Node.js 抓取工具中的内存泄漏
这是一个用 JavaScript 和 Node.js 编写的简单抓取工具,用于抓取 Wikipedia 中的元素周期表元素数据。依赖项是用于 DOM 操作的 jsdom 和 chain-gang 用于排队。
大多数时候它工作得很好(它不能优雅地处理错误),而且代码也不算太糟糕,我敢说 for 尝试,但它有一个严重的错误 - 它在任何地方都会可怕地泄漏内存每个元素占用计算机内存的 0.3% 到 0.6%,这样当它领先时,它会使用接近 20% 的内存,这显然是不可接受的。
我尝试过使用分析器,但我发现它们要么没有帮助,要么难以解释数据。我怀疑这与 processElement 的传递方式有关,但我很难将队列代码重写为更优雅的代码。
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();
This is a simple scraper written in JavaScript with Node.js, for scraping Wikipedia for periodic table element data. The dependencies are jsdom for DOM manipulation and chain-gang for queuing.
It works fine, most of the time (it doesn't handle errors gracefully), and the code isn't too bad, dare I say for a for attempt, but there is a serious fault with it - it leaks memory horribly, anywhere from 0.3% to 0.6% of the computer's memory for each element, such that by the time it gets to lead it would be using somewhere close to 20%, which is plainly unacceptable.
I've tried working with profilers, but I have either not found them to be helpful or have difficulty interpreting the data. I suspect it has something to do with the way processElement
gets passed around, but I have difficulty in rewriting the queue code into something more elegant.
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();
如果你对这篇内容有疑问,欢迎到本站社区发帖提问 参与讨论,获取更多帮助,或者扫码二维码加入 Web 技术交流群。
绑定邮箱获取回复消息
由于您还没有绑定你的真实邮箱,如果其他用户或者作者回复了您的评论,将不能在第一时间通知您!
发布评论
评论(4)
jsdom 确实存在内存泄漏,该泄漏源于节点
vm.runInContext()
背后的复制输入和复制输出逻辑。人们已经努力使用 C++ 来解决这个问题,我们希望在尝试将其推入节点之前证明该解决方案。目前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,只要您在完成后关闭窗口 (
window.close()
),这个问题就可以解决。jsdom does have a memory leak which stems from the copy in and copy out logic behind node's
vm.runInContext()
. There has been effort to fix this problem using c++ and we are hoping to prove out the solution before attempting to push it into node.A workaround for now is to spawn up a child process for each dom and close it down when you are done.
EDIT:
as of jsdom 0.2.3 this issue is fixed as long as you close the window (
window.close()
) when you are done with it.对于使用节点进行类似 jQuery 的 html 处理,我现在使用 cheerio 而不是 jsdom。到目前为止,我在几个小时内废弃和解析超过 10K 页面时还没有看到任何内存泄漏。
For jQuery-like html processing with node i use now cheerio instead of jsdom. So far, i have not seen any memory leaks while scrapping and parsing over 10K pages for a couple of hours.
我认为我有一个更好的解决方法,通过设置 window.document.innerHTML 属性来重用 jsdom 实例。解决了我的内存泄漏问题!
I think I have a better work-around, reuse your instance of jsdom by setting the window.document.innerHTML property. Solved my memory leak problems!
我知道这不是一个答案,但我也遇到了类似的问题。我有多个抓取工具同时运行,并且内存正在泄漏。
我最终使用了node-jquery而不是JSDOM
https://github.com/coolaj86/node-jquery< /a>
I know its not much of an answer but I had a similar problem. I have multiple scrapers running simultaneously and memory was getting leaked.
I have ended up using node-jquery instead of JSDOM
https://github.com/coolaj86/node-jquery