Dubbo性能调优参数及原理

Posted 2021-04-05 JAVA烂猪皮

Dubbo调用模型

常用性能调优参数

---

源码及原理分析

>>  threads

FixedThreadPool.java

public Executor getExecutor(URL url) {
    String name = url.getParameter(Constants.THREAD_NAME_KEY, Constants.DEFAULT_THREAD_NAME);    int threads = url.getParameter(Constants.THREADS_KEY, Constants.DEFAULT_THREADS);    int queues = url.getParameter(Constants.QUEUES_KEY, Constants.DEFAULT_QUEUES);    return new ThreadPoolExecutor(threads, threads, 0, TimeUnit.MILLISECONDS, 
            queues == 0 ? new SynchronousQueue<Runnable>() : 
                    (queues < 0 ? new LinkedBlockingQueue<Runnable>() : 
                            new LinkedBlockingQueue<Runnable>(queues)),            new NamedThreadFactory(name, true), new AbortPolicyWithReport(name, url));
}

---

LimitedThreadPool.java

public Executor getExecutor(URL url) {
    String name = url.getParameter(Constants.THREAD_NAME_KEY, Constants.DEFAULT_THREAD_NAME);    int cores = url.getParameter(Constants.CORE_THREADS_KEY, Constants.DEFAULT_CORE_THREADS);    int threads = url.getParameter(Constants.THREADS_KEY, Constants.DEFAULT_THREADS);    int queues = url.getParameter(Constants.QUEUES_KEY, Constants.DEFAULT_QUEUES);    return new ThreadPoolExecutor(cores, threads, Long.MAX_VALUE, TimeUnit.MILLISECONDS, 
            queues == 0 ? new SynchronousQueue<Runnable>() : 
                    (queues < 0 ? new LinkedBlockingQueue<Runnable>() : 
                            new LinkedBlockingQueue<Runnable>(queues)),            new NamedThreadFactory(name, true), new AbortPolicyWithReport(name, url));
}

其中，Constants.DEFAULT_QUEUES = 200。threads 参数配置的是业务处理线程池的最大（或核心）线程数。

 

---

>>  iothreads

NettyServer.java

@Overrideprotected void doOpen() throws Throwable {
    NettyHelper.setNettyLoggerFactory();
    ExecutorService boss = Executors.newCachedThreadPool(new NamedThreadFactory("NettyServerBoss", true));
    ExecutorService worker = Executors.newCachedThreadPool(new NamedThreadFactory("NettyServerWorker", true));
    ChannelFactory channelFactory = new NioserverSocketChannelFactory(boss, worker, getUrl().getPositiveParameter(Constants.IO_THREADS_KEY, Constants.DEFAULT_IO_THREADS));
    bootstrap = new ServerBootstrap(channelFactory);        
    final NettyHandler nettyHandler = new NettyHandler(getUrl(), this);
    channels = nettyHandler.getChannels();
    bootstrap.setPipelineFactory(new ChannelPipelineFactory() {        public ChannelPipeline getPipeline() {
            NettyCodecAdapter adapter = new NettyCodecAdapter(getCodec() ,getUrl(), NettyServer.this);
            ChannelPipeline pipeline = Channels.pipeline();
            pipeline.addLast("decoder", adapter.getDecoder());
            pipeline.addLast("encoder", adapter.getEncoder());
            pipeline.addLast("handler", nettyHandler);            return pipeline;
        }
    });    // bind
    channel = bootstrap.bind(getBindAddress());
}

>>  queues

分别在FixedThreadPool.java、LimitedThreadPool.java 和 CachedThreadPool.java 中使用，代码详情见 3.2章节。由代码可见，默认值为 0，表示使用同步阻塞队列；如果 queues 设置为小于 0 的值，则使用容量为 Integer.MAX_VALUE 的阻塞链表队列；如果为其他值，则使用指定大小的阻塞链表队列。

 

---

>>  connections

DubboProtocol.java

private ExchangeClient[] getClients(URL url){    //是否共享连接
    boolean service_share_connect = false;    int connections = url.getParameter(Constants.CONNECTIONS_KEY, 0);    //如果connections不配置，则共享连接，否则每服务每连接
    if (connections == 0){
        service_share_connect = true;
        connections = 1;
    }

    ExchangeClient[] clients = new ExchangeClient[connections];    for (int i = 0; i < clients.length; i++) {        if (service_share_connect){
            clients[i] = getSharedClient(url);
        } else {
            clients[i] = initClient(url);
        }
    }    return clients;
}

---

DubboInvoker.java

@Overrideprotected Result doInvoke(final Invocation invocation) throws Throwable {
    RpcInvocation inv = (RpcInvocation) invocation;    final String methodName = RpcUtils.getMethodName(invocation);
    inv.setAttachment(Constants.PATH_KEY, getUrl().getPath());
    inv.setAttachment(Constants.VERSION_KEY, version);
    
