001/**
002 * Licensed to the Apache Software Foundation (ASF) under one
003 * or more contributor license agreements.  See the NOTICE file
004 * distributed with this work for additional information
005 * regarding copyright ownership.  The ASF licenses this file
006 * to you under the Apache License, Version 2.0 (the
007 * "License"); you may not use this file except in compliance
008 * with the License.  You may obtain a copy of the License at
009 *
010 *     http://www.apache.org/licenses/LICENSE-2.0
011 *
012 * Unless required by applicable law or agreed to in writing, software
013 * distributed under the License is distributed on an "AS IS" BASIS,
014 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
015 * See the License for the specific language governing permissions and
016 * limitations under the License.
017 */
018package org.apache.hadoop.hbase.io.asyncfs;
019
020import static org.apache.hadoop.hbase.io.asyncfs.FanOutOneBlockAsyncDFSOutputHelper.HEART_BEAT_SEQNO;
021import static org.apache.hadoop.hbase.io.asyncfs.FanOutOneBlockAsyncDFSOutputHelper.READ_TIMEOUT;
022import static org.apache.hadoop.hbase.io.asyncfs.FanOutOneBlockAsyncDFSOutputHelper.completeFile;
023import static org.apache.hadoop.hbase.io.asyncfs.FanOutOneBlockAsyncDFSOutputHelper.endFileLease;
024import static org.apache.hadoop.hbase.io.asyncfs.FanOutOneBlockAsyncDFSOutputHelper.getStatus;
025import static org.apache.hbase.thirdparty.io.netty.handler.timeout.IdleState.READER_IDLE;
026import static org.apache.hbase.thirdparty.io.netty.handler.timeout.IdleState.WRITER_IDLE;
027import static org.apache.hadoop.hdfs.DFSConfigKeys.DFS_CLIENT_SOCKET_TIMEOUT_KEY;
028
029import java.io.IOException;
030import java.io.InterruptedIOException;
031import java.nio.ByteBuffer;
032import java.util.Collection;
033import java.util.Collections;
034import java.util.Iterator;
035import java.util.List;
036import java.util.Set;
037import java.util.concurrent.CompletableFuture;
038import java.util.concurrent.ConcurrentHashMap;
039import java.util.concurrent.ConcurrentLinkedDeque;
040import java.util.concurrent.ExecutionException;
041import java.util.concurrent.TimeUnit;
042import java.util.function.Supplier;
043
044import org.apache.hadoop.conf.Configuration;
045import org.apache.hadoop.crypto.Encryptor;
046import org.apache.hadoop.fs.Path;
047import org.apache.hadoop.hbase.io.asyncfs.FanOutOneBlockAsyncDFSOutputHelper.CancelOnClose;
048import org.apache.hadoop.hbase.util.CancelableProgressable;
049import org.apache.hadoop.hbase.util.RecoverLeaseFSUtils;
050import org.apache.hadoop.hdfs.DFSClient;
051import org.apache.hadoop.hdfs.DistributedFileSystem;
052import org.apache.hadoop.hdfs.protocol.ClientProtocol;
053import org.apache.hadoop.hdfs.protocol.DatanodeInfo;
054import org.apache.hadoop.hdfs.protocol.ExtendedBlock;
055import org.apache.hadoop.hdfs.protocol.LocatedBlock;
056import org.apache.hadoop.hdfs.protocol.datatransfer.PacketHeader;
057import org.apache.hadoop.hdfs.protocol.datatransfer.PipelineAck;
058import org.apache.hadoop.hdfs.protocol.proto.DataTransferProtos.PipelineAckProto;
059import org.apache.hadoop.hdfs.protocol.proto.DataTransferProtos.Status;
060import org.apache.hadoop.util.DataChecksum;
061import org.apache.yetus.audience.InterfaceAudience;
062
063import org.apache.hbase.thirdparty.com.google.common.base.Preconditions;
064import org.apache.hbase.thirdparty.com.google.common.base.Throwables;
065import org.apache.hbase.thirdparty.io.netty.buffer.ByteBuf;
066import org.apache.hbase.thirdparty.io.netty.buffer.ByteBufAllocator;
067import org.apache.hbase.thirdparty.io.netty.channel.Channel;
068import org.apache.hbase.thirdparty.io.netty.channel.ChannelHandler.Sharable;
069import org.apache.hbase.thirdparty.io.netty.channel.ChannelHandlerContext;
070import org.apache.hbase.thirdparty.io.netty.channel.ChannelId;
071import org.apache.hbase.thirdparty.io.netty.channel.SimpleChannelInboundHandler;
072import org.apache.hbase.thirdparty.io.netty.handler.codec.protobuf.ProtobufVarint32FrameDecoder;
073import org.apache.hbase.thirdparty.io.netty.handler.timeout.IdleStateEvent;
074import org.apache.hbase.thirdparty.io.netty.handler.timeout.IdleStateHandler;
075
076/**
077 * An asynchronous HDFS output stream implementation which fans out data to datanode and only
078 * supports writing file with only one block.
