node中process.nextTick的理解(这个文章需要多读几遍。。)

转自:http://howtonode.org/understanding-process-next-tick

Understanding process.nextTick()

I have seen quite a few people being confused about process.nextTick(). Let's take a look at whatprocess.nextTick() does, and when to use it.

As you might already know, every Node application runs on a single thread. What this means is that apart from I/O - at any time, only one task/event is processed by Node's event loop. You can imagine this event loop to be a queue of callbacks that are processed by Node on every tick of the event loop. So, even if you are running Node on a multi-core machine, you will not get any parallelism in terms of actual processing - all events will be processed only one at a time. This is why Node is a great fit for I/O bound tasks, and definitely not for CPU intensive tasks. For every I/O bound task, you can simply define a callback that will get added to the event queue. The callback will fire when the I/O operation is done, and in the mean time, the application can continue to process other I/O bound requests.

Given this model, what process.nextTick() actually does is defer the execution of an action till the next pass around the event loop. Let's take a simple example. If we had a function foo() which we wanted to invoke in the next tick, this is how we do it:

function foo() {
    console
.log('foo');
}

process
.nextTick(foo);
console
.log('bar');

If you ran the above snippet, you will notice that bar will be printed in your console before foo, as we have delayed the invokation of foo() till the next tick of the event loop:

bar
foo

In fact, you can get the same result by using setTimeout() this way:

setTimeout(foo, 0);
console
.log('bar');

However, process.nextTick() is not just a simple alias to setTimeout(fn, 0) - it's far more efficient.

Let's see where we can use process.nextTick():

Interleaving execution of a CPU intensive task with other events

Let's say we have a task compute() which needs to run almost continuously, and does some CPU intensive calculations. If we wanted to also handle other events, like serving HTTP requests in the same Node process, we can use process.nextTick() to interleave the execution of compute() with the processing of requests this way:

var http = require('http');

function compute() {
   
// performs complicated calculations continuously
   
// ...
    process
.nextTick(compute);
}

http
.createServer(function(req, res) {
     res
.writeHead(200, {'Content-Type': 'text/plain'});
     res
.end('Hello World');
}).listen(5000, '127.0.0.1');

compute
();

In this model, instead of calling compute() recursively, we use process.nextTick() to delay the execution of compute() till the next tick of the event loop. By doing so, we ensure that if any other HTTP requests are queued in the event loop, they will be processed before the next time compute() gets invoked. If we had not used process.nextTick() and had simply called compute() recursively, the program would not have been able to process any incoming HTTP requests. Try it for yourself!

So, alas, we don't really get any magical multi-core parallelism benefits by using process.nextTick(), but we can still use it to share CPU usage between different parts of our application.

Keeping callbacks truly asynchronous

When you are writing a function that takes a callback, you should always ensure that this callback is fired asynchronously. Let's look at an example which violates this convention:

function asyncFake(data, callback) {        
   
if(data === 'foo') callback(true);
   
else callback(false);
}

asyncFake
('bar', function(result) {
   
// this callback is actually called synchronously!
});

Why is this inconsistency bad? Let's consider this example taken from Node's documentation:

var client = net.connect(8124, function() { 
    console
.log('client connected');
    client
.write('world!\r\n');
});

In the above case, if for some reason, net.connect() were to become synchronous, the callback would be called immediately, and hence the client variable will not be initialized when the it's accessed by the callback to write to the client!

We can correct asyncFake() to be always asynchronous this way:

function asyncReal(data, callback) {
    process
.nextTick(function() {
        callback
(data === 'foo');      
   
});
}

When emitting events

Let's say you are writing a library that reads from a source and emits events that contains the chunks that are read. Such a library might look like this:

var EventEmitter = require('events').EventEmitter;

function StreamLibrary(resourceName) {
   
this.emit('start');

   
// read from the file, and for every chunk read, do:        
   
this.emit('data', chunkRead);      
}
StreamLibrary.prototype.__proto__ = EventEmitter.prototype;   // inherit from EventEmitter

Let's say that somewhere else, someone is listening to these events:

var stream = new StreamLibrary('fooResource');

stream
.on('start', function() {
    console
.log('Reading has started');
});

stream
.on('data', function(chunk) {
    console
.log('Received: ' + chunk);
});

In the above example, the listener will never get the start event as that event would be emitted byStreamLibrary immediately during the constructor call. At that time, we have not yet assigned a callback to the start event yet. Therefore, we would never catch this event! Once again, we can useprocess.nextTick() to defer the emit till the listener has had the chance to listen for the event.

function StreamLibrary(resourceName) {      
   
var self = this;

    process
.nextTick(function() {
       
self.emit('start');
   
});

   
// read from the file, and for every chunk read, do:        
   
this.emit('data', chunkRead);      
}

I hope that demystifies process.nextTick(). If I have missed out something, please do share in the comments.

评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值