
968   * .logs, .oldlogs, .corrupt folders.
969   * @throws IOException
970   */
971  public static List<Path> getLocalTableDirs(final FileSystem fs, final Path rootdir)
972      throws IOException {
973    // presumes any directory under hbase.rootdir is a table
974    FileStatus[] dirs = fs.listStatus(rootdir, new UserTableDirFilter(fs));
975    List<Path> tabledirs = new ArrayList<>(dirs.length);
976    for (FileStatus dir: dirs) {
977      tabledirs.add(dir.getPath());
978    }
979    return tabledirs;
980  }
981
982  /**
983   * Filter for all dirs that don't start with '.'
984   */
985  public static class RegionDirFilter extends AbstractFileStatusFilter {
986    // This pattern will accept 0.90+ style hex region dirs and older numeric region dir names.
987    final public static Pattern regionDirPattern = Pattern.compile("^[0-9a-f]*$");
988    final FileSystem fs;
989
990    public RegionDirFilter(FileSystem fs) {
991      this.fs = fs;
992    }
993
994    @Override
995    protected boolean accept(Path p, @CheckForNull Boolean isDir) {
996      if (!regionDirPattern.matcher(p.getName()).matches()) {
997        return false;
998      }
999
1000      try {
1001        return isDirectory(fs, isDir, p);
1002      } catch (IOException ioe) {
1003        // Maybe the file was moved or the fs was disconnected.
1004        LOG.warn("Skipping file " + p +" due to IOException", ioe);
1005        return false;
1006      }
1007    }
1008  }
1009
1010  /**
1011   * Given a particular table dir, return all the regiondirs inside it, excluding files such as
1012   * .tableinfo
1013   * @param fs A file system for the Path
1014   * @param tableDir Path to a specific table directory &lt;hbase.rootdir&gt;/&lt;tabledir&gt;
1015   * @return List of paths to valid region directories in table dir.
1016   * @throws IOException
1017   */
1018  public static List<Path> getRegionDirs(final FileSystem fs, final Path tableDir) throws IOException {
1019    // assumes we are in a table dir.
1020    List<FileStatus> rds = listStatusWithStatusFilter(fs, tableDir, new RegionDirFilter(fs));
1021    if (rds == null) {
1022      return new ArrayList<>();
1023    }
1024    List<Path> regionDirs = new ArrayList<>(rds.size());
1025    for (FileStatus rdfs: rds) {
1026      Path rdPath = rdfs.getPath();
1027      regionDirs.add(rdPath);
1028    }
1029    return regionDirs;
1030  }
1031
1032  public static Path getRegionDir(Path tableDir, RegionInfo region) {
1033    return new Path(tableDir, ServerRegionReplicaUtil.getRegionInfoForFs(region).getEncodedName());
1034  }
1035
1036  /**
1037   * Filter for all dirs that are legal column family names.  This is generally used for colfam
1038   * dirs &lt;hbase.rootdir&gt;/&lt;tabledir&gt;/&lt;regiondir&gt;/&lt;colfamdir&gt;.
1039   */
1040  public static class FamilyDirFilter extends AbstractFileStatusFilter {
1041    final FileSystem fs;
1042
1043    public FamilyDirFilter(FileSystem fs) {
1044      this.fs = fs;
1045    }
1046
1047    @Override
1048    protected boolean accept(Path p, @CheckForNull Boolean isDir) {
1049      try {
1050        // throws IAE if invalid
1051        HColumnDescriptor.isLegalFamilyName(Bytes.toBytes(p.getName()));
1052      } catch (IllegalArgumentException iae) {
1053        // path name is an invalid family name and thus is excluded.
1054        return false;
1055      }
1056
1057      try {
1058        return isDirectory(fs, isDir, p);
1059      } catch (IOException ioe) {
1060        // Maybe the file was moved or the fs was disconnected.