    ExchangeClient currentClient;    if (clients.length == 1) {
        currentClient = clients[0];
    } else {
        currentClient = clients[index.getAndIncrement() % clients.length];
    }    try {        boolean isAsync = RpcUtils.isAsync(getUrl(), invocation);        boolean isOneway = RpcUtils.isOneway(getUrl(), invocation);        int timeout = getUrl().getMethodParameter(methodName, Constants.TIMEOUT_KEY,Constants.DEFAULT_TIMEOUT);        if (isOneway) {            boolean isSent = getUrl().getMethodParameter(methodName, Constants.SENT_KEY, false);
            currentClient.send(inv, isSent);
            RpcContext.getContext().setFuture(null);            return new RpcResult();
        } else if (isAsync) {
            ResponseFuture future = currentClient.request(inv, timeout) ;
            RpcContext.getContext().setFuture(new FutureAdapter<Object>(future));            return new RpcResult();
        } else {
            RpcContext.getContext().setFuture(null);            return (Result) currentClient.request(inv, timeout).get();
        }
    } catch (TimeoutException e) {        throw new RpcException(RpcException.TIMEOUT_EXCEPTION, "Invoke remote method timeout. method: " + invocation.getMethodName() + ", provider: " + getUrl() + ", cause: " + e.getMessage(), e);
    } catch (RemotingException e) {        throw new RpcException(RpcException.NETWORK_EXCEPTION, "Failed to invoke remote method: " + invocation.getMethodName() + ", provider: " + getUrl() + ", cause: " + e.getMessage(), e);
    }
}

以上可见，默认值为0，表示针对每个 Provider，所有客户端共享一个长连接；否则，建立指定数量的长连接。在调用时，如果有多个长连接，则使用轮询方式获得一个长连接。

---

>>  actives

ActiveLimitFilter.java

public Result invoke(Invoker<?> invoker, Invocation invocation) throws RpcException {
    URL url = invoker.getUrl();
    String methodName = invocation.getMethodName();    int max = invoker.getUrl().getMethodParameter(methodName, Constants.ACTIVES_KEY, 0);
    RpcStatus count = RpcStatus.getStatus(invoker.getUrl(), invocation.getMethodName());    if (max > 0) {        long timeout = invoker.getUrl().getMethodParameter(invocation.getMethodName(), Constants.TIMEOUT_KEY, 0);        long start = System.currentTimeMillis();        long remain = timeout;        int active = count.getActive();        if (active >= max) {            synchronized (count) {                while ((active = count.getActive()) >= max) {                    try {
                        count.wait(remain);
                    } catch (InterruptedException e) {
                    }                    long elapsed = System.currentTimeMillis() - start;
                    remain = timeout - elapsed;                    if (remain <= 0) {                        throw new RpcException("Waiting concurrent invoke timeout in client-side for service:  "
                                               + invoker.getInterface().getName() + ", method: "
                                               + invocation.getMethodName() + ", elapsed: " + elapsed                                               + ", timeout: " + timeout + ". concurrent invokes: " + active                                               + ". max concurrent invoke limit: " + max);
                    }
                }
            }
        }
    }    try {        long begin = System.currentTimeMillis();
        RpcStatus.beginCount(url, methodName);        try {
            Result result = invoker.invoke(invocation);
            RpcStatus.endCount(url, methodName, System.currentTimeMillis() - begin, true);            return result;
        } catch (RuntimeException t) {
            RpcStatus.endCount(url, methodName, System.currentTimeMillis() - begin, false);            throw t;
        }
    } finally {        if(max>0){            synchronized (count) {
                count.notify();
            } 
        }
    }
}

Consumer 调用时，统计服务和方法维度的调用情况，如果并发数超过设置的最大值，则阻塞当前线程，直到前面有请求处理完成。

---

 >>  accepts

AbstractServer.java

@Overridepublic void connected(Channel ch) throws RemotingException {
    Collection<Channel> channels = getChannels();    if (accepts > 0 && channels.size() > accepts) {
        logger.error("Close channel " + ch + ", cause: The server " + ch.getLocalAddress() + " connections greater than max config " + accepts);
        ch.close();        return;
    }    super.connected(ch);
}

当连接数大于最大值时，关闭当前连接。

---

>>  executes

ExecuteLimitFilter.jvava

public Result invokeOrg(Invoker<?> invoker, Invocation invocation) throws RpcException {
    URL url = invoker.getUrl();
    String methodName = invocation.getMethodName();    int max = url.getMethodParameter(methodName, Constants.EXECUTES_KEY, 0);    if (max > 0) {
        RpcStatus count = RpcStatus.getStatus(url, invocation.getMethodName());        if (count.getActive() >= max) {            throw new RpcException("Failed to invoke method " + invocation.getMethodName() + " in provider " + url + ", cause: The service using threads greater than <dubbo:service executes=\"" + max + "\" /> limited.");
        }
    }    long begin = System.currentTimeMillis();    boolean isException = false;
    RpcStatus.beginCount(url, methodName);    try {
        Result result = invoker.invoke(invocation);        return result;
    } catch (Throwable t) {
        isException = true;        if(t instanceof RuntimeException) {            throw (RuntimeException) t;
        }        else {            throw new RpcException("unexpected exception when ExecuteLimitFilter", t);
        }
    }    finally {
        RpcStatus.endCount(url, methodName, System.currentTimeMillis() - begin, isException);
    }
}

Provider处理请求时，统计方法维度的调用情况，如果并发数超过设置的最大值，则阻直接抛出异常。

---

---