这里暂且仅关注连接池的实现，部分涉及Netty的channel输入输出处理、哈希轮定时器算法、事件轮询方式的区别，又或者信号量的使用等等，以后有机会会单独拿出来详解。
由于是基Netty的实现的，所以连接池实际上就是对channel的管理控制，有趣的是整个管理只用到了信号量+一个定时检测器，略微复杂的也就定时检测的逻辑，其实现方式简单且很好理解，不像httpclient里各种队列各种信号量难以理解。
先上一个简单的例子，事实上使用起来也不复杂。
public class HttpTest {
static AsyncHttpClient asyncHttpClient = Dsl
//实例化所有池和检测器
.asyncHttpClient(
Dsl.config()
.setMaxConnections(500)
.setMaxConnectionsPerHost(50)
.setPooledConnectionIdleTimeout(6000)
.setConnectionTtl(500)
.setIoThreadsCount(100)
.setConnectTimeout(60000)
.setUseNativeTransport(
System.getProperty("os.name").toLowerCase().indexOf("linux") > 0));
public static void main(String[] args) throws Exception {
List<Param> params = new ArrayList<>();
params.add(new Param("keyfrom", "XXX"));
asyncHttpClient
.prepareGet("http://fanyi.youdao.com/openapi.do")
.addQueryParams(params)
//这里进入发送请求阶段
.execute()
.toCompletableFuture()
//超时报错，或请求异常，做容错处理，抛出一个Response
.exceptionally(t -> {
return new Response() {...};
})
.thenAccept(rep -> System.out.println("RESPONSE BODY" + rep.getResponseBody()));
}
}先看看DefaultAsyncHttpClientConfig类的配置参数，这里只列出本文所需要的参数。有一点值得提一下，如果想了解Java怎么像clojure或者scala一样创建不可变对象，可以看看这个类的写法。
// timeouts
//连接超时
private final int connectTimeout;
//请求超时
private final int requestTimeout;
//读取超时，含于请求时间
private final int readTimeout;
//关闭Client前的静默时间
private final int shutdownQuietPeriod;
//关闭超时
private final int shutdownTimeout;
// keep-alive
private final boolean keepAlive;
//连接池空闲时间
private final int pooledConnectionIdleTimeout;
//定时清理空闲连接的时间
private final int connectionPoolCleanerPeriod;
//连接存活时间
private final int connectionTtl;
//最大连接数
private final int maxConnections;
//每个路由的最大连接数
private final int maxConnectionsPerHost;
//用于channel超时处理
private final ChannelPool channelPool;
private final KeepAliveStrategy keepAliveStrategy;
// internals
private final String threadPoolName;
private final int httpClientCodecMaxInitialLineLength;
private final int httpClientCodecMaxHeaderSize;
private final int httpClientCodecMaxChunkSize;
private final int chunkedFileChunkSize;
private final int webSocketMaxBufferSize;
private final int webSocketMaxFrameSize;
private final Map<ChannelOption<Object>, Object> channelOptions;
//时间轮询组类型
private final EventLoopGroup eventLoopGroup;
//是否用epoll，仅linux系统支持
private final boolean useNativeTransport;
//用于Timeout处理，建议用默认Netty的HashedWheelTimer
private final Timer nettyTimer;
private final ThreadFactory threadFactory;
private final AdditionalChannelInitializer httpAdditionalChannelInitializer;
private final AdditionalChannelInitializer wsAdditionalChannelInitializer;
private final ResponseBodyPartFactory responseBodyPartFactory;
//其实就是EventLoopGroup指定的线程数
private final int ioThreadsCount;就从这里开始，开头主要实例化ChannelManager和NettyRequestSender以及Timer三个重要组件，NettyRequestSender用于发送请求以及向ChannelManager索取channel使用权，Timer则负责另外两个组件给他的检测任务。
public final class Dsl {
public static AsyncHttpClient asyncHttpClient(DefaultAsyncHttpClientConfig.Builder configBuilder) {
//默认客户端
return new DefaultAsyncHttpClient(configBuilder.build());
}
//...
//...