1061        LOG.warn("Skipping file " + p +" due to IOException", ioe);
1062        return false;
1063      }
1064    }
1065  }
1066
1067  /**
1068   * Given a particular region dir, return all the familydirs inside it
1069   *
1070   * @param fs A file system for the Path
1071   * @param regionDir Path to a specific region directory
1072   * @return List of paths to valid family directories in region dir.
1073   * @throws IOException
1074   */
1075  public static List<Path> getFamilyDirs(final FileSystem fs, final Path regionDir) throws IOException {
1076    // assumes we are in a region dir.
1077    FileStatus[] fds = fs.listStatus(regionDir, new FamilyDirFilter(fs));
1078    List<Path> familyDirs = new ArrayList<>(fds.length);
1079    for (FileStatus fdfs: fds) {
1080      Path fdPath = fdfs.getPath();
1081      familyDirs.add(fdPath);
1082    }
1083    return familyDirs;
1084  }
1085
1086  public static List<Path> getReferenceFilePaths(final FileSystem fs, final Path familyDir) throws IOException {
1087    List<FileStatus> fds = listStatusWithStatusFilter(fs, familyDir, new ReferenceFileFilter(fs));
1088    if (fds == null) {
1089      return new ArrayList<>();
1090    }
1091    List<Path> referenceFiles = new ArrayList<>(fds.size());
1092    for (FileStatus fdfs: fds) {
1093      Path fdPath = fdfs.getPath();
1094      referenceFiles.add(fdPath);
1095    }
1096    return referenceFiles;
1097  }
1098
1099  /**
1100   * Filter for HFiles that excludes reference files.
1101   */
1102  public static class HFileFilter extends AbstractFileStatusFilter {
1103    final FileSystem fs;
1104
1105    public HFileFilter(FileSystem fs) {
1106      this.fs = fs;
1107    }
1108
1109    @Override
1110    protected boolean accept(Path p, @CheckForNull Boolean isDir) {
1111      if (!StoreFileInfo.isHFile(p)) {
1112        return false;
1113      }
1114
1115      try {
1116        return isFile(fs, isDir, p);
1117      } catch (IOException ioe) {
1118        // Maybe the file was moved or the fs was disconnected.
1119        LOG.warn("Skipping file " + p +" due to IOException", ioe);
1120        return false;
1121      }
1122    }
1123  }
1124
1125  /**
1126   * Filter for HFileLinks (StoreFiles and HFiles not included).
1127   * the filter itself does not consider if a link is file or not.
1128   */
1129  public static class HFileLinkFilter implements PathFilter {
1130
1131    @Override
1132    public boolean accept(Path p) {
1133      return HFileLink.isHFileLink(p);
1134    }
1135  }
1136
1137  public static class ReferenceFileFilter extends AbstractFileStatusFilter {
1138
1139    private final FileSystem fs;
1140
1141    public ReferenceFileFilter(FileSystem fs) {
1142      this.fs = fs;
1143    }
1144
1145    @Override
1146    protected boolean accept(Path p, @CheckForNull Boolean isDir) {
1147      if (!StoreFileInfo.isReference(p)) {
1148        return false;
1149      }
1150
1151      try {
1152        // only files can be references.
1153        return isFile(fs, isDir, p);
1154      } catch (IOException ioe) {
1155        // Maybe the file was moved or the fs was disconnected.
1156        LOG.warn("Skipping file " + p +" due to IOException", ioe);
1157        return false;
1158      }
1159    }
1160  }
1161
1162  /**
1163   * Runs through the HBase rootdir/tablename and creates a reverse lookup map for
1164   * table StoreFile names to the full Path.
1165   * <br>
1166   * Example...<br>
1167   * Key = 3944417774205889744  <br>
1168   * Value = hdfs://localhost:51169/user/userid/-ROOT-/70236052/info/3944417774205889744
1169   *
1170   * @param map map to add values.  If null, this method will create and populate one to return
1171   * @param fs  The file system to use.
1172   * @param hbaseRootDir  The root directory to scan.
1173   * @param tableName name of the table to scan.
1174   * @return Map keyed by StoreFile name with a value of the full Path.