079 * <p>
080 * Use the createOutput method in {@link FanOutOneBlockAsyncDFSOutputHelper} to create. The mainly
081 * usage of this class is implementing WAL, so we only expose a little HDFS configurations in the
082 * method. And we place it here under io package because we want to make it independent of WAL
083 * implementation thus easier to move it to HDFS project finally.
084 * <p>
085 * Note that, although we support pipelined flush, i.e, write new data and then flush before the
086 * previous flush succeeds, the implementation is not thread safe, so you should not call its
087 * methods concurrently.
088 * <p>
089 * Advantages compare to DFSOutputStream:
090 * <ol>
091 * <li>The fan out mechanism. This will reduce the latency.</li>
092 * <li>Fail-fast when connection to datanode error. The WAL implementation could open new writer
093 * ASAP.</li>
094 * <li>We could benefit from netty's ByteBuf management mechanism.</li>
095 * </ol>
096 */
097@InterfaceAudience.Private
098public class FanOutOneBlockAsyncDFSOutput implements AsyncFSOutput {
099
100  // The MAX_PACKET_SIZE is 16MB but it include the header size and checksum size. So here we set a
101  // smaller limit for data size.
102  private static final int MAX_DATA_LEN = 12 * 1024 * 1024;
103
104  private final Configuration conf;
105
106  private final DistributedFileSystem dfs;
107
108  private final DFSClient client;
109
110  private final ClientProtocol namenode;
111
112  private final String clientName;
113
114  private final String src;
115
116  private final long fileId;
117
118  private final ExtendedBlock block;
119
120  private final DatanodeInfo[] locations;
121
122  private final Encryptor encryptor;
123
124  private final List<Channel> datanodeList;
125
126  private final DataChecksum summer;
127
128  private final int maxDataLen;
129
130  private final ByteBufAllocator alloc;
131
132  private static final class Callback {
133
134    private final CompletableFuture<Long> future;
135
136    private final long ackedLength;
137
138    // should be backed by a thread safe collection
139    private final Set<ChannelId> unfinishedReplicas;
140
141    public Callback(CompletableFuture<Long> future, long ackedLength,
142        Collection<Channel> replicas) {
143      this.future = future;
144      this.ackedLength = ackedLength;
145      if (replicas.isEmpty()) {
146        this.unfinishedReplicas = Collections.emptySet();
147      } else {
148        this.unfinishedReplicas =
149          Collections.newSetFromMap(new ConcurrentHashMap<ChannelId, Boolean>(replicas.size()));
150        replicas.stream().map(c -> c.id()).forEachOrdered(unfinishedReplicas::add);
151      }
152    }
153  }
154
155  private final ConcurrentLinkedDeque<Callback> waitingAckQueue = new ConcurrentLinkedDeque<>();
156
157  private volatile long ackedBlockLength = 0L;
158
159  // this could be different from acked block length because a packet can not start at the middle of
160  // a chunk.
161  private long nextPacketOffsetInBlock = 0L;
162
163  // the length of the trailing partial chunk, this is because the packet start offset must be
164  // aligned with the length of checksum chunk so we need to resend the same data.