}
public class DefaultAsyncHttpClient implements AsyncHttpClient {
private final AsyncHttpClientConfig config;
private final AtomicBoolean closed = new AtomicBoolean(false);
//Netty初始化的起点，Bootstrap与Channel池管理都在这里声明
private final ChannelManager channelManager;
//发送请求，以及向ChannelManager索取channel的使用权
private final NettyRequestSender requestSender;
private final boolean allowStopNettyTimer;
//定时做超时处理
private final Timer nettyTimer;
public DefaultAsyncHttpClient(AsyncHttpClientConfig config) {
this.config = config;
allowStopNettyTimer = config.getNettyTimer() == null;
//默认使用HashedWheelTimer
nettyTimer = allowStopNettyTimer ? newNettyTimer() :config.getNettyTimer();
//下面两个是重点！！！
channelManager = new ChannelManager(config, nettyTimer);
requestSender = new NettyRequestSender(config, channelManager,nettyTimer, new AsyncHttpClientState(closed));
//给Bootstraps配置解析器，以及处理接收服务端发送的消息的处理器
channelManager.configureBootstraps(requestSender);
}
private Timer newNettyTimer() {
HashedWheelTimer timer = new HashedWheelTimer();
timer.start();
return timer;
}
}这里是重中之重，网络配置、连接池、IO线程池、轮询方式都是在这儿产生的。
public class ChannelManager {
private final AsyncHttpClientConfig config;
private final EventLoopGroup eventLoopGroup;
private final boolean allowReleaseEventLoopGroup;
private final Bootstrap httpBootstrap;
private final Bootstrap wsBootstrap;
private final long handshakeTimeout;
private final IOException tooManyConnections;
private final IOException tooManyConnectionsPerHost;
//定时清理不符合标准的channel
private final ChannelPool channelPool;
//netty自带的用于管理channel的管理器
private final ChannelGroup openChannels;
private final ConcurrentHashMap<Channel, Object> channelId2PartitionKey = new ConcurrentHashMap<>();
//是否开启最大总连接数
private final boolean maxTotalConnectionsEnabled;
//最大连接数
private final Semaphore freeChannels;
//是否开启每个路由最大连接数
private final boolean maxConnectionsPerHostEnabled;
//每个路由最大连接数
private final ConcurrentHashMap<Object, Semaphore> freeChannelsPerHost = new ConcurrentHashMap<>();
private AsyncHttpClientHandler wsHandler;
public ChannelManager(final AsyncHttpClientConfig config, Timer nettyTimer) {
this.config = config;
//忽略一小段关于ssl的
//ChannelPool是用于检测已经实例化的channel的健康状况，如果不合格会直接close掉
ChannelPool channelPool = config.getChannelPool();
if (channelPool == null) {
if (config.isKeepAlive()) {
//这是默认使用的，事实上多数场景不需要我们自己实现
channelPool = new DefaultChannelPool(config, nettyTimer);
} else {
channelPool = NoopChannelPool.INSTANCE;
}
}
this.channelPool = channelPool;
tooManyConnections = trimStackTrace(new TooManyConnectionsException(config.getMaxConnections()));
tooManyConnectionsPerHost = trimStackTrace(new TooManyConnectionsPerHostException(config.getMaxConnectionsPerHost()));
maxTotalConnectionsEnabled = config.getMaxConnections() > 0;
maxConnectionsPerHostEnabled = config.getMaxConnectionsPerHost() > 0;
if (maxTotalConnectionsEnabled || maxConnectionsPerHostEnabled) {
//管理已经被实例化的channel
openChannels = new DefaultChannelGroup("asyncHttpClient", GlobalEventExecutor.INSTANCE) {
//重写删除方法，因为删除channel时，需要释放信号量
@Override
public boolean remove(Object o) {
boolean removed = super.remove(o);
if (removed) {
//释放总连接池的信号量
if (maxTotalConnectionsEnabled)
freeChannels.release();
//释放路由连接池的信号量
if (maxConnectionsPerHostEnabled) {
Object partitionKey = channelId2PartitionKey.remove(Channel.class.cast(o));
if (partitionKey != null) {
Semaphore hostFreeChannels = freeChannelsPerHost.get(partitionKey);
if (hostFreeChannels != null)
hostFreeChannels.release();
}
}
}
return removed;
}
};
//信号量数为最大连接数
freeChannels = new Semaphore(config.getMaxConnections());
} else {
openChannels = new DefaultChannelGroup("asyncHttpClient", GlobalEventExecutor.INSTANCE);
freeChannels = null;
}
handshakeTimeout = config.getHandshakeTimeout();
ThreadFactory threadFactory = config.getThreadFactory() != null ? config.getThreadFactory() : new DefaultThreadFactory(config.getThreadPoolName());
allowReleaseEventLoopGroup = config.getEventLoopGroup() == null;
ChannelFactory<? extends Channel> channelFactory;
if (allowReleaseEventLoopGroup) {
//这个只能在linux下使用
if (config.isUseNativeTransport()) {
eventLoopGroup = newEpollEventLoopGroup(config.getIoThreadsCount(), threadFactory);
channelFactory = getEpollSocketChannelFactory();
} else {
//通常默认走这个！
eventLoopGroup = new NioEventLoopGroup(config.getIoThreadsCount(), //瞧！IO线程数就是时间轮询的线程数
threadFactory);
channelFactory = NioSocketChannelFactory.INSTANCE;
}
} else {
//...