1175   * @throws IOException When scanning the directory fails.
1176   * @throws InterruptedException
1177   */
1178  public static Map<String, Path> getTableStoreFilePathMap(Map<String, Path> map,
1179  final FileSystem fs, final Path hbaseRootDir, TableName tableName)
1180  throws IOException, InterruptedException {
1181    return getTableStoreFilePathMap(map, fs, hbaseRootDir, tableName, null, null, null);
1182  }
1183
1184  /**
1185   * Runs through the HBase rootdir/tablename and creates a reverse lookup map for
1186   * table StoreFile names to the full Path.  Note that because this method can be called
1187   * on a 'live' HBase system that we will skip files that no longer exist by the time
1188   * we traverse them and similarly the user of the result needs to consider that some
1189   * entries in this map may not exist by the time this call completes.
1190   * <br>
1191   * Example...<br>
1192   * Key = 3944417774205889744  <br>
1193   * Value = hdfs://localhost:51169/user/userid/-ROOT-/70236052/info/3944417774205889744
1194   *
1195   * @param resultMap map to add values.  If null, this method will create and populate one to return
1196   * @param fs  The file system to use.
1197   * @param hbaseRootDir  The root directory to scan.
1198   * @param tableName name of the table to scan.
1199   * @param sfFilter optional path filter to apply to store files
1200   * @param executor optional executor service to parallelize this operation
1201   * @param errors ErrorReporter instance or null
1202   * @return Map keyed by StoreFile name with a value of the full Path.
1203   * @throws IOException When scanning the directory fails.
1204   * @throws InterruptedException
1205   */
1206  public static Map<String, Path> getTableStoreFilePathMap(
1207      Map<String, Path> resultMap,
1208      final FileSystem fs, final Path hbaseRootDir, TableName tableName, final PathFilter sfFilter,
1209      ExecutorService executor, final ErrorReporter errors) throws IOException, InterruptedException {
1210
1211    final Map<String, Path> finalResultMap =
1212        resultMap == null ? new ConcurrentHashMap<>(128, 0.75f, 32) : resultMap;
1213
1214    // only include the directory paths to tables
1215    Path tableDir = FSUtils.getTableDir(hbaseRootDir, tableName);
1216    // Inside a table, there are compaction.dir directories to skip.  Otherwise, all else
1217    // should be regions.
1218    final FamilyDirFilter familyFilter = new FamilyDirFilter(fs);
1219    final Vector<Exception> exceptions = new Vector<>();
1220
1221    try {
1222      List<FileStatus> regionDirs = FSUtils.listStatusWithStatusFilter(fs, tableDir, new RegionDirFilter(fs));
1223      if (regionDirs == null) {
1224        return finalResultMap;
1225      }
1226
1227      final List<Future<?>> futures = new ArrayList<>(regionDirs.size());
1228
1229      for (FileStatus regionDir : regionDirs) {
1230        if (null != errors) {
1231          errors.progress();
1232        }
1233        final Path dd = regionDir.getPath();
1234
1235        if (!exceptions.isEmpty()) {
1236          break;
1237        }
1238
1239        Runnable getRegionStoreFileMapCall = new Runnable() {
1240          @Override
1241          public void run() {
1242            try {
1243              HashMap<String,Path> regionStoreFileMap = new HashMap<>();
1244              List<FileStatus> familyDirs = FSUtils.listStatusWithStatusFilter(fs, dd, familyFilter);
1245              if (familyDirs == null) {
1246                if (!fs.exists(dd)) {
1247                  LOG.warn("Skipping region because it no longer exists: " + dd);
1248                } else {
1249                  LOG.warn("Skipping region because it has no family dirs: " + dd);
1250                }
1251                return;
1252              }
1253              for (FileStatus familyDir : familyDirs) {
1254                if (null != errors) {
1255                  errors.progress();
1256                }
1257                Path family = familyDir.getPath();
1258                if (family.getName().equals(HConstants.RECOVERED_EDITS_DIR)) {
1259                  continue;