165  private int trailingPartialChunkLength = 0;
166
167  private long nextPacketSeqno = 0L;
168
169  private ByteBuf buf;
170
171  private final SendBufSizePredictor sendBufSizePRedictor = new SendBufSizePredictor();
172
173  // State for connections to DN
174  private enum State {
175    STREAMING, CLOSING, BROKEN, CLOSED
176  }
177
178  private volatile State state;
179
180  // all lock-free to make it run faster
181  private void completed(Channel channel) {
182    for (Iterator<Callback> iter = waitingAckQueue.iterator(); iter.hasNext();) {
183      Callback c = iter.next();
184      // if the current unfinished replicas does not contain us then it means that we have already
185      // acked this one, let's iterate to find the one we have not acked yet.
186      if (c.unfinishedReplicas.remove(channel.id())) {
187        if (c.unfinishedReplicas.isEmpty()) {
188          // we need to remove first before complete the future. It is possible that after we
189          // complete the future the upper layer will call close immediately before we remove the
190          // entry from waitingAckQueue and lead to an IllegalStateException. And also set the
191          // ackedBlockLength first otherwise we may use a wrong length to commit the block. This
192          // may lead to multiple remove and assign but is OK. The semantic of iter.remove is
193          // removing the entry returned by calling previous next, so if the entry has already been
194          // removed then it is a no-op, and for the assign, the values are the same so no problem.
195          iter.remove();
196          ackedBlockLength = c.ackedLength;
197          // the future.complete check is to confirm that we are the only one who grabbed the work,
198          // otherwise just give up and return.
199          if (c.future.complete(c.ackedLength)) {
200            // also wake up flush requests which have the same length.
201            while (iter.hasNext()) {
202              Callback maybeDummyCb = iter.next();
203              if (maybeDummyCb.ackedLength == c.ackedLength) {
204                iter.remove();
205                maybeDummyCb.future.complete(c.ackedLength);
206              } else {
207                break;
208              }
209            }
210          }
211        }
212        return;
213      }
214    }
215  }
216
217  // this usually does not happen which means it is not on the critical path so make it synchronized
218  // so that the implementation will not burn up our brain as there are multiple state changes and
219  // checks.
220  private synchronized void failed(Channel channel, Supplier<Throwable> errorSupplier) {
221    if (state == State.BROKEN || state == State.CLOSED) {
222      return;
223    }
224    if (state == State.CLOSING) {
225      Callback c = waitingAckQueue.peekFirst();
226      if (c == null || !c.unfinishedReplicas.contains(channel.id())) {
227        // nothing, the endBlock request has already finished.
228        return;
229      }
230    }
231    // disable further write, and fail all pending ack.
232    state = State.BROKEN;
233    Throwable error = errorSupplier.get();
234    for (Iterator<Callback> iter = waitingAckQueue.iterator(); iter.hasNext();) {
235      Callback c = iter.next();
236      // find the first sync request which we have not acked yet and fail all the request after it.
237      if (!c.unfinishedReplicas.contains(channel.id())) {
238        continue;
239      }
240      for (;;) {
241        c.future.completeExceptionally(error);
242        if (!iter.hasNext()) {
243          break;
244        }
245        c = iter.next();
246      }
247      break;
248    }
249    datanodeList.forEach(ch -> ch.close());
250  }
251
252  @Sharable
253  private final class AckHandler extends SimpleChannelInboundHandler<PipelineAckProto> {
254
255    private final int timeoutMs;
256
257    public AckHandler(int timeoutMs) {
258      this.timeoutMs = timeoutMs;
259    }
260
261    @Override
262    protected void channelRead0(ChannelHandlerContext ctx, PipelineAckProto ack) throws Exception {
263      Status reply = getStatus(ack);
264      if (reply != Status.SUCCESS) {
