Example #1
0
 /**
  * Update the maps baseDirSize and baseDirNumberSubDir when deleting cache.
  *
  * @param cacheStatus cache status of the cache is deleted
  */
 private static void deleteCacheInfoUpdate(CacheStatus cacheStatus) {
   if (!cacheStatus.isInited()) {
     // if it is not created yet, do nothing.
     return;
   }
   synchronized (baseDirSize) {
     Long dirSize = baseDirSize.get(cacheStatus.getBaseDir());
     if (dirSize != null) {
       dirSize -= cacheStatus.size;
       baseDirSize.put(cacheStatus.getBaseDir(), dirSize);
     }
   }
   synchronized (baseDirNumberSubDir) {
     Integer dirSubDir = baseDirNumberSubDir.get(cacheStatus.getBaseDir());
     if (dirSubDir != null) {
       dirSubDir--;
       baseDirNumberSubDir.put(cacheStatus.getBaseDir(), dirSubDir);
     }
   }
 }
Example #2
0
 /**
  * Update the maps baseDirSize and baseDirNumberSubDir when adding cache.
  *
  * @param cacheStatus cache status of the cache is added
  */
 private static void addCacheInfoUpdate(CacheStatus cacheStatus) {
   long cacheSize = cacheStatus.size;
   synchronized (baseDirSize) {
     Long dirSize = baseDirSize.get(cacheStatus.getBaseDir());
     if (dirSize == null) {
       dirSize = Long.valueOf(cacheSize);
     } else {
       dirSize += cacheSize;
     }
     baseDirSize.put(cacheStatus.getBaseDir(), dirSize);
   }
   synchronized (baseDirNumberSubDir) {
     Integer dirSubDir = baseDirNumberSubDir.get(cacheStatus.getBaseDir());
     if (dirSubDir == null) {
       dirSubDir = 1;
     } else {
       dirSubDir += 1;
     }
     baseDirNumberSubDir.put(cacheStatus.getBaseDir(), dirSubDir);
   }
 }
Example #3
0
  /**
   * This is the opposite of getlocalcache. When you are done with using the cache, you need to
   * release the cache
   *
   * @param cache The cache URI to be released
   * @param conf configuration which contains the filesystem the cache is contained in.
   * @throws IOException
   */
  public static void releaseCache(URI cache, Configuration conf, long timeStamp)
      throws IOException {
    String cacheId = getKey(cache, conf, timeStamp);

    synchronized (cachedArchives) {
      CacheStatus lcacheStatus = cachedArchives.get(cacheId);
      if (lcacheStatus == null) {
        LOG.warn(
            "Cannot find localized cache: " + cache + " (key: " + cacheId + ") in releaseCache!");
        return;
      }
      lcacheStatus.refcount--;
    }
  }
Example #4
0
 private static void deleteCache(Configuration conf, MRAsyncDiskService asyncDiskService)
     throws IOException {
   List<CacheStatus> deleteSet = new LinkedList<CacheStatus>();
   // try deleting cache Status with refcount of zero
   synchronized (cachedArchives) {
     for (Iterator<String> it = cachedArchives.keySet().iterator(); it.hasNext(); ) {
       String cacheId = (String) it.next();
       CacheStatus lcacheStatus = cachedArchives.get(cacheId);
       if (lcacheStatus.refcount == 0) {
         // delete this cache entry from the global list
         // and mark the localized file for deletion
         deleteSet.add(lcacheStatus);
         it.remove();
       }
     }
   }
   // do the deletion asynchronously, after releasing the global lock
   Thread cacheFileCleaner =
       new Thread(new CacheFileCleanTask(asyncDiskService, FileSystem.getLocal(conf), deleteSet));
   cacheFileCleaner.start();
 }
Example #5
0
  /**
   * Get the locally cached file or archive; it could either be previously cached (and valid) or
   * copy it from the {@link FileSystem} now.
   *
   * @param cache the cache to be localized, this should be specified as new
   *     URI(hdfs://hostname:port/absolute_path_to_file#LINKNAME). If no schema or hostname:port is
   *     provided the file is assumed to be in the filesystem being used in the Configuration
   * @param conf The Confguration file which contains the filesystem
   * @param subDir The sub cache Dir where you want to localize the files/archives
   * @param fileStatus The file status on the dfs.
   * @param isArchive if the cache is an archive or a file. In case it is an archive with a .zip or
   *     .jar or .tar or .tgz or .tar.gz extension it will be unzipped/unjarred/untarred
   *     automatically and the directory where the archive is unzipped/unjarred/untarred is returned
   *     as the Path. In case of a file, the path to the file is returned
   * @param confFileStamp this is the hdfs file modification timestamp to verify that the file to be
   *     cached hasn't changed since the job started
   * @param fileLength this is the length of the cache file
   * @param currentWorkDir this is the directory where you would want to create symlinks for the
   *     locally cached files/archives
   * @param honorSymLinkConf if this is false, then the symlinks are not created even if conf says
   *     so (this is required for an optimization in task launches
   * @param lDirAllocator LocalDirAllocator of the tracker
   * @return the path to directory where the archives are unjarred in case of archives, the path to
   *     the file where the file is copied locally
   * @throws IOException
   */
  private static Path getLocalCache(
      URI cache,
      Configuration conf,
      Path subDir,
      FileStatus fileStatus,
      boolean isArchive,
      long confFileStamp,
      long fileLength,
      Path currentWorkDir,
      boolean honorSymLinkConf,
      MRAsyncDiskService asyncDiskService,
      LocalDirAllocator lDirAllocator)
      throws IOException {
    String key = getKey(cache, conf, confFileStamp);