}
//用于http请求的bootstrap
httpBootstrap = newBootstrap(channelFactory, eventLoopGroup, config);
//用于WebSocket请求的bootstrap
wsBootstrap = newBootstrap(channelFactory, eventLoopGroup, config);
httpBootstrap.option(ChannelOption.AUTO_READ, false);
}
}
实例化完ChannelManager后，就轮到请求发送器，这里先看看所需要的参数，具体执行的方法在后面说。
public final class NettyRequestSender {
private final AsyncHttpClientConfig config;
private final ChannelManager channelManager;
private final Timer nettyTimer;
private final AsyncHttpClientState clientState;
private final NettyRequestFactory requestFactory;
public NettyRequestSender(AsyncHttpClientConfig config,//
ChannelManager channelManager,//
Timer nettyTimer,//
AsyncHttpClientState clientState) {
this.config = config;
this.channelManager = channelManager;
this.nettyTimer = nettyTimer;
this.clientState = clientState;
requestFactory = new NettyRequestFactory(config);
}
}
再回来看看ChannelManager构造方法中使用的工厂方法newBootstrap(channelFactory, eventLoopGroup, config)，这是支持整个ahc运作的代码，对于用Netty实现网络客户端来说，这个配置很有参考价值，所以也贴上来一起观赏！
private Bootstrap newBootstrap(ChannelFactory<? extends Channel> channelFactory, EventLoopGroup eventLoopGroup, AsyncHttpClientConfig config) {
@SuppressWarnings("deprecation")
Bootstrap bootstrap = new Bootstrap().channelFactory(channelFactory)
//客户端只有worker线程池，ServerBootstrap则需要boss和worker
.group(eventLoopGroup)
//设置内存分配器，我的理解是关于堆内存模型的，可用于对Netty的优化
.option(ChannelOption.ALLOCATOR, config.getAllocator() != null ? config.getAllocator() : ByteBufAllocator.DEFAULT)
//是否使用tcp的Nagle算法，文件传输可以选择使用
.option(ChannelOption.TCP_NODELAY, config.isTcpNoDelay())
//重复使用本地地址端口
.option(ChannelOption.SO_REUSEADDR, config.isSoReuseAddress())//
.option(ChannelOption.AUTO_CLOSE, false);
if (config.getConnectTimeout() > 0) {
bootstrap.option(ChannelOption.CONNECT_TIMEOUT_MILLIS, config.getConnectTimeout());
}
if (config.getSoLinger() >= 0) {
//就是一个设置延迟关闭时间的参数，用于保证数据发送完成
bootstrap.option(ChannelOption.SO_LINGER, config.getSoLinger());
}
if (config.getSoSndBuf() >= 0) {
bootstrap.option(ChannelOption.SO_SNDBUF, config.getSoSndBuf());
}
if (config.getSoRcvBuf() >= 0) {
bootstrap.option(ChannelOption.SO_RCVBUF, config.getSoRcvBuf());
}
//自定义配置
for (Entry<ChannelOption<Object>, Object> entry : config.getChannelOptions().entrySet()) {
bootstrap.option(entry.getKey(), entry.getValue());
}
return bootstrap;
}
//下面则是管道的配置
public void configureBootstraps(NettyRequestSender requestSender) {
//ahc自定义的ChannelInboundHandler，异步方式获取服务端返回的数据
//我们自己获取数据后的核心业务逻辑，也在这里开始
final AsyncHttpClientHandler httpHandler = new HttpHandler(config, this, requestSender);
wsHandler = new WebSocketHandler(config, this, requestSender);
final NoopHandler pinnedEntry = new NoopHandler();
httpBootstrap.handler(new ChannelInitializer<Channel>() {
@Override
protected void initChannel(Channel ch) throws Exception {
ChannelPipeline pipeline = ch.pipeline()//
.addLast(PINNED_ENTRY, pinnedEntry)//
.addLast(HTTP_CLIENT_CODEC, newHttpClientCodec())//
.addLast(INFLATER_HANDLER, newHttpContentDecompressor())//
.addLast(CHUNKED_WRITER_HANDLER, new ChunkedWriteHandler())//
.addLast(AHC_HTTP_HANDLER, httpHandler);
if (config.getHttpAdditionalChannelInitializer() != null)
config.getHttpAdditionalChannelInitializer().initChannel(ch);
}
});
wsBootstrap.handler(new ChannelInitializer<Channel>() {
@Override
protected void initChannel(Channel ch) throws Exception {
ChannelPipeline pipeline = ch.pipeline()//
.addLast(PINNED_ENTRY, pinnedEntry)//
.addLast(HTTP_CLIENT_CODEC, newHttpClientCodec())//
.addLast(AHC_WS_HANDLER, wsHandler);
if (config.getWsAdditionalChannelInitializer() != null)
config.getWsAdditionalChannelInitializer().initChannel(ch);
}
});
}一切工作准备就绪，现在可以请求了！怎么构建请求就不打算讲了，可以自行阅读RequestBuilderBase类。执行execute()方法，正式开始请求，往下看DefaultAsyncHttpClient.executeRequest()怎么创建连接的。
