/**
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership. The ASF licenses this file
* to you under the Apache License, Version 2.0 (the
* "License"); you may not use this file except in compliance
* with the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.apache.hadoop.hbase.io.asyncfs;
import static io.netty.handler.timeout.IdleState.READER_IDLE;
import static io.netty.handler.timeout.IdleState.WRITER_IDLE;
import static org.apache.hadoop.hbase.io.asyncfs.FanOutOneBlockAsyncDFSOutputHelper.HEART_BEAT_SEQNO;
import static org.apache.hadoop.hbase.io.asyncfs.FanOutOneBlockAsyncDFSOutputHelper.READ_TIMEOUT;
import static org.apache.hadoop.hbase.io.asyncfs.FanOutOneBlockAsyncDFSOutputHelper.completeFile;
import static org.apache.hadoop.hbase.io.asyncfs.FanOutOneBlockAsyncDFSOutputHelper.endFileLease;
import static org.apache.hadoop.hbase.io.asyncfs.FanOutOneBlockAsyncDFSOutputHelper.getStatus;
import static org.apache.hadoop.hdfs.DFSConfigKeys.DFS_CLIENT_SOCKET_TIMEOUT_KEY;
import io.netty.buffer.ByteBuf;
import io.netty.buffer.ByteBufAllocator;
import io.netty.channel.Channel;
import io.netty.channel.ChannelHandler.Sharable;
import io.netty.channel.ChannelHandlerContext;
import io.netty.channel.EventLoop;
import io.netty.channel.SimpleChannelInboundHandler;
import io.netty.handler.codec.protobuf.ProtobufDecoder;
import io.netty.handler.codec.protobuf.ProtobufVarint32FrameDecoder;
import io.netty.handler.timeout.IdleStateEvent;
import io.netty.handler.timeout.IdleStateHandler;
import io.netty.util.concurrent.Promise;
import io.netty.util.concurrent.PromiseCombiner;
import java.io.IOException;
import java.nio.ByteBuffer;
import java.util.ArrayDeque;
import java.util.Collection;
import java.util.Collections;
import java.util.Deque;
import java.util.IdentityHashMap;
import java.util.List;
import java.util.Set;
import java.util.concurrent.CompletableFuture;
import java.util.concurrent.TimeUnit;
import java.util.function.Supplier;
import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.crypto.Encryptor;
import org.apache.hadoop.fs.Path;
import org.apache.hadoop.hbase.classification.InterfaceAudience;
import org.apache.hadoop.hbase.io.asyncfs.FanOutOneBlockAsyncDFSOutputHelper.CancelOnClose;
import org.apache.hadoop.hbase.util.CancelableProgressable;
import org.apache.hadoop.hbase.util.FSUtils;
import org.apache.hadoop.hdfs.DFSClient;
import org.apache.hadoop.hdfs.DistributedFileSystem;
import org.apache.hadoop.hdfs.protocol.ClientProtocol;
import org.apache.hadoop.hdfs.protocol.DatanodeInfo;
import org.apache.hadoop.hdfs.protocol.LocatedBlock;
import org.apache.hadoop.hdfs.protocol.datatransfer.PacketHeader;
import org.apache.hadoop.hdfs.protocol.datatransfer.PipelineAck;
import org.apache.hadoop.hdfs.protocol.proto.DataTransferProtos.PipelineAckProto;
import org.apache.hadoop.hdfs.protocol.proto.DataTransferProtos.Status;
import org.apache.hadoop.util.DataChecksum;
import com.google.common.annotations.VisibleForTesting;
/**
* An asynchronous HDFS output stream implementation which fans out data to datanode and only
* supports writing file with only one block.
* <p>
* Use the createOutput method in {@link FanOutOneBlockAsyncDFSOutputHelper} to create. The mainly
* usage of this class is implementing WAL, so we only expose a little HDFS configurations in the
* method. And we place it here under util package because we want to make it independent of WAL
* implementation thus easier to move it to HDFS project finally.
* <p>
* Note that, all connections to datanode will run in the same {@link EventLoop} which means we only
* need one thread here. But be careful, we do some blocking operations in {@link #close()} and
* {@link #recoverAndClose(CancelableProgressable)} methods, so do not call them inside
* {@link EventLoop}. And for {@link #write(byte[])} {@link #write(byte[], int, int)},
* {@link #buffered()} and {@link #flush(boolean)}, if you call them outside {@link EventLoop},
* there will be an extra context-switch.
