记低版本okhttp超时会导致死锁

一、问题起源

在处理一次生产环境cpu拉满问题时，把日志拉下来看发现很多http请求调用出错，项目使用的是okhttp 3.8.1版本。

二、问题描述

问题出在okhttp3.Dispatcher.finished(Dispatcher.java:201)代码如下：

void finished(AsyncCall call) {finished(runningAsyncCalls, call, true);
}
void finished(RealCall call) {finished(runningSyncCalls, call, false);
}
private <T> void finished(Deque<T> calls, T call, boolean promoteCalls) {int runningCallsCount;Runnable idleCallback;synchronized (this) { //201行if (!calls.remove(call)) throw new AssertionError("Call wasn't in-flight!");if (promoteCalls) promoteCalls();runningCallsCount = runningCallsCount();idleCallback = this.idleCallback;}if (runningCallsCount == 0 && idleCallback != null) {idleCallback.run();}
}private void promoteCalls() {if (runningAsyncCalls.size() >= maxRequests) return; // Already running max capacity.if (readyAsyncCalls.isEmpty()) return; // No ready calls to promote.for (Iterator<AsyncCall> i = readyAsyncCalls.iterator(); i.hasNext(); ) {AsyncCall call = i.next();if (runningCallsForHost(call) < maxRequestsPerHost) {i.remove();runningAsyncCalls.add(call);executorService().execute(call);}if (runningAsyncCalls.size() >= maxRequests) return; // Reached max capacity.}
}

三、分析代码

在OkHttpClient中final Dispatcher dispatcher; 作为成员对象，而我们代码中OkHttpClient作为连接池是单例的，这里是对dispatcher做synchronized。

追踪代码发现，在finished的调用方法中，我们方法中使用的是异步AsyncCall，而这里synchronized方法中的promoteCalls被置为true。所以会调用promoteCalls()方法， 而promoteCalls()方法中会继续调用executorService().execute(call);，就是这里，问题大了，synchronized中执行http请求，那上面代码中的超时不就长时间占用锁了？怪不得进程blocked了。

关于线程的BLOCKED，需要知道：

• java.lang.Thread.State: BLOCKED：等待监视器锁而被阻塞的线程的线程状态，当进入 synchronized 块/方法或者在调用 wait()被唤醒/超时之后重新进入 synchronized 块/方法， 但是锁被其它线程占有，这个时候被操作系统挂起，状态为阻塞状态。若是有线程长时间处于 BLOCKED 状态，要考虑是否发生了死锁（deadlock）的情况。
• blocked的线程不会消耗cpu，但频繁的频繁切换线程上下文会导致cpu过高。线程被频繁唤醒,而又由于抢占锁失败频繁地被挂起. 因此也会带来大量的上下文切换, 消耗系统的cpu资源。

四、解决方案

okttp关于这个问题已经有过解答：

Dispatcher no longer has quadratic behaviour by iamdanfox · Pull Request #4581 · square/okhttp · GitHub

[improvement] okhttp 3.12.0 -> 3.13.1, to pick up perf improvements to okhttp3.Dispatcher by iamdanfox · Pull Request #940 · palantir/conjure-java-runtime · GitHub

解决方案就简单多了：升级okhttp到3.14.9，虽然目前最新稳定版本为4.9.3，但是OkHttp 4发布,从Java切换到Kotlin。谨慎一点，还是小版本升级吧。

在3.14.9中，这部分代码被优化为：

private boolean promoteAndExecute() {assert (!Thread.holdsLock(this));List<AsyncCall> executableCalls = new ArrayList<>();boolean isRunning;synchronized (this) {for (Iterator<AsyncCall> i = readyAsyncCalls.iterator(); i.hasNext(); ) {AsyncCall asyncCall = i.next();if (runningAsyncCalls.size() >= maxRequests) break; // Max capacity.if (asyncCall.callsPerHost().get() >= maxRequestsPerHost) continue; // Host max capacity.i.remove();asyncCall.callsPerHost().incrementAndGet();executableCalls.add(asyncCall);runningAsyncCalls.add(asyncCall);}isRunning = runningCallsCount() > 0;}for (int i = 0, size = executableCalls.size(); i < size; i++) {AsyncCall asyncCall = executableCalls.get(i);asyncCall.executeOn(executorService());}return isRunning;
}

执行HTTP请求被移出了synchronized方法了。