1260                }
1261                // now in family, iterate over the StoreFiles and
1262                // put in map
1263                FileStatus[] familyStatus = fs.listStatus(family);
1264                for (FileStatus sfStatus : familyStatus) {
1265                  if (null != errors) {
1266                    errors.progress();
1267                  }
1268                  Path sf = sfStatus.getPath();
1269                  if (sfFilter == null || sfFilter.accept(sf)) {
1270                    regionStoreFileMap.put( sf.getName(), sf);
1271                  }
1272                }
1273              }
1274              finalResultMap.putAll(regionStoreFileMap);
1275            } catch (Exception e) {
1276              LOG.error("Could not get region store file map for region: " + dd, e);
1277              exceptions.add(e);
1278            }
1279          }
1280        };
1281
1282        // If executor is available, submit async tasks to exec concurrently, otherwise
1283        // just do serial sync execution
1284        if (executor != null) {
1285          Future<?> future = executor.submit(getRegionStoreFileMapCall);
1286          futures.add(future);
1287        } else {
1288          FutureTask<?> future = new FutureTask<>(getRegionStoreFileMapCall, null);
1289          future.run();
1290          futures.add(future);
1291        }
1292      }
1293
1294      // Ensure all pending tasks are complete (or that we run into an exception)
1295      for (Future<?> f : futures) {
1296        if (!exceptions.isEmpty()) {
1297          break;
1298        }
1299        try {
1300          f.get();
1301        } catch (ExecutionException e) {
1302          LOG.error("Unexpected exec exception!  Should've been caught already.  (Bug?)", e);
1303          // Shouldn't happen, we already logged/caught any exceptions in the Runnable
1304        }
1305      }
1306    } catch (IOException e) {
1307      LOG.error("Cannot execute getTableStoreFilePathMap for " + tableName, e);
1308      exceptions.add(e);
1309    } finally {
1310      if (!exceptions.isEmpty()) {
1311        // Just throw the first exception as an indication something bad happened
1312        // Don't need to propagate all the exceptions, we already logged them all anyway
1313        Throwables.propagateIfInstanceOf(exceptions.firstElement(), IOException.class);
1314        throw Throwables.propagate(exceptions.firstElement());
1315      }
1316    }
1317
1318    return finalResultMap;
1319  }
1320
1321  public static int getRegionReferenceFileCount(final FileSystem fs, final Path p) {
1322    int result = 0;
1323    try {
1324      for (Path familyDir:getFamilyDirs(fs, p)){
1325        result += getReferenceFilePaths(fs, familyDir).size();
1326      }
1327    } catch (IOException e) {
1328      LOG.warn("Error Counting reference files.", e);
1329    }
1330    return result;
1331  }
1332
1333  /**
1334   * Runs through the HBase rootdir and creates a reverse lookup map for
1335   * table StoreFile names to the full Path.
1336   * <br>
1337   * Example...<br>
1338   * Key = 3944417774205889744  <br>
1339   * Value = hdfs://localhost:51169/user/userid/-ROOT-/70236052/info/3944417774205889744
1340   *
1341   * @param fs  The file system to use.
1342   * @param hbaseRootDir  The root directory to scan.
1343   * @return Map keyed by StoreFile name with a value of the full Path.
1344   * @throws IOException When scanning the directory fails.
1345   * @throws InterruptedException
1346   */
1347  public static Map<String, Path> getTableStoreFilePathMap(
1348    final FileSystem fs, final Path hbaseRootDir)
1349  throws IOException, InterruptedException {
1350    return getTableStoreFilePathMap(fs, hbaseRootDir, null, null, null);
1351  }
1352
1353  /**
1354   * Runs through the HBase rootdir and creates a reverse lookup map for
1355   * table StoreFile names to the full Path.
1356   * <br>
1357   * Example...<br>
1358   * Key = 3944417774205889744  <br>
1359   * Value = hdfs://localhost:51169/user/userid/-ROOT-/70236052/info/3944417774205889744
1360   *
1361   * @param fs  The file system to use.