* <p>
* Advantages compare to DFSOutputStream:
* <ol>
* <li>The fan out mechanism. This will reduce the latency.</li>
* <li>The asynchronous WAL could also run in the same EventLoop, we could just call write and flush
* inside the EventLoop thread, so generally we only have one thread to do all the things.</li>
* <li>Fail-fast when connection to datanode error. The WAL implementation could open new writer
* ASAP.</li>
* <li>We could benefit from netty's ByteBuf management mechanism.</li>
* </ol>
*/
@InterfaceAudience.Private
public class FanOutOneBlockAsyncDFSOutput implements AsyncFSOutput {
// The MAX_PACKET_SIZE is 16MB but it include the header size and checksum size. So here we set a
// smaller limit for data size.
private static final int MAX_DATA_LEN = 12 * 1024 * 1024;
private final Configuration conf;
private final FSUtils fsUtils;
private final DistributedFileSystem dfs;
private final DFSClient client;
private final ClientProtocol namenode;
private final String clientName;
private final String src;
private final long fileId;
private final LocatedBlock locatedBlock;
private final Encryptor encryptor;
private final EventLoop eventLoop;
private final List<Channel> datanodeList;
private final DataChecksum summer;
private final int maxDataLen;
private final ByteBufAllocator alloc;
private static final class Callback {
private final Promise<Void> promise;
private final long ackedLength;
private final Set<Channel> unfinishedReplicas;
public Callback(Promise<Void> promise, long ackedLength, Collection<Channel> replicas) {
this.promise = promise;
this.ackedLength = ackedLength;
if (replicas.isEmpty()) {
this.unfinishedReplicas = Collections.emptySet();
} else {
this.unfinishedReplicas =
Collections.newSetFromMap(new IdentityHashMap<Channel, Boolean>(replicas.size()));
this.unfinishedReplicas.addAll(replicas);
}
}
}
private final Deque<Callback> waitingAckQueue = new ArrayDeque<>();
// this could be different from acked block length because a packet can not start at the middle of
// a chunk.
private long nextPacketOffsetInBlock = 0L;
private long nextPacketSeqno = 0L;
private ByteBuf buf;
// buf's initial capacity - 4KB
private int capacity = 4 * 1024;
// LIMIT is 128MB
private static final int LIMIT = 128 * 1024 * 1024;
private enum State {
STREAMING, CLOSING, BROKEN, CLOSED
}
private State state;
private void completed(Channel channel) {
if (waitingAckQueue.isEmpty()) {
return;
}
for (Callback c : waitingAckQueue) {
if (c.unfinishedReplicas.remove(channel)) {
if (c.unfinishedReplicas.isEmpty()) {
c.promise.trySuccess(null);
// since we will remove the Callback entry from waitingAckQueue if its unfinishedReplicas
// is empty, so this could only happen at the head of waitingAckQueue, so we just call
// removeFirst here.
waitingAckQueue.removeFirst();
// also wake up flush requests which have the same length.
for (Callback cb; (cb = waitingAckQueue.peekFirst()) != null;) {
if (cb.ackedLength == c.ackedLength) {
cb.promise.trySuccess(null);
waitingAckQueue.removeFirst();
} else {
break;
}
}
}
return;
}
}
}
private void failed(Channel channel, Supplier<Throwable> errorSupplier) {
if (state == State.BROKEN || state == State.CLOSED) {
return;
}
if (state == State.CLOSING) {
Callback c = waitingAckQueue.peekFirst();
if (c == null || !c.unfinishedReplicas.contains(channel)) {
// nothing, the endBlock request has already finished.
return;
}
}
// disable further write, and fail all pending ack.
