使用 XMLHttpRequest 的内存高效消息块处理

Memory efficient message chunk processing using a XMLHttpRequest

我有一个 XMLHttpRequest 和一个 progress 事件处理程序,它正在请求一个分块页面,该页面不断发送添加消息块。如果我不设置 responseType,我可以在每个 progress 事件中访问 XMLHttpRequestresponse 属性 并处理额外的消息块。这种做法的问题是浏览器必须将整个响应保存在内存中,最终,浏览器会因为这种内存浪费而崩溃。

所以,我尝试了一个responseType of arraybuffer,希望能对buffer进行分片,避免之前过多的内存浪费。不幸的是,此时 progress 事件处理程序不再能够读取 XMLHttpRequestresponse 属性。 progress 事件的事件参数也不包含缓冲区。这是我对此尝试的一个简短、独立的示例(这是为 node.js 编写的):

var http = require('http');

// -- The server.

http.createServer(function(req, res) {
  if (req.url === '/stream') return serverStream(res);
  serverMain(res);
}).listen(3000);

// -- The server functions to send a HTML page with the client code, or a stream.

function serverMain(res) {
  res.writeHead(200, {'Content-Type': 'text/html'});
  res.write('<html><body>Hello World</body><script>');
  res.end(client.toString() + ';client();</script></html>');
}

function serverStream(res) {
  res.writeHead(200, {'Content-Type': 'text/html'});
  setInterval(function() {
    res.write('Hello World<br />\n');
  }, 1000);
}

// -- The client code which runs in the browser.

function client() {
  var xhr = new XMLHttpRequest();
  xhr.addEventListener('progress', function() {
    if (!xhr.response) return console.log('progress without response :-(');
    console.log('progress: ' + xhr.response.size);
  }, false);
  xhr.open('GET', '/stream', true);
  xhr.responseType = 'arraybuffer';
  xhr.send();
}

progress 事件处理程序无法访问我想要的 response。如何以内存高效的方式处理浏览器中的消息块?请不要建议一个WebSocket。我不希望仅使用一个来处理只读的消息块流。

XMLHttpRequest 似乎并不是为这种用法而设计的。显而易见的解决方案是轮询,这是 XMLHttpRequest 的一种流行用法,但我猜您不想错过流中会在调用之间滑动的数据。

关于我的问题, you answered

基于这个前提,我建议:

想法:合作并发侦听器

  1. 连接到流并设置进度侦听器(称为 listenerA())。
  2. 当一个块到达时,对其进行处理并输出。保留对 listenerA() 收到的第一个和最后一个块的 ID 的引用。计算收到了多少块listenerA()
  3. listenerA() 收到一定数量的块后,生成另一个 "thread"(连接 + 侦听器,listenerB())与第一个并行执行步骤 1 和 2,但是将处理后的数据保存在缓冲区中而不是输出它。
  4. listenerA()接收到与[=20=接收到的第一个块具有相同id的块时,向listenerB()发送信号,丢弃第一个连接并终止listenerA().
  5. listenerB()收到来自listenerA()的终止信号时,将缓冲区转储到输出并继续正常处理。
  6. 在与以前相同的条件下生成 listenerB() listenerC()
  7. 根据需要使用尽可能多的连接 + 侦听器重复。

通过使用两个重叠的连接,您可以防止丢失单个连接然后重新连接可能导致的块丢失。

备注

  • 这假设所有连接的数据流都相同,并且不引入一些个性化设置。
  • 根据流的输出速率和连接延迟,从一个连接转换到另一个连接期间的缓冲区转储可能会很明显。
  • 您还可以测量总响应大小而不是块数来决定何时切换到新连接。
  • 可能需要保留完整的块 ID 列表以进行比较,而不仅仅是第一个和最后一个,因为我们无法保证重叠的时间。
  • XMLHttpRequestresponseType 必须设置为默认值 "" 或 "text",为 return 文本。其他数据类型不会 return 部分 response。参见 https://xhr.spec.whatwg.org/#the-response-attribute

测试服务器在node.js

以下代码是一个 node.js 服务器,它输出一致的元素流以用于测试目的。您可以打开多个连接,跨会话的输出将相同,减去可能的服务器延迟。

http://localhost:5500/stream

将 return id 为递增数字的数据

http://localhost:5500/streamRandom

将 return 数据,其中 id 是随机的 40 个字符长的字符串。这是为了测试不能依赖 id 来订购数据的场景。

var crypto = require('crypto');