public class BoundRequestBuilder extends RequestBuilderBase<BoundRequestBuilder> {
private final AsyncHttpClient client;
public ListenableFuture<Response> execute() {
return client.executeRequest(build(), new AsyncCompletionHandlerBase());
}
}
public class DefaultAsyncHttpClient implements AsyncHttpClient {
@Override
public <T> ListenableFuture<T> executeRequest(Request request, AsyncHandler<T> handler) {
if (config.getRequestFilters().isEmpty()) {
return execute(request, handler);
} else {
//不考虑设置请求过滤器的情况
}
}
private <T> ListenableFuture<T> execute(Request request, final AsyncHandler<T> asyncHandler) {
try {
//把请求参数，和读取数据后的回调一同塞给请求发送器
return requestSender.sendRequest(request, asyncHandler, null, false);
} catch (Exception e) {
asyncHandler.onThrowable(e);
return new ListenableFuture.CompletedFailure<>(e);
}
}
}OK～～上面列出NettyRequestSender需要什么参数，现在再来看看怎么做的？ 下面的方法中，重点关注sendRequestWithNewChannel，它包括了如何新建channel、连接，抢占信号量
public <T> ListenableFuture<T> sendRequest(final Request request,//
final AsyncHandler<T> asyncHandler,//
NettyResponseFuture<T> future,//
boolean performingNextRequest) {
//...
ProxyServer proxyServer = getProxyServer(config, request);
//使用SSL代理或者ws
if (proxyServer != null && (request.getUri().isSecured() || request.getUri().isWebSocket()) && !isConnectDone(request, future))
//暂时忽略另外两个创建连接的方式
else
//我们的例子用的是GET，所以执行该方法
return sendRequestWithCertainForceConnect(request, asyncHandler, future, performingNextRequest, proxyServer, false);
}
private <T> ListenableFuture<T> sendRequestWithCertainForceConnect(//
Request request,//
AsyncHandler<T> asyncHandler,//
NettyResponseFuture<T> future,//注意，这时候传进来是null
boolean performingNextRequest,//
ProxyServer proxyServer,//
boolean forceConnect) {
//把所有请求信息保证在一个响应回调对象里
NettyResponseFuture<T> newFuture = newNettyRequestAndResponseFuture(request, asyncHandler, future, proxyServer, forceConnect);
//这里视图根据这个请求去拿去channel，过程有点漫长，回头再来解释
Channel channel = getOpenChannel(future, request, proxyServer, asyncHandler);
if (Channels.isChannelValid(channel))
return sendRequestWithOpenChannel(request, proxyServer, newFuture, asyncHandler, channel);
else
return sendRequestWithNewChannel(request, proxyServer, newFuture, asyncHandler, performingNextRequest);
}
private Channel getOpenChannel(NettyResponseFuture<?> future, Request request, ProxyServer proxyServer, AsyncHandler<?> asyncHandler) {
//future并没有channel，对于什么时候channel是可复用的，一直没搞明白，所以我基本默认每次都要新建一个channel
if (future != null && future.isReuseChannel() && Channels.isChannelValid(future.channel()))
return future.channel();
//视图在channelManager中找到可用对象
else
return pollPooledChannel(request, proxyServer, asyncHandler);
}
private <T> ListenableFuture<T> sendRequestWithOpenChannel(Request request, ProxyServer proxy, NettyResponseFuture<T> future, AsyncHandler<T> asyncHandler, Channel channel) {
if (asyncHandler instanceof AsyncHandlerExtensions)
AsyncHandlerExtensions.class.cast(asyncHandler).onConnectionPooled(channel);
//启动请求超时，在writeRequest中，会启动读取超时
TimeoutsHolder timeoutsHolder = scheduleRequestTimeout(future);
timeoutsHolder.initRemoteAddress((InetSocketAddress) channel.remoteAddress());
future.setChannelState(ChannelState.POOLED);
future.attachChannel(channel, false);
Channels.setAttribute(channel, future);
if (Channels.isChannelValid(channel)) {
writeRequest(future, channel);
} else {
handleUnexpectedClosedChannel(channel, future);
}
return future;
}
//把这里当作一个请求连接的开始
private <T> ListenableFuture<T> sendRequestWithNewChannel(//
Request request,//
ProxyServer proxy,//
NettyResponseFuture<T> future,//
AsyncHandler<T> asyncHandler,//
boolean performingNextRequest) {
Realm realm = future.getRealm();
Realm proxyRealm = future.getProxyRealm();
//...