state = State.BROKEN;
Throwable error = errorSupplier.get();
waitingAckQueue.stream().forEach(c -> c.promise.tryFailure(error));
waitingAckQueue.clear();
datanodeList.forEach(ch -> ch.close());
}
@Sharable
private final class AckHandler extends SimpleChannelInboundHandler<PipelineAckProto> {
private final int timeoutMs;
public AckHandler(int timeoutMs) {
this.timeoutMs = timeoutMs;
}
@Override
protected void channelRead0(ChannelHandlerContext ctx, PipelineAckProto ack) throws Exception {
Status reply = getStatus(ack);
if (reply != Status.SUCCESS) {
failed(ctx.channel(), () -> new IOException("Bad response " + reply + " for block "
+ locatedBlock.getBlock() + " from datanode " + ctx.channel().remoteAddress()));
return;
}
if (PipelineAck.isRestartOOBStatus(reply)) {
failed(ctx.channel(), () -> new IOException("Restart response " + reply + " for block "
+ locatedBlock.getBlock() + " from datanode " + ctx.channel().remoteAddress()));
return;
}
if (ack.getSeqno() == HEART_BEAT_SEQNO) {
return;
}
completed(ctx.channel());
}
@Override
public void channelInactive(ChannelHandlerContext ctx) throws Exception {
failed(ctx.channel(),
() -> new IOException("Connection to " + ctx.channel().remoteAddress() + " closed"));
}
@Override
public void exceptionCaught(ChannelHandlerContext ctx, Throwable cause) throws Exception {
failed(ctx.channel(), () -> cause);
}
@Override
public void userEventTriggered(ChannelHandlerContext ctx, Object evt) throws Exception {
if (evt instanceof IdleStateEvent) {
IdleStateEvent e = (IdleStateEvent) evt;
if (e.state() == READER_IDLE) {
failed(ctx.channel(),
() -> new IOException("Timeout(" + timeoutMs + "ms) waiting for response"));
} else if (e.state() == WRITER_IDLE) {
PacketHeader heartbeat = new PacketHeader(4, 0, HEART_BEAT_SEQNO, false, 0, false);
int len = heartbeat.getSerializedSize();
ByteBuf buf = alloc.buffer(len);
heartbeat.putInBuffer(buf.nioBuffer(0, len));
buf.writerIndex(len);
ctx.channel().writeAndFlush(buf);
}
return;
}
super.userEventTriggered(ctx, evt);
}
}
private void setupReceiver(int timeoutMs) {
AckHandler ackHandler = new AckHandler(timeoutMs);
for (Channel ch : datanodeList) {
ch.pipeline().addLast(
new IdleStateHandler(timeoutMs, timeoutMs / 2, 0, TimeUnit.MILLISECONDS),
new ProtobufVarint32FrameDecoder(),
new ProtobufDecoder(PipelineAckProto.getDefaultInstance()), ackHandler);
ch.config().setAutoRead(true);
}
}
FanOutOneBlockAsyncDFSOutput(Configuration conf, FSUtils fsUtils, DistributedFileSystem dfs,
DFSClient client, ClientProtocol namenode, String clientName, String src, long fileId,
LocatedBlock locatedBlock, Encryptor encryptor, EventLoop eventLoop,
List<Channel> datanodeList, DataChecksum summer, ByteBufAllocator alloc) {
this.conf = conf;
this.fsUtils = fsUtils;
this.dfs = dfs;
this.client = client;
this.namenode = namenode;
this.fileId = fileId;
this.clientName = clientName;
this.src = src;
this.locatedBlock = locatedBlock;
this.encryptor = encryptor;
this.eventLoop = eventLoop;
this.datanodeList = datanodeList;
this.summer = summer;
this.maxDataLen = MAX_DATA_LEN - (MAX_DATA_LEN % summer.getBytesPerChecksum());
this.alloc = alloc;
this.buf = alloc.directBuffer(capacity);
this.state = State.STREAMING;
setupReceiver(conf.getInt(DFS_CLIENT_SOCKET_TIMEOUT_KEY, READ_TIMEOUT));
}
private void writeInt0(int i) {
buf.ensureWritable(4);
buf.writeInt(i);
}
@Override
public void writeInt(int i) {
if (eventLoop.inEventLoop()) {
writeInt0(i);
} else {
eventLoop.submit(() -> writeInt0(i));
}
}
private void write0(ByteBuffer bb) {
buf.ensureWritable(bb.remaining());
