[{"_1":2,"_3":-5,"_4":-5},"loaderData",{"_5":6,"_7":8},"actionData","errors","root",{"_11":-5},"routes/user.$name.collections",{"_9":10,"_11":12,"_13":14,"_15":16,"_17":18,"_19":16},"topics",[35],"user",{"_20":21,"_13":14,"_22":23,"_24":25,"_26":27,"_28":29,"_30":16,"_31":32,"_33":34},"loginname","mcdaddy","page",1,"limit",20,"total","id","9778","avatar_url","http://www.gravatar.com/avatar/f19b5dda5747a45ed7a0dfa97061f72a?size=48","githubUsername","","create_at","2013-09-08 15:29:38.351","score",5,"topic_count","reply_count",0,"is_star",false,{"_20":36,"_37":38,"_39":-5,"_40":41,"_42":43,"_44":45,"_46":34,"_47":34,"_31":48,"_49":50,"_26":51,"_52":53},"139","author_id","9","tab","content","nodejs的核心之一就是非阻塞的异步IO，于是想知道它是怎么实现的，挖了下nodejs源码，找到些答案，在此跟大家分享下。首先，我用了一段js代码test-fs-read.js做测试，代码如下：\r\n

var path = require('path'),\r\n
fs = require('fs'),\r\n
filepath = path.join(__dirname, 'experiment.log'),\r\n
fd = fs.openSync(filepath, 'r');\r\n
\r\n
fs.read(fd, 12*1024*1024, 0, 'utf-8', function(err, str, bytesRead) {\r\n
    console.log('[main thread] execute read callback');\r\n
});\r\n
console.log('[main thread] execute operation after read');\r\n

\r\n
这段代码的异步IO操作就在fs.read的调用上，读取的experiment.log是一个12M的文本文件，所谓的异步，大家大概能想得到运行时会先打印\r\n
[main thread] execute operation after read\r\n
然后打印回调函数中的\r\n
[main thread] execute read callback\r\n
\r\n
但大家也许还听说过，nodejs是单线程的，那又是怎么实现异步IO的呢？读文件操作是在哪里执行的呢？读完又是怎么调用回调函数的呢？猜想读文件可能是在另一个线程中完成的，读完后通过事件通知nodejs执行回调。为了一探究竟，debug了一把nodejs和libeio源码，重新编译后，运行测试代码node test-fs-read.js，输出如下：\r\n
\r\n
$\"\"$ \r\n
\r\n
可以看到，nodejs的IO操作是通过调用libeio库完成的，debug从fs.read开始，js代码经过v8编译后，fs.read会调用node_file.cc中的Read方法，测试代码的运行经历了以下步骤：\r\n
\r\n
1 node_file.cc中的Read方法调用libeio（eio.c）的eio_read， read请求被放入请求队列req_queue中。\r\n
\r\n
2 主线程创建了1个eio线程，此时主线程的read调用返回。\r\n
\r\n
3 eio线程从req_queue中取出1个请求，开始执行read IO\r\n
\r\n
4 主线程继续执行read调用后的其它操作。\r\n
\r\n
5 主线程poll eio，从响应队列res_queue取已经完成的请求，此时res_queue为空，主线程stop poll\r\n
\r\n
6 eio线程完成了read IO，read请求被放入响应队列res_queue中，并且向主线程发送libev事件want_poll（通过主线程初始化eio时提供的回调函数）。\r\n
\r\n
7 eio线程从req_queue中取下一个请求，此时已经没有请求。\r\n
\r\n
8 主线程响应want_poll事件，从res_queue中取出1个请求，取出请求后res_queue变为空，主线程发送done_poll事件。\r\n
\r\n
9 主线程执行请求的callback函数。\r\n
\r\n
还需要说明的是，当同时有多个IO请求时，主线程会创建多个eio线程，以提高IO请求的处理速度。\r\n
为了更清晰的看到nodejs的IO执行过程，图示如下，序号仅用来标示流程，与上述步骤序号并无对应关系。\r\n
\r\n
$\"\"$ \r\n
\r\n
最后总结几条，不当之处还请大家指正。\r\n
\r\n
1 nodejs通过libev事件得到IO执行状态，而不是轮询，提高了CPU利用率。 \r\n
\r\n
\r\n
2 虽然nodejs是单线程的，但它的IO操作是多线程的，多个IO请求会创建多个libeio线程（最多4个），使通常情况的IO操作性能得到提高。\r\n
\r\n
3 但是当IO操作情况比较复杂的时候，有可能造成线程竞争状态，导致IO性能降低；而且libeio最多创建4个线程，当同时有大量IO请求时，实际性能有待测量。另外，由于每个IO请求对应一个libeio的数据结构，当同时有大量IO操作驻留在系统中时候，会增加内存开销。\r\n
\r\n
4 Libeio为了实现异步IO功能，带来了额外的管理，当IO数据量比较小的时候，整体性能不一定比同步IO好。","title","nodejs异步IO的实现","last_reply_at","2016-05-10 07:45:38.737","good","top",16,"visit_count",39043,"2011-01-30 10:12:05","author",{"_13":54,"_22":55},"changlin","http://www.gravatar.com/avatar/f2a6e3bb024b3ee7b387df2f05d2c608?s=48"]