js异步之惑
本文地址:https://blog.whyun.com/posts/js
1.异步是啥
与异步对应的就是同步,对于同步我们很好理解,就是代码顺序执行。但是一到异步代码,很多人多少有些理不清。异步,从功能上讲,就是在背后偷偷的执行,不堵塞当前运行的代码;从实现上讲,能够这么做的,就只能靠在当前运行代码中另一起线程或者进程了。举一个使用线程来实现异步的例子:
public class MyAsync extends Thread {private volatile boolean done = false;public void run() {while (!done) {//子线程中的循环System.out.println("thread out x");try {Thread.sleep(500);} catch (InterruptedException e) {e.printStackTrace();// Thread.currentThread().interrupt();}}}public synchronized void setDone() {done = true;}public static void main(String argv[]) {MyAsync t = new MyAsync();t.start();//起子线程long last = System.currentTimeMillis();int count = 0;while(true) {//主线程中循环long now = System.currentTimeMillis();if (now - last > 1000 * 1) {last = now;count ++;System.out.println("the " + count + "th count.");}if (count > 2) {break;}}t.setDone();}}
代码1.1 线程示例
对于代码1.1的运行结果,有可能是这个样子的:
thread out x
thread out x
thread out x
the 1th count.
thread out x
thread out x
the 2th count.
thread out x
thread out x
the 3th count.
代码27-38行起是主线程循环,7-15行是子线程循环,你多运行几次,就可以看出两个循环的输出是很随机的,但是不管运行多少次两个循环的输出都是交叉在一起的。这样我们就可以推断出,运行主线程的代码的时候,子线程的代码也在背后偷偷的运行,说白了两者是并行运行的。
2.js中的异步
就是因为异步这个特性,js如今被大家推崇,下面用一个小例子来演示一下js中异步的使用:
function synchronizedCode() {var last = new Date().getTime();var count = 0;while (true) {var now = new Date().getTime();if (now - last > 1000 * 2) {last = now;count++;console.log('the %dth count.',count);}if (count > 9) {console.log('exist while.');break;}}}(function() {setTimeout(function() {console.log('setTimeout 0 occured first.');},0);setTimeout(function() {console.log('setTimeout 0 occured second.');},0);synchronizedCode();})();
代码2.1 setTimeout的例子
我们运行代码2.1,然后不管运行多少次,输出都是这个样子的:
the 1th count.
the 2th count.
the 3th count.
the 4th count.
the 5th count.
exist while.
setTimeout 0 occured first.
setTimeout 0 occured second.
输出2.1
跟java中的异步和同步代码会交叉输出相比,js中的异步其实是排好队输出的。由于js是单线程执行代码的,所以没有那种交叉输出的效果。那么还有一个问题,while循环明明运行了5秒钟,为何在这期间那两个setTimeout一直没有运行呢?这就和js代码的异步机制有关了。js代码中有帧的概念,对于同步代码是在当前帧运行的,异步代码是在下一帧运行的。对于代码2.1我们给代码运行画一幅图的话,应该是这样的:
[外链图片转存失败,源站可能有防盗链机制,建议将图片保存下来直接上传(img-xN4Fg3k9-1676965585638)(https://raw.githubusercontent.com/yunnysunny/yunnysunny.github.io/master/images/js_frame.png)]
图2.1 js帧结构
那么为什么是第一个setTimeout先触发,第二个后触发呢,难道仅仅由于先后顺序?我们把第一个setTimeout改为setTimeout(function() {console.log('setTimeout 0 occured first.');},100);
,那么输出的时候就会是先输出setTimeout 0 occured second.
,在输出setTimeout 0 occured first.
。也就是说在第二帧setTimeout的回调的执行顺序不仅与代码顺序有关还和延迟时间有关。
在node.js中还有一个特殊的API,就是process.nextTick
,虽然已经不推荐使用了,但是已经可以在很多代码中看到它的身影。例如如下代码:
(function() {setTimeout(function() {console.log('setTimeout 0 occured first.');},0);setTimeout(function() {console.log('setTimeout 0 occured second.');},0);process.nextTick(function() {console.log('nextTick occured.');});synchronizedCode();})();
代码2.2
运行后输出:
the 1th count.
the 2th count.
the 3th count.
the 4th count.
the 5th count.
exist while.
nextTick occured.
setTimeout 0 occured first.
setTimeout 0 occured second.