1362   * @param hbaseRootDir  The root directory to scan.
1363   * @param sfFilter optional path filter to apply to store files
1364   * @param executor optional executor service to parallelize this operation
1365   * @param errors ErrorReporter instance or null
1366   * @return Map keyed by StoreFile name with a value of the full Path.
1367   * @throws IOException When scanning the directory fails.
1368   * @throws InterruptedException
1369   */
1370  public static Map<String, Path> getTableStoreFilePathMap(
1371    final FileSystem fs, final Path hbaseRootDir, PathFilter sfFilter,
1372    ExecutorService executor, ErrorReporter errors)
1373  throws IOException, InterruptedException {
1374    ConcurrentHashMap<String, Path> map = new ConcurrentHashMap<>(1024, 0.75f, 32);
1375
1376    // if this method looks similar to 'getTableFragmentation' that is because
1377    // it was borrowed from it.
1378
1379    // only include the directory paths to tables
1380    for (Path tableDir : FSUtils.getTableDirs(fs, hbaseRootDir)) {
1381      getTableStoreFilePathMap(map, fs, hbaseRootDir,
1382          FSUtils.getTableName(tableDir), sfFilter, executor, errors);
1383    }
1384    return map;
1385  }
1386
1387  /**
1388   * Filters FileStatuses in an array and returns a list
1389   *
1390   * @param input   An array of FileStatuses
1391   * @param filter  A required filter to filter the array
1392   * @return        A list of FileStatuses
1393   */
1394  public static List<FileStatus> filterFileStatuses(FileStatus[] input,
1395      FileStatusFilter filter) {
1396    if (input == null) return null;
1397    return filterFileStatuses(Iterators.forArray(input), filter);
1398  }
1399
1400  /**
1401   * Filters FileStatuses in an iterator and returns a list
1402   *
1403   * @param input   An iterator of FileStatuses
1404   * @param filter  A required filter to filter the array
1405   * @return        A list of FileStatuses
1406   */
1407  public static List<FileStatus> filterFileStatuses(Iterator<FileStatus> input,
1408      FileStatusFilter filter) {
1409    if (input == null) return null;
1410    ArrayList<FileStatus> results = new ArrayList<>();
1411    while (input.hasNext()) {
1412      FileStatus f = input.next();
1413      if (filter.accept(f)) {
1414        results.add(f);
1415      }
1416    }
1417    return results;
1418  }
1419
1420  /**
1421   * Calls fs.listStatus() and treats FileNotFoundException as non-fatal
1422   * This accommodates differences between hadoop versions, where hadoop 1
1423   * does not throw a FileNotFoundException, and return an empty FileStatus[]
1424   * while Hadoop 2 will throw FileNotFoundException.
1425   *
1426   * @param fs file system
1427   * @param dir directory
1428   * @param filter file status filter
1429   * @return null if dir is empty or doesn't exist, otherwise FileStatus list
1430   */
1431  public static List<FileStatus> listStatusWithStatusFilter(final FileSystem fs,
1432      final Path dir, final FileStatusFilter filter) throws IOException {
1433    FileStatus [] status = null;
1434    try {
1435      status = fs.listStatus(dir);
1436    } catch (FileNotFoundException fnfe) {
1437      // if directory doesn't exist, return null
1438      if (LOG.isTraceEnabled()) {
1439        LOG.trace(dir + " doesn't exist");
1440      }
1441    }
1442
1443    if (status == null || status.length < 1)  {
1444      return null;
1445    }
1446
1447    if (filter == null) {
1448      return Arrays.asList(status);
1449    } else {
1450      List<FileStatus> status2 = filterFileStatuses(status, filter);
1451      if (status2 == null || status2.isEmpty()) {
1452        return null;
1453      } else {
1454        return status2;
1455      }
1456    }
1457  }
1458
1459  /**
1460   * Throw an exception if an action is not permitted by a user on a file.