//为做连接做准备
Bootstrap bootstrap = channelManager.getBootstrap(request.getUri(), proxy);
//用于索取channel
Object partitionKey = future.getPartitionKey();
final boolean acquireChannelLock = !performingNextRequest;
try {
//抢占信号量
if (acquireChannelLock) {
channelManager.acquireChannelLock(partitionKey);
}
} catch (Throwable t) {
abort(null, future, getCause(t));
return future;
}
//开启请求超时定时器
scheduleRequestTimeout(future);
//域名解析
RequestHostnameResolver.INSTANCE.resolve(request, proxy, asyncHandler)//
.addListener(new SimpleFutureListener<List<InetSocketAddress>>() {
@Override
//域名解析后得到的IP地址列表
protected void onSuccess(List<InetSocketAddress> addresses) {
NettyConnectListener<T> connectListener = new NettyConnectListener<>(future, NettyRequestSender.this, channelManager, acquireChannelLock, partitionKey);
//不要怀疑！这里开始连接了！！！
NettyChannelConnector connector = new NettyChannelConnector(request.getLocalAddress(), addresses, asyncHandler, clientState, config);
if (!future.isDone()) {
connector.connect(bootstrap, connectListener);
} else if (acquireChannelLock) {
//如果future已经完成，则释放信号量
channelManager.releaseChannelLock(partitionKey);
}
}
@Override
protected void onFailure(Throwable cause) {
//失败，释放信号
if (acquireChannelLock) {
channelManager.releaseChannelLock(partitionKey);
}
abort(null, future, getCause(cause));
}
});
return future;
}NettyChannelConnector负责对远程IP创建连接，一旦连接成功，NettyConnectListener就会调用requestSender向服务端发送数据。
public class NettyChannelConnector {
public void connect(final Bootstrap bootstrap, final NettyConnectListener<?> connectListener) {
//获取DNS后的IP地址
final InetSocketAddress remoteAddress = remoteAddresses.get(i);
if (asyncHandlerExtensions != null)
asyncHandlerExtensions.onTcpConnectAttempt(remoteAddress);
try {
connect0(bootstrap, connectListener, remoteAddress);
} catch (RejectedExecutionException e) {
if (clientState.isClosed()) {
connectListener.onFailure(null, e);
}
}
}
private void connect0(Bootstrap bootstrap, final NettyConnectListener<?> connectListener, InetSocketAddress remoteAddress) {
bootstrap.connect(remoteAddress, localAddress)//
.addListener(new SimpleChannelFutureListener() {
@Override
public void onSuccess(Channel channel) {
if (asyncHandlerExtensions != null) {
asyncHandlerExtensions.onTcpConnectSuccess(remoteAddress, channel);
}
//如果有设置连接的存活时间，则初始化channelId，在ChannelPool中自检有用到
if (connectionTtlEnabled) {
Channels.initChannelId(channel);
}
connectListener.onSuccess(channel, remoteAddress);
}
@Override
public void onFailure(Channel channel, Throwable t) {
if (asyncHandlerExtensions != null)
asyncHandlerExtensions.onTcpConnectFailure(remoteAddress, t);
//如果连接失败，则尝试连接下一个IP
boolean retry = pickNextRemoteAddress();
if (retry)
NettyChannelConnector.this.connect(bootstrap, connectListener);
else
connectListener.onFailure(channel, t);
}
});
}
}连接成功，就来到这里，拿到channel，准备向服务器发送数据！
public final class NettyConnectListener<T> {
public void onSuccess(Channel channel, InetSocketAddress remoteAddress) {
Channels.setInactiveToken(channel);
TimeoutsHolder timeoutsHolder = future.getTimeoutsHolder();
if (futureIsAlreadyCancelled(channel)) {
return;
}
Request request = future.getTargetRequest();
Uri uri = request.getUri();
timeoutsHolder.initRemoteAddress(remoteAddress);