输出2.2
之所以nextTick排在所有异步的最前面,是由于nextTick是在第一帧运行的,而其他的都是在第二帧运行的。也就是说代码运行情况是这个样子的:
[外链图片转存失败,源站可能有防盗链机制,建议将图片保存下来直接上传(img-LlpydYcE-1676965585639)(https://raw.githubusercontent.com/yunnysunny/yunnysunny.github.io/master/images/js_frame2.png)]
图2.2 js帧结构
接下来再举几个异步API的例子,这次我们添加setImmediate
和mkdir
两个函数:
var synchronizedCode = require('./sync');(function() {setTimeout(function() {console.log('setTimeout 0 occured first.');},0);setTimeout(function() {console.log('setTimeout 0 occured second.');},0);process.nextTick(function() {console.log('nextTick occured.');});setImmediate(function() {console.log('setImmediate occured.')});var fs = require('fs');var crypto = require('crypto');var rand = crypto.pseudoRandomBytes(8).toString('hex');fs.mkdir('d:\\temp\\xx'+'\\'+rand,function(err) {if (err) {console.log(err,'错误',err.code);} else {console.log('create directory success.');} });synchronizedCode();})();
代码2.3
那么他的输出就应该是这样的:
the 1th count.
the 2th count.
the 3th count.
the 4th count.
the 5th count.
exist while.
nextTick occured.
setTimeout 0 occured first.
setTimeout 0 occured second.
setImmediate occured.
create directory success.
输出2.3
等等,问题来了,这里最后一句才打印create directory success
,那么是不是程序是在最后一步才创建的文件夹呢,如果真是这样,就有点低效了,起码这个创建文件的工作被那个while循环给延迟了得有5秒钟。不过幸好,这个想法是错误的!node.js中使用libuv来IO或者CPU计算量大的操作,而在libuv中处理这些耗时的操作都是用线程来解决,以避免堵塞住js线程(这一点和android的程序设计思路类似,android开发中使用子线程来处理耗时逻辑,避免对主线程造成卡顿)。这里我们来演示一个libuv的异步处理,在异步处理中模拟一个耗时操作:
#include <node.h>#include <string>#ifdef WINDOWS_SPECIFIC_DEFINE#include <windows.h>typedef DWORD ThreadId;#else#include <unistd.h>#include <pthread.h>typedef unsigned int ThreadId;#endifusing namespace v8;Handle<Value> async_hello(const Arguments& args);//不在js主线程,,在uv线程池内被调用void call_work(uv_work_t* req);//回调函数void call_work_after(uv_work_t* req);static ThreadId __getThreadId() {ThreadId nThreadID;#ifdef WINDOWS_SPECIFIC_DEFINEnThreadID = GetCurrentProcessId();nThreadID = (nThreadID << 16) + GetCurrentThreadId();#elsenThreadID = getpid();nThreadID = (nThreadID << 16) + pthread_self();#endifreturn nThreadID;}static void __tsleep(unsigned int millisecond) {#ifdef WINDOWS_SPECIFIC_DEFINE::Sleep(millisecond);#elseusleep(millisecond*1000);#endif}//定义一个结构体,存储异步请求信息struct Baton {//存储回调函数,使用Persistent来声明,让系统不会在函数结束后自动回收//当回调成功后,需要执行dispose释放空间Persistent<Function> callback;// 错误控制,保护错误信息和错误状态bool error;std::string error_message;//存储js传入的参数,字符串std::string input_string;//存放返回参数,字符串std::string result;};Handle<Value> async_hello(const Arguments& args) {printf("\n%s Thread id : gettid() == %d\n",__FUNCTION__,__getThreadId());HandleScope scope;if(args.Length() < 2) { ThrowException(Exception::TypeError(String::New("Wrong number of arguments"))); return scope.Close(Undefined()); } if (!args[0]->IsString() || !args[1]->IsFunction()) {return ThrowException(Exception::TypeError(String::New("Wrong number of arguments")));}// 强制转换成函数变量Local<Function> callback = Local<Function>::Cast(args[1]);Baton* baton = new Baton();baton->error = false;baton->callback = Persistent<Function>::New(callback);v8::String::Utf8Value param1(args[0]->ToString());baton->input_string = std::string(*param1); uv_work_t *req = new uv_work_t();req->data = baton;int status = uv_queue_work(uv_default_loop(), req, call_work,(uv_after_work_cb)call_work_after);assert(status == 0);return Undefined();}//在该函数内模拟处理过程 ,如i/o阻塞或者cpu高消耗情形的处理。