1461   *
1462   * @param ugi
1463   *          the user
1464   * @param file
1465   *          the file
1466   * @param action
1467   *          the action
1468   */
1469  public static void checkAccess(UserGroupInformation ugi, FileStatus file,
1470      FsAction action) throws AccessDeniedException {
1471    if (ugi.getShortUserName().equals(file.getOwner())) {
1472      if (file.getPermission().getUserAction().implies(action)) {
1473        return;
1474      }
1475    } else if (contains(ugi.getGroupNames(), file.getGroup())) {
1476      if (file.getPermission().getGroupAction().implies(action)) {
1477        return;
1478      }
1479    } else if (file.getPermission().getOtherAction().implies(action)) {
1480      return;
1481    }
1482    throw new AccessDeniedException("Permission denied:" + " action=" + action
1483        + " path=" + file.getPath() + " user=" + ugi.getShortUserName());
1484  }
1485
1486  private static boolean contains(String[] groups, String user) {
1487    for (String group : groups) {
1488      if (group.equals(user)) {
1489        return true;
1490      }
1491    }
1492    return false;
1493  }
1494
1495  /**
1496   * This function is to scan the root path of the file system to get the
1497   * degree of locality for each region on each of the servers having at least
1498   * one block of that region.
1499   * This is used by the tool {@link org.apache.hadoop.hbase.master.RegionPlacementMaintainer}
1500   *
1501   * @param conf
1502   *          the configuration to use
1503   * @return the mapping from region encoded name to a map of server names to
1504   *           locality fraction
1505   * @throws IOException
1506   *           in case of file system errors or interrupts
1507   */
1508  public static Map<String, Map<String, Float>> getRegionDegreeLocalityMappingFromFS(
1509      final Configuration conf) throws IOException {
1510    return getRegionDegreeLocalityMappingFromFS(
1511        conf, null,
1512        conf.getInt(THREAD_POOLSIZE, DEFAULT_THREAD_POOLSIZE));
1513
1514  }
1515
1516  /**
1517   * This function is to scan the root path of the file system to get the
1518   * degree of locality for each region on each of the servers having at least
1519   * one block of that region.
1520   *
1521   * @param conf
1522   *          the configuration to use
1523   * @param desiredTable
1524   *          the table you wish to scan locality for
1525   * @param threadPoolSize
1526   *          the thread pool size to use
1527   * @return the mapping from region encoded name to a map of server names to
1528   *           locality fraction
1529   * @throws IOException
1530   *           in case of file system errors or interrupts
1531   */
1532  public static Map<String, Map<String, Float>> getRegionDegreeLocalityMappingFromFS(
1533      final Configuration conf, final String desiredTable, int threadPoolSize)
1534      throws IOException {
1535    Map<String, Map<String, Float>> regionDegreeLocalityMapping = new ConcurrentHashMap<>();
1536    getRegionLocalityMappingFromFS(conf, desiredTable, threadPoolSize, null,
1537        regionDegreeLocalityMapping);
1538    return regionDegreeLocalityMapping;
1539  }
1540
1541  /**
1542   * This function is to scan the root path of the file system to get either the
1543   * mapping between the region name and its best locality region server or the
1544   * degree of locality of each region on each of the servers having at least
1545   * one block of that region. The output map parameters are both optional.