265        failed(ctx.channel(), () -> new IOException("Bad response " + reply + " for block " +
266          block + " from datanode " + ctx.channel().remoteAddress()));
267        return;
268      }
269      if (PipelineAck.isRestartOOBStatus(reply)) {
270        failed(ctx.channel(), () -> new IOException("Restart response " + reply + " for block " +
271          block + " from datanode " + ctx.channel().remoteAddress()));
272        return;
273      }
274      if (ack.getSeqno() == HEART_BEAT_SEQNO) {
275        return;
276      }
277      completed(ctx.channel());
278    }
279
280    @Override
281    public void channelInactive(ChannelHandlerContext ctx) throws Exception {
282      if (state == State.CLOSED) {
283        return;
284      }
285      failed(ctx.channel(),
286        () -> new IOException("Connection to " + ctx.channel().remoteAddress() + " closed"));
287    }
288
289    @Override
290    public void exceptionCaught(ChannelHandlerContext ctx, Throwable cause) throws Exception {
291      failed(ctx.channel(), () -> cause);
292    }
293
294    @Override
295    public void userEventTriggered(ChannelHandlerContext ctx, Object evt) throws Exception {
296      if (evt instanceof IdleStateEvent) {
297        IdleStateEvent e = (IdleStateEvent) evt;
298        if (e.state() == READER_IDLE) {
299          failed(ctx.channel(),
300            () -> new IOException("Timeout(" + timeoutMs + "ms) waiting for response"));
301        } else if (e.state() == WRITER_IDLE) {
302          PacketHeader heartbeat = new PacketHeader(4, 0, HEART_BEAT_SEQNO, false, 0, false);
303          int len = heartbeat.getSerializedSize();
304          ByteBuf buf = alloc.buffer(len);
305          heartbeat.putInBuffer(buf.nioBuffer(0, len));
306          buf.writerIndex(len);
307          ctx.channel().writeAndFlush(buf);
308        }
309        return;
310      }
311      super.userEventTriggered(ctx, evt);
312    }
313  }
314
315  private void setupReceiver(int timeoutMs) {
316    AckHandler ackHandler = new AckHandler(timeoutMs);
317    for (Channel ch : datanodeList) {
318      ch.pipeline().addLast(
319        new IdleStateHandler(timeoutMs, timeoutMs / 2, 0, TimeUnit.MILLISECONDS),
320        new ProtobufVarint32FrameDecoder(),
321        new ProtobufDecoder(PipelineAckProto.getDefaultInstance()), ackHandler);
322      ch.config().setAutoRead(true);
323    }
324  }
325
326  FanOutOneBlockAsyncDFSOutput(Configuration conf,DistributedFileSystem dfs,
327      DFSClient client, ClientProtocol namenode, String clientName, String src, long fileId,
328      LocatedBlock locatedBlock, Encryptor encryptor, List<Channel> datanodeList,
329      DataChecksum summer, ByteBufAllocator alloc) {
330    this.conf = conf;
331    this.dfs = dfs;
332    this.client = client;
333    this.namenode = namenode;
334    this.fileId = fileId;
335    this.clientName = clientName;
336    this.src = src;
337    this.block = locatedBlock.getBlock();
338    this.locations = locatedBlock.getLocations();
339    this.encryptor = encryptor;
340    this.datanodeList = datanodeList;
341    this.summer = summer;
342    this.maxDataLen = MAX_DATA_LEN - (MAX_DATA_LEN % summer.getBytesPerChecksum());
343    this.alloc = alloc;
344    this.buf = alloc.directBuffer(sendBufSizePRedictor.initialSize());
345    this.state = State.STREAMING;
346    setupReceiver(conf.getInt(DFS_CLIENT_SOCKET_TIMEOUT_KEY, READ_TIMEOUT));
347  }
348
349  @Override
350  public void writeInt(int i) {
351    buf.ensureWritable(4);
352    buf.writeInt(i);
353  }
354
355  @Override
356  public void write(ByteBuffer bb) {
357    buf.ensureWritable(bb.remaining());
358    buf.writeBytes(bb);
359  }
360
361  @Override
362  public void write(byte[] b) {
363    write(b, 0, b.length);
364  }
365
366  @Override
367  public void write(byte[] b, int off, int len) {
368    buf.ensureWritable(len);
369    buf.writeBytes(b, off, len);
370  }
371
372  @Override
373  public int buffered() {
374    return buf.readableBytes();
375  }
376
377  @Override
378  public DatanodeInfo[] getPipeline() {
379    return locations;
380  }
381
382  private void flushBuffer(CompletableFuture<Long> future, ByteBuf dataBuf,
383      long nextPacketOffsetInBlock, boolean syncBlock) {