// 注意不能使用v8 api,这个线程不是在js主线程内void call_work(uv_work_t* req) {printf("\n%s Thread id : gettid() == %d\n",__FUNCTION__,__getThreadId());Baton* baton = static_cast<Baton*>(req->data);for (int i=0;i<15;i++) {__tsleep(1000);printf("sleep 1 seconds in uv_work\n");}baton->result = baton->input_string+ "--->hello world from c++";}//执行完任务,进行回调的时候,返回到 js主线程void call_work_after(uv_work_t* req) {printf("\n%s Thread id : gettid() == %d\n",__FUNCTION__,__getThreadId()); HandleScope scope;Baton* baton = static_cast<Baton*>(req->data);if (baton->error) {Local<Value> err = Exception::Error(String::New(baton->error_message.c_str()));// 准备回调函数的参数const unsigned argc = 1;Local<Value> argv[argc] = { err };//捕捉回调中的错误,在Node环境中可利用process.on('uncaughtException')捕捉错误 TryCatch try_catch;baton->callback->Call(Context::GetCurrent()->Global(), argc, argv);if (try_catch.HasCaught()) {node::FatalException(try_catch);}} else {const unsigned argc = 2;Local<Value> argv[argc] = {Local<Value>::New(Null()),Local<Value>::New(String::New(baton->result.c_str()))};TryCatch try_catch;baton->callback->Call(Context::GetCurrent()->Global(), argc, argv);if (try_catch.HasCaught()) {node::FatalException(try_catch);}}//注意一定释放baton->callback.Dispose();// 处理完毕,清除对象和空间delete baton;delete req;}void RegisterModule(Handle<Object> target) {target->Set(String::NewSymbol("async_hello"),FunctionTemplate::New(async_hello)->GetFunction());}NODE_MODULE(binding, RegisterModule);
代码2.4
编写的测试代码,将其运行,就可以看出函数call_work
根本就不在js线程中执行,也就是说它是可以和js线程并行的;函数call_work_after
中能够触发js中的回调函数,将处理完的结果交给js。下面就编译上面代码(需要node-gyp支持,执行node-gpy rebuild
进行编译),来验证上述结论:
(function() {setTimeout(function() {synchronizedCode('timer1',3);},0);setTimeout(function() {console.log('setTimeout 0 occured second.');},0);process.nextTick(function() {console.log('nextTick occured.');});var addon = require('./addon/build/Release/binding');addon.async_hello("good",function(err, result) {console.log('node addon result',result);});setImmediate(function() {console.log('setImmediate occured.')});var fs = require('fs');var crypto = require('crypto');var rand = crypto.pseudoRandomBytes(8).toString('hex');fs.mkdir('d:\\temp\\xx'+'\\'+rand,function(err) {if (err) {console.log(err,'错误',err.code);} else {console.log('create directory success.');} });synchronizedCode();})();
代码2.5
我们在代码2.5中引入了代码2.4中的c++扩展,其输出内容如下
async_hello Thread id : gettid() == 284953360call_work Thread id : gettid() == 284958096
sleep 1 seconds in uv_work
sleep 1 seconds in uv_work
the 1th count.
sleep 1 seconds in uv_work
sleep 1 seconds in uv_work
the 2th count.
sleep 1 seconds in uv_work
sleep 1 seconds in uv_work
the 3th count.
sleep 1 seconds in uv_work
sleep 1 seconds in uv_work
the 4th count.
sleep 1 seconds in uv_work
sleep 1 seconds in uv_work
the 5th count.
exist while.
nextTick occured.
sleep 1 seconds in uv_work
sleep 1 seconds in uv_work
the 1th count in [timer1].
sleep 1 seconds in uv_work
sleep 1 seconds in uv_work
the 2th count in [timer1].
sleep 1 seconds in uv_work
the 3th count in [timer1].
the 4th count in [timer1].
exist while in [timer1].
setTimeout 0 occured second.
setImmediate occured.
create directory success.call_work_after Thread id : gettid() == 284953360
node addon result good--->hello world from c++
输出2.5
我们终于看到开篇提到的类似java代码的交叉输出的效果了。libuv在处理任务时根本就和js不在一个线程中,所以才出现了libuv线程和js线程交叉输出的效果。我们在梳理一下代码2.5的异步流程,那么可以用下面这个图来展示出来:
[外链图片转存失败,源站可能有防盗链机制,建议将图片保存下来直接上传(img-8dnfecwO-1676965585639)(https://raw.githubusercontent.com/yunnysunny/yunnysunny.github.io/master/images/js_frame3.png)]
图2.3
在node中维护了一个回调的队列,那为什么调用插件的回调排在队列的最后面呢,是有于我们在代码2.4中故意将其代码设置成15秒之后才执行完成,这要远远长于其他的那些回调,所以它只能被追加到回调队列的最后一位。在第二帧中,node中的事件轮询依次将这些回调队列中的任务取出来,进行触发调用。可以看出回调队列是个先进先出的结构。注意回调是按照队列中排队顺序执行的,同时它们执行的环境是js线程,要知道js线程可是个单线程,也就是说一旦某个回调中的同步代码耗时比较长,那么它后面的回调任务就得一直等着它运行完成,所以说一定不要在回调中写特别耗时的同步代码。
3. 代码
本文用到代码发布在http://git.oschina.net/yunnysunny/async-tutorial-code
本文链接:https://my.lmcjl.com/post/3080.html
4 评论