// init + update nodeId
var nodeId     = 0;
var nodeIdRand = '0000000000000000000000000000000000000000';

setInterval(function() {

    // regular id
    ++nodeId;

    //random id
    nodeIdRand = crypto.createHash('sha1').update(nodeId.toString()).digest('hex');
}, 1000);


// create server  (port 5500)
var http = require('http');
http.createServer(function(req, res) {

  if(req.url === '/stream') {
      return serverStream(res);
  }
  else if(req.url === '/streamRandom') {
      return serverStream(res, true);
  }
}).listen(5500);


// serve nodeId
function serverStream(res, rand) {

    // headers
    res.writeHead(200, {
        'Content-Type'                : 'text/plain',
        'Access-Control-Allow-Origin' : '*',
    });

    // remember last served id
    var last = null;

    // output interval
    setInterval(function() {

        // output on new node
        if(last != nodeId) {
            res.write('[node id="'+(rand ? nodeIdRand : nodeId)+'"]');
            last = nodeId;
        }
    }, 250);
}

概念验证,使用上述 node.js 服务器代码

<!DOCTYPE html>
<html>
    <head>
        <meta http-equiv="Content-Type" content="text/html; charset=utf-8"/>
    </head>
    <body>
        <button id="stop">stop</button>
        <div id="output"></div>
        <script>

/*
Listening to a never ending page load (http stream) without running out of
memory by using concurrent overlapping connections to prevent loss of data,
using only xmlHttpRequest, under the condition that the data can be identified.

listen arguments
    url         url of the http stream
    chunkMax    number of chunks to receive before switching to new connection

listen properties
    output           a reference to a DOM element with id "output"
    queue            an array filled with non-duplicate received chunks and metadata
    lastFetcherId    an incrementing number used to assign an id to new fetchers
    fetchers         an array listing all active fetchers

listen methods
    fire        internal use    fire an event
    stop        external use    stop all connections
    fetch       internal use    starts a new connection
    fetchRun    internal use    initialize a new fetcher object

Usage

    var myListen = new listen('http://localhost:5500/streamRandom', 20);
        will listen to url "http://localhost:5500/streamRandom"
        will switch connections every 20 chunks

    myListen.stop()
        will stop all connections in myListen
*/
function listen(url, chunkMax) {

    // main ref
    var that = this;

    // output element
    that.output = document.getElementById('output');

    // main queue
    that.queue = [];

    // last fetcher id
    that.lastFetcherId = 0;

    // list of fetchers
    that.fetchers = [];




    //********************************************************* event dispatcher
    that.fire = function(name, data) {
        document.dispatchEvent(new CustomEvent(name, {'detail':data}));
    }




    //******************************************************** kill all fetchers
    that.stop = function() {
        that.fire('fetch-kill', -1);
    }




    //************************************************************** url fetcher
    that.fetch = function(fetchId, url, fetchRef) {

        //console.log('start fetcher #'+fetchId);
        var len = 0;
        var xhr = new XMLHttpRequest();
        var cb_progress;
        var cb_kill;


        // progress listener
        xhr.addEventListener('progress', cb_progress = function(e) {

            // extract chunk data
            var chunkData = xhr.response.substr(len);

            // chunk id
            var chunkId = chunkData.match(/id="([a-z0-9]+)"/)[1];

            // update response end point
            len = xhr.response.length;

            // signal end of chunk processing
            that.fire('chunk-ready', {
                'fetchId'   : fetchId,
                'fetchRef'  : fetchRef,
                'chunkId'   : chunkId,
                'chunkData' : chunkData,
            });
        }, false);


        // kill switch
        document.addEventListener('fetch-kill', cb_kill = function(e) {

            // kill this fetcher or all fetchers (-1)
            if(e.detail == fetchId || e.detail == -1) {

                //console.log('kill fetcher #'+fetchId);

                xhr.removeEventListener('progress', cb_progress);
                document.removeEventListener('fetch-kill', cb_kill);

                xhr.abort();
                that.fetchers.shift(); // remove oldest fetcher from list
                xhr = null;
                delete xhr;
            }
        }, false);


        // go
        xhr.open('GET', url, true);
        xhr.responseType = 'text';
        xhr.send();
    };




    //****************************************************** start a new fetcher
    that.fetchRun = function() {