384    int dataLen = dataBuf.readableBytes();
385    int chunkLen = summer.getBytesPerChecksum();
386    int trailingPartialChunkLen = dataLen % chunkLen;
387    int numChecks = dataLen / chunkLen + (trailingPartialChunkLen != 0 ? 1 : 0);
388    int checksumLen = numChecks * summer.getChecksumSize();
389    ByteBuf checksumBuf = alloc.directBuffer(checksumLen);
390    summer.calculateChunkedSums(dataBuf.nioBuffer(), checksumBuf.nioBuffer(0, checksumLen));
391    checksumBuf.writerIndex(checksumLen);
392    PacketHeader header = new PacketHeader(4 + checksumLen + dataLen, nextPacketOffsetInBlock,
393        nextPacketSeqno, false, dataLen, syncBlock);
394    int headerLen = header.getSerializedSize();
395    ByteBuf headerBuf = alloc.buffer(headerLen);
396    header.putInBuffer(headerBuf.nioBuffer(0, headerLen));
397    headerBuf.writerIndex(headerLen);
398    Callback c = new Callback(future, nextPacketOffsetInBlock + dataLen, datanodeList);
399    waitingAckQueue.addLast(c);
400    // recheck again after we pushed the callback to queue
401    if (state != State.STREAMING && waitingAckQueue.peekFirst() == c) {
402      future.completeExceptionally(new IOException("stream already broken"));
403      // it's the one we have just pushed or just a no-op
404      waitingAckQueue.removeFirst();
405      return;
406    }
407    datanodeList.forEach(ch -> {
408      ch.write(headerBuf.retainedDuplicate());
409      ch.write(checksumBuf.retainedDuplicate());
410      ch.writeAndFlush(dataBuf.retainedDuplicate());
411    });
412    checksumBuf.release();
413    headerBuf.release();
414    dataBuf.release();
415    nextPacketSeqno++;
416  }
417
418  private void flush0(CompletableFuture<Long> future, boolean syncBlock) {
419    if (state != State.STREAMING) {
420      future.completeExceptionally(new IOException("stream already broken"));
421      return;
422    }
423    int dataLen = buf.readableBytes();
424    if (dataLen == trailingPartialChunkLength) {
425      // no new data
426      long lengthAfterFlush = nextPacketOffsetInBlock + dataLen;
427      Callback lastFlush = waitingAckQueue.peekLast();
428      if (lastFlush != null) {
429        Callback c = new Callback(future, lengthAfterFlush, Collections.emptyList());
430        waitingAckQueue.addLast(c);
431        // recheck here if we have already removed the previous callback from the queue
432        if (waitingAckQueue.peekFirst() == c) {
433          // all previous callbacks have been removed
434          // notice that this does mean we will always win here because the background thread may
435          // have already started to mark the future here as completed in the completed or failed
436          // methods but haven't removed it from the queue yet. That's also why the removeFirst
437          // call below may be a no-op.
438          if (state != State.STREAMING) {
439            future.completeExceptionally(new IOException("stream already broken"));
440          } else {
441            future.complete(lengthAfterFlush);
442          }
443          // it's the one we have just pushed or just a no-op
444          waitingAckQueue.removeFirst();
445        }
446      } else {
447        // we must have acked all the data so the ackedBlockLength must be same with
448        // lengthAfterFlush
449        future.complete(lengthAfterFlush);
450      }
451      return;
452    }
453
454    if (encryptor != null) {
455      ByteBuf encryptBuf = alloc.directBuffer(dataLen);
456      buf.readBytes(encryptBuf, trailingPartialChunkLength);
457      int toEncryptLength = dataLen - trailingPartialChunkLength;
458      try {
459        encryptor.encrypt(buf.nioBuffer(trailingPartialChunkLength, toEncryptLength),
460          encryptBuf.nioBuffer(trailingPartialChunkLength, toEncryptLength));
461      } catch (IOException e) {
462        encryptBuf.release();
463        future.completeExceptionally(e);
464        return;
465      }
466      encryptBuf.writerIndex(dataLen);
467      buf.release();
468      buf = encryptBuf;
469    }
470
471    if (dataLen > maxDataLen) {
472      // We need to write out the data by multiple packets as the max packet allowed is 16M.
