Node.js scraper中的内存泄漏

Question

这是一个用Node.js用JavaScript编写的简单刮刀,用于抓取维基百科的周期表元素数据.依赖关系是用于DOM操作的jsdom和用于排队的chain-gang.

它工作得很好,大部分时间(它不能优雅地处理错误),并且代码也不错,我敢说是为了尝试,但是它有一个严重的错误 - 它可以在任何地方泄漏内存从每个元素的计算机内存的0.3%到0.6%,这样当它达到领先时它将使用接近20%的某个地方,这显然是不可接受的.

我曾尝试使用分析器,但我要么没有发现它们有用,要么难以解释数据.我怀疑它与processElement传递方式有关,但我很难将队列代码重写为更优雅的东西.

var fs = require('fs'),
    path = require('path'),
    jsdom = require("jsdom"),
    parseUrl = require('url').parse,
    chainGang = require('chain-gang');

var chain = chainGang.create({
    workers: 1
});

var Settings = {
    periodicUrl: 'http://en.wikipedia.org/wiki/Template:Periodic_table',
    periodicSelector: '#bodyContent > table:first',
    pathPrefix: 'data/',
    ignoredProperties: ['Pronunciation']
};

function writeToFile(output) {
    var keys = 0;

    // Huge nests for finding the name of the element... yeah
    for(var i in output) {
        if(typeof output[i] === 'object' && output[i] !== null){
            for(var l in output[i]) {
                if(l.toLowerCase() === 'name') {
                    var name = output[i][l];
                }
            }

            keys += Object.keys(output[i]).length;
        }
    }

    console.log('Scraped ' + keys + ' properties for ' + name);
    console.log('Writing to ' + Settings.pathPrefix + name + '.json');
    fs.writeFile(Settings.pathPrefix + name + '.json', JSON.stringify(output));
}

// Generic create task function to create a task function that
// would be passed to the chain gang
function createTask (url, callback) {
    console.log('Task added - ' + url);

    return function(worker){
        console.log('Requesting: ' +url);

        jsdom.env(url, [
            'jquery.min.js' // Local copy of jQuery
        ], function(errors, window) {
            if(errors){
                console.log('Error! ' + errors)
                createTask(url, callback);
            } else {
                // Give me thy $
                var $ = window.$;

                // Cleanup - remove unneeded elements
                $.fn.cleanup = function() {
                    return this.each(function(){
                        $(this).find('sup.reference, .IPA').remove().end()
                            .find('a, b, i, small, span').replaceWith(function(){
                                return this.innerHTML;
                            }).end()
                            .find('br').replaceWith(' ');
                    });
                }

                callback($);
            }

            worker.finish();
        });
    }
}

function processElement ($){
    var infoBox = $('.infobox'),
        image = infoBox.find('tr:contains("Appearance") + tr img:first'),
        description = $('#toc').prevAll('p').cleanup(),
        headers = infoBox.find('tr:contains("properties")'),
        output = {
            Appearance: image.attr('src'),
            Description: $('.infobox + p').cleanup().html()
        };

    headers.each(function(){
        var that = this,
            title = this.textContent.trim(),
            rowspan = 0,
            rowspanHeading = '';

        output[title] = {};

        $(this).nextUntil('tr:has(th:only-child)').each(function(){
            var t = $(this).cleanup(),
                headingEle = t.children('th'),
                data = t.children('td').html().trim();

            if(headingEle.length) {
                var heading = headingEle.html().trim();
            }

            // Skip to next heading if current property is ignored
            if(~Settings.ignoredProperties.indexOf(heading)) {
                return true;
            }

            if (rowspan) {
                output[title][rowspanHeading][data.split(':')[0].trim()] = data.split(':')[1].trim();
                rowspan--;
            } else if (headingEle.attr('rowspan')){
                rowspan = headingEle.attr('rowspan') - 1;
                rowspanHeading = heading;

                output[title][heading] = {};
                output[title][heading][data.split(':')[0]] = data.split(':')[1];
            } else if (~heading.indexOf(',')){
                data = data.split(',');

                heading.split(',').forEach(function(v, i){
                    output[title][v.trim()] = data[i].trim();
                });
            } else {
                output[title][heading] = data;
            }
        });
    });

    writeToFile(output);
}

function fetchElements(elements) {
    elements.forEach(function(value){
        // Element URL used here as task id (second argument)
        chain.add(createTask(value, processElement), value);
    });
}

function processTable($){
    var elementArray = $(Settings.periodicSelector).find('td').map(function(){
        var t = $(this),
            atomicN = parseInt(t.text(), 10);

        if(atomicN && t.children('a').length) {
            var elementUrl = 'http://' + parseUrl(Settings.periodicUrl).host + t.children('a:first').attr('href');

            console.log(atomicN, t.children('a:first').attr('href').split('/').pop(), elementUrl);
            return elementUrl;
        }
    }).get();

    fetchElements(elementArray);
    fs.writeFile(Settings.pathPrefix + 'elements.json', JSON.stringify(elementArray));
}

// Get table - init
function getPeriodicList(){
    var elementsList = Settings.pathPrefix + 'elements.json';

    if(path.existsSync(elementsList)){
        var fileData = JSON.parse(fs.readFileSync(elementsList, 'utf8'));
        fetchElements(fileData);
    } else {
        chain.add(createTask(Settings.periodicUrl, processTable));
    }
}

getPeriodicList();

Answer 1

jsdom确实有一个内存泄漏源于节点的复制和复制逻辑vm.runInContext().已经努力使用c ++解决这个问题,我们希望在尝试将其推入节点之前证明解决方案.

现在的解决方法是为每个dom生成一个子进程,并在完成后将其关闭.

编辑:

从jsdom 0.2.3开始,只要window.close()你完成它就关闭窗口(),这个问题就解决了.

Answer 2

对于类似jQuery的html处理,使用节点我现在使用cheerio而不是jsdom.到目前为止,我在几个小时内报废和解析超过10K的页面时没有看到任何内存泄漏.