if (future.getProxyServer() == null && uri.isSecured()) {
//直接无视
} else {
writeRequest(channel);
}
}
private void writeRequest(Channel channel) {
if (futureIsAlreadyCancelled(channel)) {
return;
}
//在这设置属性，在读取服务器数据的httphandler里面有用到
Channels.setAttribute(channel, future);
//注册到ChannelGroup中
channelManager.registerOpenChannel(channel, partitionKey);
//设置为不复用channel
future.attachChannel(channel, false);
//发送请求数据
//这个方法就不贴上来了，没什么意思
//方法里最后将启动读取超时scheduleReadTimeout(future);意味将进入HttpHandler读取服务端数据
requestSender.writeRequest(future, channel);
}
}
读取数据一切顺利后，就会走下面这个私有方法，将channel送入channelpool里，等待生命的结束！
public final class HttpHandler extends AsyncHttpClientHandler {
private void finishUpdate(final NettyResponseFuture<?> future, Channel channel, boolean expectOtherChunks) throws IOException {
future.cancelTimeouts();
boolean keepAlive = future.isKeepAlive();
//这里继续读取后面的数据块，最后channel被设置了回调，依然调用下面的tryToOfferChannelToPool方法
if (expectOtherChunks && keepAlive)
channelManager.drainChannelAndOffer(channel, future);
else
channelManager.tryToOfferChannelToPool(channel, future.getAsyncHandler(), keepAlive, future.getPartitionKey());
try {
future.done();
} catch (Exception t) {}
}
}
tryToOfferChannelToPool 是ChannelManager的方法，主要将依然活跃的channel送入生命倒数器中，还记得connectionTtl么，这个参数在这就起作用了！
public final void tryToOfferChannelToPool(Channel channel, AsyncHandler<?> asyncHandler, boolean keepAlive, Object partitionKey) {
//长连接，或者依然活跃的
if (channel.isActive() && keepAlive) {
//丢弃被设置的属性
Channels.setDiscard(channel);
if (asyncHandler instanceof AsyncHandlerExtensions)
AsyncHandlerExtensions.class.cast(asyncHandler).onConnectionOffer(channel);
//尝试塞进pool里
if (channelPool.offer(channel, partitionKey)) {
if (maxConnectionsPerHostEnabled)
//我没明白这个映射到底是干嘛用的
channelId2PartitionKey.putIfAbsent(channel, partitionKey);
} else {
//被pool驳回，就直接关闭掉！！
closeChannel(channel);
}
} else {
//已经死亡或者不是长连接，直接关闭！！
closeChannel(channel);
}
}到这里，关于channel已经接近尾声了，细心的童鞋可能发现，信号量呢？！不用释放么？！其实在关闭channel的时候，已经释放了，这是因为 ChannelGroup 的作用，在将channel注册(add方法)到group的时候，已经在其上面加了关闭的监听器，一旦close就执行remove，实例化 ChannelGroup 时已经将remove(channel)重写，可以倒回去看是不是已经释放了信号量，也可以看看 ChannelGroup 源码是不是在add时候添加了监听器。
不过，这里只是接近尾声，没意味就结束了，还有存活的channel被塞到 ChannelPool 进行生命的倒计时
public final class DefaultChannelPool implements ChannelPool {
private final ConcurrentHashMap<Object, ConcurrentLinkedDeque<IdleChannel>> partitions = new ConcurrentHashMap<>();
private final ConcurrentHashMap<ChannelId, ChannelCreation> channelId2Creation;
private final AtomicBoolean isClosed = new AtomicBoolean(false);
private final Timer nettyTimer;
private final int connectionTtl;
private final boolean connectionTtlEnabled;
private final int maxIdleTime;
private final boolean maxIdleTimeEnabled;
private final long cleanerPeriod;
private final PoolLeaseStrategy poolLeaseStrategy;
public DefaultChannelPool(int maxIdleTime,//
int connectionTtl,//
PoolLeaseStrategy poolLeaseStrategy,//
Timer nettyTimer,//
int cleanerPeriod) {
this.maxIdleTime = maxIdleTime;
this.connectionTtl = connectionTtl;
connectionTtlEnabled = connectionTtl > 0;
channelId2Creation = connectionTtlEnabled ? new ConcurrentHashMap<>() : null;
this.nettyTimer = nettyTimer;
