Package org.apache.spark.network.shuffle

Source Code of org.apache.spark.network.shuffle.ExternalShuffleBlockManager$AppExecId

/*
* 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.spark.network.shuffle;

import java.io.DataInputStream;
import java.io.File;
import java.io.FileInputStream;
import java.io.IOException;
import java.util.Iterator;
import java.util.Map;
import java.util.concurrent.ConcurrentMap;
import java.util.concurrent.Executor;
import java.util.concurrent.Executors;

import com.google.common.annotations.VisibleForTesting;
import com.google.common.base.Objects;
import com.google.common.collect.Maps;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;

import org.apache.spark.network.buffer.FileSegmentManagedBuffer;
import org.apache.spark.network.buffer.ManagedBuffer;
import org.apache.spark.network.shuffle.protocol.ExecutorShuffleInfo;
import org.apache.spark.network.util.JavaUtils;

/**
* Manages converting shuffle BlockIds into physical segments of local files, from a process outside
* of Executors. Each Executor must register its own configuration about where it stores its files
* (local dirs) and how (shuffle manager). The logic for retrieval of individual files is replicated
* from Spark's FileShuffleBlockManager and IndexShuffleBlockManager.
*
* Executors with shuffle file consolidation are not currently supported, as the index is stored in
* the Executor's memory, unlike the IndexShuffleBlockManager.
*/
public class ExternalShuffleBlockManager {
  private final Logger logger = LoggerFactory.getLogger(ExternalShuffleBlockManager.class);

  // Map containing all registered executors' metadata.
  private final ConcurrentMap<AppExecId, ExecutorShuffleInfo> executors;

  // Single-threaded Java executor used to perform expensive recursive directory deletion.
  private final Executor directoryCleaner;

  public ExternalShuffleBlockManager() {
    // TODO: Give this thread a name.
    this(Executors.newSingleThreadExecutor());
  }

  // Allows tests to have more control over when directories are cleaned up.
  @VisibleForTesting
  ExternalShuffleBlockManager(Executor directoryCleaner) {
    this.executors = Maps.newConcurrentMap();
    this.directoryCleaner = directoryCleaner;
  }

  /** Registers a new Executor with all the configuration we need to find its shuffle files. */
  public void registerExecutor(
      String appId,
      String execId,
      ExecutorShuffleInfo executorInfo) {
    AppExecId fullId = new AppExecId(appId, execId);
    logger.info("Registered executor {} with {}", fullId, executorInfo);
    executors.put(fullId, executorInfo);
  }

  /**
   * Obtains a FileSegmentManagedBuffer from a shuffle block id. We expect the blockId has the
   * format "shuffle_ShuffleId_MapId_ReduceId" (from ShuffleBlockId), and additionally make
   * assumptions about how the hash and sort based shuffles store their data.
   */
  public ManagedBuffer getBlockData(String appId, String execId, String blockId) {
    String[] blockIdParts = blockId.split("_");
    if (blockIdParts.length < 4) {
      throw new IllegalArgumentException("Unexpected block id format: " + blockId);
    } else if (!blockIdParts[0].equals("shuffle")) {
      throw new IllegalArgumentException("Expected shuffle block id, got: " + blockId);
    }
    int shuffleId = Integer.parseInt(blockIdParts[1]);
    int mapId = Integer.parseInt(blockIdParts[2]);
    int reduceId = Integer.parseInt(blockIdParts[3]);

    ExecutorShuffleInfo executor = executors.get(new AppExecId(appId, execId));
    if (executor == null) {
      throw new RuntimeException(
        String.format("Executor is not registered (appId=%s, execId=%s)", appId, execId));
    }

    if ("org.apache.spark.shuffle.hash.HashShuffleManager".equals(executor.shuffleManager)) {
      return getHashBasedShuffleBlockData(executor, blockId);
    } else if ("org.apache.spark.shuffle.sort.SortShuffleManager".equals(executor.shuffleManager)) {
      return getSortBasedShuffleBlockData(executor, shuffleId, mapId, reduceId);
    } else {
      throw new UnsupportedOperationException(
        "Unsupported shuffle manager: " + executor.shuffleManager);
    }
  }

  /**
   * Removes our metadata of all executors registered for the given application, and optionally
   * also deletes the local directories associated with the executors of that application in a
   * separate thread.
   *
   * It is not valid to call registerExecutor() for an executor with this appId after invoking
   * this method.
   */
  public void applicationRemoved(String appId, boolean cleanupLocalDirs) {
    logger.info("Application {} removed, cleanupLocalDirs = {}", appId, cleanupLocalDirs);
    Iterator<Map.Entry<AppExecId, ExecutorShuffleInfo>> it = executors.entrySet().iterator();
    while (it.hasNext()) {
      Map.Entry<AppExecId, ExecutorShuffleInfo> entry = it.next();
      AppExecId fullId = entry.getKey();
      final ExecutorShuffleInfo executor = entry.getValue();