1546   *
1547   * @param conf
1548   *          the configuration to use
1549   * @param desiredTable
1550   *          the table you wish to scan locality for
1551   * @param threadPoolSize
1552   *          the thread pool size to use
1553   * @param regionToBestLocalityRSMapping
1554   *          the map into which to put the best locality mapping or null
1555   * @param regionDegreeLocalityMapping
1556   *          the map into which to put the locality degree mapping or null,
1557   *          must be a thread-safe implementation
1558   * @throws IOException
1559   *           in case of file system errors or interrupts
1560   */
1561  private static void getRegionLocalityMappingFromFS(
1562      final Configuration conf, final String desiredTable,
1563      int threadPoolSize,
1564      Map<String, String> regionToBestLocalityRSMapping,
1565      Map<String, Map<String, Float>> regionDegreeLocalityMapping)
1566      throws IOException {
1567    FileSystem fs =  FileSystem.get(conf);
1568    Path rootPath = FSUtils.getRootDir(conf);
1569    long startTime = EnvironmentEdgeManager.currentTime();
1570    Path queryPath;
1571    // The table files are in ${hbase.rootdir}/data/<namespace>/<table>/*
1572    if (null == desiredTable) {
1573      queryPath = new Path(new Path(rootPath, HConstants.BASE_NAMESPACE_DIR).toString() + "/*/*/*/");
1574    } else {
1575      queryPath = new Path(FSUtils.getTableDir(rootPath, TableName.valueOf(desiredTable)).toString() + "/*/");
1576    }
1577
1578    // reject all paths that are not appropriate
1579    PathFilter pathFilter = new PathFilter() {
1580      @Override
1581      public boolean accept(Path path) {
1582        // this is the region name; it may get some noise data
1583        if (null == path) {
1584          return false;
1585        }
1586
1587        // no parent?
1588        Path parent = path.getParent();
1589        if (null == parent) {
1590          return false;
1591        }
1592
1593        String regionName = path.getName();
1594        if (null == regionName) {
1595          return false;
1596        }
1597
1598        if (!regionName.toLowerCase(Locale.ROOT).matches("[0-9a-f]+")) {
1599          return false;
1600        }
1601        return true;
1602      }
1603    };
1604
1605    FileStatus[] statusList = fs.globStatus(queryPath, pathFilter);
1606
1607    if (null == statusList) {
1608      return;
1609    } else {
1610      LOG.debug("Query Path: " + queryPath + " ; # list of files: " +
1611          statusList.length);
1612    }
1613
1614    // lower the number of threads in case we have very few expected regions
1615    threadPoolSize = Math.min(threadPoolSize, statusList.length);
1616
1617    // run in multiple threads
1618    ThreadPoolExecutor tpe = new ThreadPoolExecutor(threadPoolSize,
1619        threadPoolSize, 60, TimeUnit.SECONDS,
1620        new ArrayBlockingQueue<>(statusList.length));
1621    try {
1622      // ignore all file status items that are not of interest
1623      for (FileStatus regionStatus : statusList) {
1624        if (null == regionStatus) {
1625          continue;
1626        }
1627
1628        if (!regionStatus.isDirectory()) {
1629          continue;
1630        }
1631
1632        Path regionPath = regionStatus.getPath();
1633        if (null == regionPath) {
1634          continue;
1635        }
1636
1637        tpe.execute(new FSRegionScanner(fs, regionPath,
1638            regionToBestLocalityRSMapping, regionDegreeLocalityMapping));
1639      }
1640    } finally {
1641      tpe.shutdown();
1642      int threadWakeFrequency = conf.getInt(HConstants.THREAD_WAKE_FREQUENCY,
1643          60 * 1000);
1644      try {
1645        // here we wait until TPE terminates, which is either naturally or by
1646        // exceptions in the execution of the threads
1647        while (!tpe.awaitTermination(threadWakeFrequency,
1648            TimeUnit.MILLISECONDS)) {
1649          // printing out rough estimate, so as to not introduce
1650          // AtomicInteger
1651          LOG.info("Locality checking is underway: { Scanned Regions : "
1652              + tpe.getCompletedTaskCount() + "/"
1653              + tpe.getTaskCount() + " }");
1654        }
1655      } catch (InterruptedException e) {
1656        throw (InterruptedIOException)new InterruptedIOException().initCause(e);
1657      }
1658    }
1659
1660    long overhead = EnvironmentEdgeManager.currentTime() - startTime;
1661    String overheadMsg = "Scan DFS for locality info takes " + overhead + " ms";
1662
1663    LOG.info(overheadMsg);
1664  }
1665
1666  /**
1667   * Do our short circuit read setup.
1668   * Checks buffer size to use and whether to do checksumming in hbase or hdfs.