        // new id
        var id = ++that.lastFetcherId;

        //console.log('create fetcher #'+id);

        // create fetcher with new id
        var fetchRef = {
            'id'           : id,    // self id
            'queue'        : [],    // internal queue
            'chunksIds'    : [],    // retrieved ids, also used to count
            'hasSuccessor' : false, // keep track of next fetcher spawn
            'ignoreId'     : null,  // when set, ignore chunks until this id is received (this id included)
        };
        that.fetchers.push(fetchRef);

        // run fetcher
        that.fetch(id, url, fetchRef);
    };




    //************************************************ a fetcher returns a chunk
    document.addEventListener('chunk-ready', function(e) {

        // shorthand
        var f = e.detail;

        // ignore flag is not set, process chunk
        if(f.fetchRef.ignoreId == null) {

            // store chunk id
            f.fetchRef.chunksIds.push(f.chunkId);

            // create queue item
            var queueItem = {'id':f.chunkId, 'data':f.chunkData};

            // chunk is received from oldest fetcher
            if(f.fetchId == that.fetchers[0].id) {

                // send to main queue
                that.queue.push(queueItem);

                // signal queue insertion
                that.fire('queue-new');
            }
            // not oldest fetcher
            else {

                // use fetcher internal queue
                f.fetchRef.queue.push(queueItem);
            }
        }
        // ignore flag is set, current chunk id the one to ignore
        else if(f.fetchRef.ignoreId == f.chunkId) {

            // disable ignore flag
            f.fetchRef.ignoreId = null;
        }







        //******************** check chunks count for fetcher, threshold reached
        if(f.fetchRef.chunksIds.length >= chunkMax && !f.fetchRef.hasSuccessor) {

            // remember the spawn
            f.fetchRef.hasSuccessor = true;

            // spawn new fetcher
            that.fetchRun();
        }




        /***********************************************************************
        check if the first chunk of the second oldest fetcher exists in the
        oldest fetcher.
        If true, then they overlap and we can kill the oldest fetcher
        ***********************************************************************/
        if(
            // is this the oldest fetcher ?
            f.fetchId == that.fetchers[0].id
            // is there a successor ?
            && that.fetchers[1]
            // has oldest fetcher received the first chunk of its successor ?
            && that.fetchers[0].chunksIds.indexOf(
                that.fetchers[1].chunksIds[0]
            ) > -1
        ) {

            // get index of last chunk of the oldest fetcher within successor queue
            var lastChunkId    = that.fetchers[0].chunksIds[that.fetchers[0].chunksIds.length-1]
            var lastChunkIndex = that.fetchers[1].chunksIds.indexOf(lastChunkId);

            // successor has not reached its parent last chunk
            if(lastChunkIndex < 0) {

                // discard whole queue
                that.fetchers[1].queue     = [];
                that.fetchers[1].chunksIds = [];

                // set ignore id in successor to future discard duplicates
                that.fetchers[1].ignoreId = lastChunkId;
            }
            // there is overlap
            else {

                /**
                console.log('triming queue start: '+that.fetchers[1].queue.length
                    +"   "+(lastChunkIndex+1)
                    +"   "+(that.fetchers[1].queue.length-1)
                );
                /**/
                var trimStart = lastChunkIndex+1;
                var trimEnd   = that.fetchers[1].queue.length-1;

                // trim queue
                that.fetchers[1].queue = that.fetchers[1].queue.splice(trimStart, trimEnd);
                that.fetchers[1].chunksIds = that.fetchers[1].chunksIds.splice(trimStart, trimEnd);

                //console.log('triming queue end: '+that.fetchers[1].queue.length);
            }

            // kill oldest fetcher
            that.fire('fetch-kill', that.fetchers[0].id);
        }





    }, false);




    //***************************************************** main queue processor
    document.addEventListener('queue-new', function(e) {

        // process chunks in queue
        while(that.queue.length > 0) {

            // get chunk and remove from queue
            var chunk = that.queue.shift();

            // output item to document
            if(that.output) {
                that.output.innerHTML += "<br />"+chunk.data;
            }
        }
    }, false);



    //****************************************************** start first fetcher
    that.fetchRun();
};


// run
var process = new listen('http://localhost:5500/streamRandom', 20);

// bind global kill switch to button
document.getElementById('stop').addEventListener('click', process.stop, false);

        </script>
    </body>
</html>