buf.writeBytes(bb);
}
@Override
public void write(ByteBuffer bb) {
if (eventLoop.inEventLoop()) {
write0(bb);
} else {
eventLoop.submit(() -> write0(bb));
}
}
@Override
public void write(byte[] b) {
write(b, 0, b.length);
}
private void write0(byte[] b, int off, int len) {
buf.ensureWritable(len);
buf.writeBytes(b, off, len);
}
@Override
public void write(byte[] b, int off, int len) {
if (eventLoop.inEventLoop()) {
write0(b, off, len);
} else {
eventLoop.submit(() -> write0(b, off, len)).syncUninterruptibly();
}
}
@Override
public int buffered() {
if (eventLoop.inEventLoop()) {
return buf.readableBytes();
} else {
return eventLoop.submit(() -> buf.readableBytes()).syncUninterruptibly().getNow().intValue();
}
}
@Override
public DatanodeInfo[] getPipeline() {
return locatedBlock.getLocations();
}
private Promise<Void> flushBuffer(ByteBuf dataBuf, long nextPacketOffsetInBlock,
boolean syncBlock) {
int dataLen = dataBuf.readableBytes();
int chunkLen = summer.getBytesPerChecksum();
int trailingPartialChunkLen = dataLen % chunkLen;
int numChecks = dataLen / chunkLen + (trailingPartialChunkLen != 0 ? 1 : 0);
int checksumLen = numChecks * summer.getChecksumSize();
ByteBuf checksumBuf = alloc.directBuffer(checksumLen);
summer.calculateChunkedSums(dataBuf.nioBuffer(), checksumBuf.nioBuffer(0, checksumLen));
checksumBuf.writerIndex(checksumLen);
PacketHeader header = new PacketHeader(4 + checksumLen + dataLen, nextPacketOffsetInBlock,
nextPacketSeqno, false, dataLen, syncBlock);
int headerLen = header.getSerializedSize();
ByteBuf headerBuf = alloc.buffer(headerLen);
header.putInBuffer(headerBuf.nioBuffer(0, headerLen));
headerBuf.writerIndex(headerLen);
long ackedLength = nextPacketOffsetInBlock + dataLen;
Promise<Void> promise = eventLoop.<Void> newPromise().addListener(future -> {
if (future.isSuccess()) {
locatedBlock.getBlock().setNumBytes(ackedLength);
}
});
waitingAckQueue.addLast(new Callback(promise, ackedLength, datanodeList));
for (Channel ch : datanodeList) {
ch.write(headerBuf.duplicate().retain());
ch.write(checksumBuf.duplicate().retain());
ch.writeAndFlush(dataBuf.duplicate().retain());
}
checksumBuf.release();
headerBuf.release();
dataBuf.release();
nextPacketSeqno++;
return promise;
}
private void flush0(CompletableFuture<Long> future, boolean syncBlock) {
if (state != State.STREAMING) {
future.completeExceptionally(new IOException("stream already broken"));
return;
}
int dataLen = buf.readableBytes();
if (encryptor != null) {
ByteBuf encryptBuf = alloc.directBuffer(dataLen);
try {
encryptor.encrypt(buf.nioBuffer(buf.readerIndex(), dataLen),
encryptBuf.nioBuffer(0, dataLen));
} catch (IOException e) {
encryptBuf.release();
future.completeExceptionally(e);
return;
}
encryptBuf.writerIndex(dataLen);
buf.release();
buf = encryptBuf;
}
long lengthAfterFlush = nextPacketOffsetInBlock + dataLen;
if (lengthAfterFlush == locatedBlock.getBlock().getNumBytes()) {
// no new data, just return
future.complete(locatedBlock.getBlock().getNumBytes());
return;
}
Callback c = waitingAckQueue.peekLast();
if (c != null && lengthAfterFlush == c.ackedLength) {
// just append it to the tail of waiting ack queue,, do not issue new hflush request.
waitingAckQueue.addLast(new Callback(eventLoop.<Void> newPromise().addListener(f -> {
if (f.isSuccess()) {
future.complete(lengthAfterFlush);
} else {
future.completeExceptionally(f.cause());
}
}), lengthAfterFlush, Collections.<Channel> emptyList()));
return;
}
Promise<Void> promise;
if (dataLen > maxDataLen) {
// We need to write out the data by multiple packets as the max packet allowed is 16M.