      // Only touch executors associated with the appId that was removed.
      if (appId.equals(fullId.appId)) {
        it.remove();

        if (cleanupLocalDirs) {
          logger.info("Cleaning up executor {}'s {} local dirs", fullId, executor.localDirs.length);

          // Execute the actual deletion in a different thread, as it may take some time.
          directoryCleaner.execute(new Runnable() {
            @Override
            public void run() {
              deleteExecutorDirs(executor.localDirs);
            }
          });
        }
      }
    }
  }

  /**
   * Synchronously deletes each directory one at a time.
   * Should be executed in its own thread, as this may take a long time.
   */
  private void deleteExecutorDirs(String[] dirs) {
    for (String localDir : dirs) {
      try {
        JavaUtils.deleteRecursively(new File(localDir));
        logger.debug("Successfully cleaned up directory: " + localDir);
      } catch (Exception e) {
        logger.error("Failed to delete directory: " + localDir, e);
      }
    }
  }

  /**
   * Hash-based shuffle data is simply stored as one file per block.
   * This logic is from FileShuffleBlockManager.
   */
  // TODO: Support consolidated hash shuffle files
  private ManagedBuffer getHashBasedShuffleBlockData(ExecutorShuffleInfo executor, String blockId) {
    File shuffleFile = getFile(executor.localDirs, executor.subDirsPerLocalDir, blockId);
    return new FileSegmentManagedBuffer(shuffleFile, 0, shuffleFile.length());
  }

  /**
   * Sort-based shuffle data uses an index called "shuffle_ShuffleId_MapId_0.index" into a data file
   * called "shuffle_ShuffleId_MapId_0.data". This logic is from IndexShuffleBlockManager,
   * and the block id format is from ShuffleDataBlockId and ShuffleIndexBlockId.
   */
  private ManagedBuffer getSortBasedShuffleBlockData(
    ExecutorShuffleInfo executor, int shuffleId, int mapId, int reduceId) {
    File indexFile = getFile(executor.localDirs, executor.subDirsPerLocalDir,
      "shuffle_" + shuffleId + "_" + mapId + "_0.index");

    DataInputStream in = null;
    try {
      in = new DataInputStream(new FileInputStream(indexFile));
      in.skipBytes(reduceId * 8);
      long offset = in.readLong();
      long nextOffset = in.readLong();
      return new FileSegmentManagedBuffer(
        getFile(executor.localDirs, executor.subDirsPerLocalDir,
          "shuffle_" + shuffleId + "_" + mapId + "_0.data"),
        offset,
        nextOffset - offset);
    } catch (IOException e) {
      throw new RuntimeException("Failed to open file: " + indexFile, e);
    } finally {
      if (in != null) {
        JavaUtils.closeQuietly(in);
      }
    }
  }

  /**
   * Hashes a filename into the corresponding local directory, in a manner consistent with
   * Spark's DiskBlockManager.getFile().
   */
  @VisibleForTesting
  static File getFile(String[] localDirs, int subDirsPerLocalDir, String filename) {
    int hash = JavaUtils.nonNegativeHash(filename);
    String localDir = localDirs[hash % localDirs.length];
    int subDirId = (hash / localDirs.length) % subDirsPerLocalDir;
    return new File(new File(localDir, String.format("%02x", subDirId)), filename);
  }

  /** Simply encodes an executor's full ID, which is appId + execId. */
  private static class AppExecId {
    final String appId;
    final String execId;

    private AppExecId(String appId, String execId) {
      this.appId = appId;
      this.execId = execId;
    }

    @Override
    public boolean equals(Object o) {
      if (this == o) return true;
      if (o == null || getClass() != o.getClass()) return false;

      AppExecId appExecId = (AppExecId) o;
      return Objects.equal(appId, appExecId.appId) && Objects.equal(execId, appExecId.execId);
    }

    @Override
    public int hashCode() {
      return Objects.hashCode(appId, execId);
    }

    @Override
    public String toString() {
      return Objects.toStringHelper(this)
        .add("appId", appId)
        .add("execId", execId)
        .toString();
    }
  }
}
TOP

Related Classes of org.apache.spark.network.shuffle.ExternalShuffleBlockManager$AppExecId

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.