1669   * @param conf
1670   */
1671  public static void setupShortCircuitRead(final Configuration conf) {
1672    // Check that the user has not set the "dfs.client.read.shortcircuit.skip.checksum" property.
1673    boolean shortCircuitSkipChecksum =
1674      conf.getBoolean("dfs.client.read.shortcircuit.skip.checksum", false);
1675    boolean useHBaseChecksum = conf.getBoolean(HConstants.HBASE_CHECKSUM_VERIFICATION, true);
1676    if (shortCircuitSkipChecksum) {
1677      LOG.warn("Configuration \"dfs.client.read.shortcircuit.skip.checksum\" should not " +
1678        "be set to true." + (useHBaseChecksum ? " HBase checksum doesn't require " +
1679        "it, see https://issues.apache.org/jira/browse/HBASE-6868." : ""));
1680      assert !shortCircuitSkipChecksum; //this will fail if assertions are on
1681    }
1682    checkShortCircuitReadBufferSize(conf);
1683  }
1684
1685  /**
1686   * Check if short circuit read buffer size is set and if not, set it to hbase value.
1687   * @param conf
1688   */
1689  public static void checkShortCircuitReadBufferSize(final Configuration conf) {
1690    final int defaultSize = HConstants.DEFAULT_BLOCKSIZE * 2;
1691    final int notSet = -1;
1692    // DFSConfigKeys.DFS_CLIENT_READ_SHORTCIRCUIT_BUFFER_SIZE_KEY is only defined in h2
1693    final String dfsKey = "dfs.client.read.shortcircuit.buffer.size";
1694    int size = conf.getInt(dfsKey, notSet);
1695    // If a size is set, return -- we will use it.
1696    if (size != notSet) return;
1697    // But short circuit buffer size is normally not set.  Put in place the hbase wanted size.
1698    int hbaseSize = conf.getInt("hbase." + dfsKey, defaultSize);
1699    conf.setIfUnset(dfsKey, Integer.toString(hbaseSize));
1700  }
1701
1702  /**
1703   * @param c
1704   * @return The DFSClient DFSHedgedReadMetrics instance or null if can't be found or not on hdfs.
1705   * @throws IOException
1706   */
1707  public static DFSHedgedReadMetrics getDFSHedgedReadMetrics(final Configuration c)
1708      throws IOException {
1709    if (!isHDFS(c)) return null;
1710    // getHedgedReadMetrics is package private. Get the DFSClient instance that is internal
1711    // to the DFS FS instance and make the method getHedgedReadMetrics accessible, then invoke it
1712    // to get the singleton instance of DFSHedgedReadMetrics shared by DFSClients.
1713    final String name = "getHedgedReadMetrics";
1714    DFSClient dfsclient = ((DistributedFileSystem)FileSystem.get(c)).getClient();
1715    Method m;
1716    try {
1717      m = dfsclient.getClass().getDeclaredMethod(name);
1718    } catch (NoSuchMethodException e) {
1719      LOG.warn("Failed find method " + name + " in dfsclient; no hedged read metrics: " +
1720          e.getMessage());
1721      return null;
1722    } catch (SecurityException e) {
1723      LOG.warn("Failed find method " + name + " in dfsclient; no hedged read metrics: " +
1724          e.getMessage());
1725      return null;
1726    }
1727    m.setAccessible(true);
1728    try {
1729      return (DFSHedgedReadMetrics)m.invoke(dfsclient);
1730    } catch (IllegalAccessException e) {
1731      LOG.warn("Failed invoking method " + name + " on dfsclient; no hedged read metrics: " +
1732          e.getMessage());
1733      return null;
1734    } catch (IllegalArgumentException e) {
1735      LOG.warn("Failed invoking method " + name + " on dfsclient; no hedged read metrics: " +
1736          e.getMessage());
1737      return null;
1738    } catch (InvocationTargetException e) {
1739      LOG.warn("Failed invoking method " + name + " on dfsclient; no hedged read metrics: " +
1740          e.getMessage());
1741      return null;
1742    }
1743  }
1744
1745}