PromiseCombiner combiner = new PromiseCombiner();
long nextSubPacketOffsetInBlock = nextPacketOffsetInBlock;
for (int remaining = dataLen; remaining > 0;) {
int toWriteDataLen = Math.min(remaining, maxDataLen);
combiner.add(flushBuffer(buf.readRetainedSlice(toWriteDataLen), nextSubPacketOffsetInBlock,
syncBlock));
nextSubPacketOffsetInBlock += toWriteDataLen;
remaining -= toWriteDataLen;
}
promise = eventLoop.newPromise();
combiner.finish(promise);
} else {
promise = flushBuffer(buf.retain(), nextPacketOffsetInBlock, syncBlock);
}
promise.addListener(f -> {
if (f.isSuccess()) {
future.complete(lengthAfterFlush);
} else {
future.completeExceptionally(f.cause());
}
});
int trailingPartialChunkLen = dataLen % summer.getBytesPerChecksum();
ByteBuf newBuf = alloc.directBuffer(guess(dataLen)).ensureWritable(trailingPartialChunkLen);
if (trailingPartialChunkLen != 0) {
buf.readerIndex(dataLen - trailingPartialChunkLen).readBytes(newBuf, trailingPartialChunkLen);
}
buf.release();
this.buf = newBuf;
nextPacketOffsetInBlock += dataLen - trailingPartialChunkLen;
}
/**
* Flush the buffer out to datanodes.
* @param syncBlock will call hsync if true, otherwise hflush.
* @return A CompletableFuture that hold the acked length after flushing.
*/
public CompletableFuture<Long> flush(boolean syncBlock) {
CompletableFuture<Long> future = new CompletableFuture<>();
if (eventLoop.inEventLoop()) {
flush0(future, syncBlock);
} else {
eventLoop.execute(() -> flush0(future, syncBlock));
}
return future;
}
private void endBlock(Promise<Void> promise, long size) {
if (state != State.STREAMING) {
promise.tryFailure(new IOException("stream already broken"));
return;
}
if (!waitingAckQueue.isEmpty()) {
promise.tryFailure(new IllegalStateException("should call flush first before calling close"));
return;
}
state = State.CLOSING;
PacketHeader header = new PacketHeader(4, size, nextPacketSeqno, true, 0, false);
buf.release();
buf = null;
int headerLen = header.getSerializedSize();
ByteBuf headerBuf = alloc.directBuffer(headerLen);
header.putInBuffer(headerBuf.nioBuffer(0, headerLen));
headerBuf.writerIndex(headerLen);
waitingAckQueue.add(new Callback(promise, size, datanodeList));
datanodeList.forEach(ch -> ch.writeAndFlush(headerBuf.duplicate().retain()));
headerBuf.release();
}
/**
* The close method when error occurred. Now we just call recoverFileLease.
*/
@Override
public void recoverAndClose(CancelableProgressable reporter) throws IOException {
assert !eventLoop.inEventLoop();
datanodeList.forEach(ch -> ch.closeFuture().awaitUninterruptibly());
endFileLease(client, fileId);
fsUtils.recoverFileLease(dfs, new Path(src), conf,
reporter == null ? new CancelOnClose(client) : reporter);
}
/**
* End the current block and complete file at namenode. You should call
* {@link #recoverAndClose(CancelableProgressable)} if this method throws an exception.
*/
@Override
public void close() throws IOException {
assert !eventLoop.inEventLoop();
Promise<Void> promise = eventLoop.newPromise();
eventLoop.execute(() -> endBlock(promise, nextPacketOffsetInBlock + buf.readableBytes()));
promise.addListener(f -> datanodeList.forEach(ch -> ch.close())).syncUninterruptibly();
datanodeList.forEach(ch -> ch.closeFuture().awaitUninterruptibly());
completeFile(client, namenode, src, clientName, locatedBlock.getBlock(), fileId);
}
@VisibleForTesting
int guess(int bytesWritten) {
// if the bytesWritten is greater than the current capacity
// always increase the capacity in powers of 2.
if (bytesWritten > this.capacity) {
// Ensure we don't cross the LIMIT
if ((this.capacity << 1) <= LIMIT) {
// increase the capacity in the range of power of 2
this.capacity = this.capacity << 1;
}
} else {
// if we see that the bytesWritten is lesser we could again decrease
// the capacity by dividing it by 2 if the bytesWritten is satisfied by
// that reduction
if ((this.capacity >> 1) >= bytesWritten) {
this.capacity = this.capacity >> 1;
}
}
return this.capacity;
}
}