maxIdleTimeEnabled = maxIdleTime > 0;
this.poolLeaseStrategy = poolLeaseStrategy;
//在cleanerPeriod清理周期时间、connectionTtl连接存活时间、maxIdleTime最大空闲时间中选择最小的
this.cleanerPeriod = Math.min(cleanerPeriod定时清理周期, Math.min(connectionTtlEnabled ? connectionTtl : Integer.MAX_VALUE, maxIdleTimeEnabled ? maxIdleTime : Integer.MAX_VALUE));
//如果开启了连接存活时间，或者最大空闲时间，则实例化空闲channel检测
if (connectionTtlEnabled || maxIdleTimeEnabled)
scheduleNewIdleChannelDetector(new IdleChannelDetector());
}
private void scheduleNewIdleChannelDetector(TimerTask task) {
nettyTimer.newTimeout(task, cleanerPeriod, TimeUnit.MILLISECONDS);
}
private final class IdleChannelDetector implements TimerTask {
//挖出已经不满足条件的channel
private List<IdleChannel> expiredChannels(ConcurrentLinkedDeque<IdleChannel> partition, long now) {
List<IdleChannel> idleTimeoutChannels = null;
for (IdleChannel idleChannel : partition) {
//空闲时间是否过期
boolean isIdleTimeoutExpired = isIdleTimeoutExpired(idleChannel, now);
//channel是否还活跃
boolean isRemotelyClosed = isRemotelyClosed(idleChannel.channel);
//存活时间是否过期
boolean isTtlExpired = isTtlExpired(idleChannel.channel, now);
//满足其中一个条件，加入即将被关闭的channel队列
if (isIdleTimeoutExpired || isRemotelyClosed || isTtlExpired) {
if (idleTimeoutChannels == null)
idleTimeoutChannels = new ArrayList<>(1);
idleTimeoutChannels.add(idleChannel);
}
}
return idleTimeoutChannels != null ? idleTimeoutChannels : Collections.<IdleChannel> emptyList();
}
//关闭expiredChannels筛选出来的队列，并返回一个已被close的channel队列
private List<IdleChannel> closeChannels(List<IdleChannel> candidates) {
List<IdleChannel> closedChannels = null;
for (int i = 0; i < candidates.size(); i++) {
IdleChannel idleChannel = candidates.get(i);
//如果未被占有，则直接close；如果中间出现有被占有的channel，实例化closedChannels，并将之前被close的channel塞进其中
if (idleChannel.takeOwnership()) {
close(idleChannel.channel);
if (closedChannels != null) {
closedChannels.add(idleChannel);
}
} //注意，这里只会被执行一次，closedChannels被实例化后不会再执行
else if (closedChannels == null) {
closedChannels = new ArrayList<>(candidates.size());
for (int j = 0; j < i; j++)
closedChannels.add(candidates.get(j));
}
}
//如果closedChannels为null，代表已经关闭candidates所有channel，原封不动返回
//如果closedChannels非null，代表被占用的channel没有close并继续存活在candidates，所以返回被close了的channel队列closedChannels
return closedChannels != null ? closedChannels : candidates;
}
public void run(Timeout timeout) throws Exception {
if (isClosed.get())
return;
//检测器的启动时间
long start = unpreciseMillisTime();
int closedCount = 0;
int totalCount = 0;
//遍历每个路由的被塞到ChannelPool的channel队列
for (ConcurrentLinkedDeque<IdleChannel> partition : partitions.values()) {
List<IdleChannel> closedChannels = closeChannels(expiredChannels(partition, start));
//非空且开启了连接存活时间的channel且被close的channel，全部从channelId2Creation和partition中去除
if (!closedChannels.isEmpty()) {
if (connectionTtlEnabled) {
for (IdleChannel closedChannel : closedChannels)
channelId2Creation.remove(channelId(closedChannel.channel));
}
partition.removeAll(closedChannels);
closedCount += closedChannels.size();
}
}
//退出并继续下一轮检测
scheduleNewIdleChannelDetector(timeout.task());
}
}
//存放空闲channel
private static final class IdleChannel {
final Channel channel;
final long start;
final AtomicBoolean owned = new AtomicBoolean(false);