473      long nextSubPacketOffsetInBlock = nextPacketOffsetInBlock;
474      for (int remaining = dataLen;;) {
475        if (remaining < maxDataLen) {
476          flushBuffer(future, buf.readRetainedSlice(remaining), nextSubPacketOffsetInBlock,
477            syncBlock);
478          break;
479        } else {
480          flushBuffer(new CompletableFuture<>(), buf.readRetainedSlice(maxDataLen),
481            nextSubPacketOffsetInBlock, syncBlock);
482          remaining -= maxDataLen;
483          nextSubPacketOffsetInBlock += maxDataLen;
484        }
485      }
486    } else {
487      flushBuffer(future, buf.retain(), nextPacketOffsetInBlock, syncBlock);
488    }
489    trailingPartialChunkLength = dataLen % summer.getBytesPerChecksum();
490    ByteBuf newBuf = alloc.directBuffer(sendBufSizePRedictor.guess(dataLen))
491        .ensureWritable(trailingPartialChunkLength);
492    if (trailingPartialChunkLength != 0) {
493      buf.readerIndex(dataLen - trailingPartialChunkLength).readBytes(newBuf,
494        trailingPartialChunkLength);
495    }
496    buf.release();
497    this.buf = newBuf;
498    nextPacketOffsetInBlock += dataLen - trailingPartialChunkLength;
499  }
500
501  /**
502   * Flush the buffer out to datanodes.
503   * @param syncBlock will call hsync if true, otherwise hflush.
504   * @return A CompletableFuture that hold the acked length after flushing.
505   */
506  @Override
507  public CompletableFuture<Long> flush(boolean syncBlock) {
508    CompletableFuture<Long> future = new CompletableFuture<>();
509    flush0(future, syncBlock);
510    return future;
511  }
512
513  private void endBlock() throws IOException {
514    Preconditions.checkState(waitingAckQueue.isEmpty(),
515      "should call flush first before calling close");
516    if (state != State.STREAMING) {
517      throw new IOException("stream already broken");
518    }
519    state = State.CLOSING;
520    long finalizedLength = ackedBlockLength;
521    PacketHeader header = new PacketHeader(4, finalizedLength, nextPacketSeqno, true, 0, false);
522    buf.release();
523    buf = null;
524    int headerLen = header.getSerializedSize();
525    ByteBuf headerBuf = alloc.directBuffer(headerLen);
526    header.putInBuffer(headerBuf.nioBuffer(0, headerLen));
527    headerBuf.writerIndex(headerLen);
528    CompletableFuture<Long> future = new CompletableFuture<>();
529    waitingAckQueue.add(new Callback(future, finalizedLength, datanodeList));
530    datanodeList.forEach(ch -> ch.writeAndFlush(headerBuf.retainedDuplicate()));
531    headerBuf.release();
532    try {
533      future.get();
534    } catch (InterruptedException e) {
535      throw (IOException) new InterruptedIOException().initCause(e);
536    } catch (ExecutionException e) {
537      Throwable cause = e.getCause();
538      Throwables.propagateIfPossible(cause, IOException.class);
539      throw new IOException(cause);
540    }
541  }
542
543  /**
544   * The close method when error occurred. Now we just call recoverFileLease.
545   */
546  @Override
547  public void recoverAndClose(CancelableProgressable reporter) throws IOException {
548    if (buf != null) {
549      buf.release();
550      buf = null;
551    }
552    datanodeList.forEach(ch -> ch.close());
553    datanodeList.forEach(ch -> ch.closeFuture().awaitUninterruptibly());
554    endFileLease(client, fileId);
555    RecoverLeaseFSUtils.recoverFileLease(dfs, new Path(src), conf,
556      reporter == null ? new CancelOnClose(client) : reporter);
557  }
558
559  /**
560   * End the current block and complete file at namenode. You should call
561   * {@link #recoverAndClose(CancelableProgressable)} if this method throws an exception.
562   */
563  @Override
564  public void close() throws IOException {
565    endBlock();
566    state = State.CLOSED;
567    datanodeList.forEach(ch -> ch.close());
568    datanodeList.forEach(ch -> ch.closeFuture().awaitUninterruptibly());
569    block.setNumBytes(ackedBlockLength);
570    completeFile(client, namenode, src, clientName, block, fileId);
571  }
572
573  @Override
574  public boolean isBroken() {
575    return state == State.BROKEN;
576  }
577}