    CacheStatus lcacheStatus;
    Path localizedPath;
    synchronized (cachedArchives) {
      lcacheStatus = cachedArchives.get(key);
      if (lcacheStatus == null) {
        // was never localized
        Path uniqueParentDir = new Path(subDir, String.valueOf(random.nextLong()));
        String cachePath = new Path(uniqueParentDir, makeRelative(cache, conf)).toString();
        Path localPath = lDirAllocator.getLocalPathForWrite(cachePath, fileLength, conf);
        lcacheStatus =
            new CacheStatus(
                new Path(localPath.toString().replace(cachePath, "")), localPath, uniqueParentDir);
        cachedArchives.put(key, lcacheStatus);
      }
      lcacheStatus.refcount++;
    }
    boolean initSuccessful = false;
    try {
      synchronized (lcacheStatus) {
        if (!lcacheStatus.isInited()) {
          localizedPath = localizeCache(conf, cache, confFileStamp, lcacheStatus, isArchive);
          lcacheStatus.initComplete();
        } else {
          if (fileStatus != null) {
            localizedPath =
                checkCacheStatusValidity(
                    conf, cache, confFileStamp, lcacheStatus, fileStatus, isArchive);
          } else {
            // if fileStatus is null, then the md5 must be correct
            // so there is no need to check for cache validity
            localizedPath = lcacheStatus.localizedLoadPath;
          }
        }
        createSymlink(conf, cache, lcacheStatus, isArchive, currentWorkDir, honorSymLinkConf);
      }

      // try deleting stuff if you can
      long size = 0;
      int numberSubDir = 0;
      synchronized (lcacheStatus) {
        synchronized (baseDirSize) {
          Long get = baseDirSize.get(lcacheStatus.getBaseDir());
          if (get != null) {
            size = get.longValue();
          } else {
            LOG.warn("Cannot find size of baseDir: " + lcacheStatus.getBaseDir());
          }
        }
        synchronized (baseDirNumberSubDir) {
          Integer get = baseDirNumberSubDir.get(lcacheStatus.getBaseDir());
          if (get != null) {
            numberSubDir = get.intValue();
          } else {
            LOG.warn("Cannot find subdirectories limit of baseDir: " + lcacheStatus.getBaseDir());
          }
        }
      }
      // setting the cache size to a default of 10GB
      long allowedSize = conf.getLong("local.cache.size", DEFAULT_CACHE_SIZE);
      long allowedNumberSubDir =
          conf.getLong("local.cache.numbersubdir", DEFAULT_CACHE_SUBDIR_LIMIT);
      if (allowedSize < size || allowedNumberSubDir < numberSubDir) {
        // try some cache deletions
        LOG.debug(
            "Start deleting released cache because"
                + " [size, allowedSize, numberSubDir, allowedNumberSubDir] ="
                + " ["
                + size
                + ", "
                + allowedSize
                + ", "
                + numberSubDir
                + ", "
                + allowedNumberSubDir
                + "]");
        deleteCache(conf, asyncDiskService);
      }
      initSuccessful = true;
      return localizedPath;
    } finally {
      if (!initSuccessful) {
        synchronized (cachedArchives) {
          lcacheStatus.refcount--;
        }
      }
    }
  }