IdleChannel(Channel channel, long start) {
this.channel = assertNotNull(channel, "channel");
this.start = start;
}
public boolean takeOwnership() {
return owned.compareAndSet(false, true);
}
@Override
public boolean equals(Object o) {...}
@Override
public int hashCode() {...}
}
//存放channel的创建时间
private static final class ChannelCreation {
final long creationTime;
final Object partitionKey;
ChannelCreation(long creationTime, Object partitionKey) {
this.creationTime = creationTime;
this.partitionKey = partitionKey;
}
}
}这里才是channel的终结！！！
channel被终结了，但有些还存活的channel还在请求的路上，还有很重要的两点没说到，就是 请求超时 和 读取超时。 每个NettyResponseFuture都持有一个TimeoutsHolder来计算 requestTimeout 和 readTimeout 是否过期。在ResponseFuture获取连接后，以及获取成功向服务器发送数据后，都会分别启动请求超时和读取超时两个定时器。通过阅读源码，可以发现 requestTimeout 其实是包括了 readTimeout，如果请求剩余时间小于读取超时时间时，startReadTimeout是不会启动readTimeout定时器的。下面只贴上TimeoutsHolder的部分源码，RequestTimeoutTimerTask和ReadTimeoutTimerTask可以自行阅读。
对于这两个参数，需要说明一点就是，一旦超时过期，channel和future都会被close掉，如果读超设置比请超长则是无意义的，只会以requestTimeout为准。
public class TimeoutsHolder {
private final AtomicBoolean cancelled = new AtomicBoolean();
private final Timer nettyTimer;
private final NettyRequestSender requestSender;
private final long requestTimeoutMillisTime;
private final int readTimeoutValue;
private volatile NettyResponseFuture<?> nettyResponseFuture;
public final Timeout requestTimeout;
public volatile Timeout readTimeout;
public TimeoutsHolder(Timer nettyTimer, NettyResponseFuture<?> nettyResponseFuture, NettyRequestSender requestSender, AsyncHttpClientConfig config) {
this.nettyTimer = nettyTimer;
this.nettyResponseFuture = nettyResponseFuture;
this.requestSender = requestSender;
this.readTimeoutValue = config.getReadTimeout();
int requestTimeoutInMs = nettyResponseFuture.getTargetRequest().getRequestTimeout();
//每个请求都可以独立设置请求超时时间
if (requestTimeoutInMs == 0) {
requestTimeoutInMs = config.getRequestTimeout();
}
if (requestTimeoutInMs != -1) {
//请求的到期时间，启动请求超时定时器
requestTimeoutMillisTime = unpreciseMillisTime() + requestTimeoutInMs;
requestTimeout = newTimeout(new RequestTimeoutTimerTask(nettyResponseFuture, requestSender, this, requestTimeoutInMs), requestTimeoutInMs);
} else {
requestTimeoutMillisTime = -1L;
requestTimeout = null;
}
}
public void startReadTimeout() {
if (readTimeoutValue != -1) {
startReadTimeout(null);
}
}
void startReadTimeout(ReadTimeoutTimerTask task) {
//如果requestTimeout不为null，或者requestTimeout还没有过期并且读取超时时间<请求剩余时间
if (requestTimeout == null || (!requestTimeout.isExpired() && readTimeoutValue < (requestTimeoutMillisTime - unpreciseMillisTime()))) {
if (task == null) {
task = new ReadTimeoutTimerTask(nettyResponseFuture, requestSender, this, readTimeoutValue);
}
Timeout readTimeout = newTimeout(task, readTimeoutValue);
this.readTimeout = readTimeout;
} else if (task != null) {
task.clean();
}
}
}最后最后最后。。。
来总结一下ahc的连接池实现，很明显的一点整个过程都是对Channel的管理，而且对于连接的抢占则使用了Semaphore，这再方便不过了！！！对于信号量的释放，Netty的ChannelGroup有很大的功劳，它提供了最优雅的方式关闭channel并且释放信号量。除此之外，一堆的超时限制任务需要一个定时任务容器执行，Netty又提供了一个在面对大量任务依然稳坐泰山的HashedWheelTimer，有机会专门来说说这一个。还有就是DefaultChannelPool对存活时间的检测，实在是通俗易懂，而且基于前面说的几点，实现起来也相当方便。
如果遇到基于netty的网络编程开发，对于连接资源的管理ahc确实提供了一套不错的思路，不仅对客户端，服务端也是可以试一试的！
190
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
