3rd-party-cpu-abuser 中文文档教程
Why
很多网站充斥着广告和其他第 3 方,它们都试图滥用 CPU,导致用户因为页面太慢而离开。 JS Profiling 通常是与您的第 3 方开始谈判(或删除他们......)的关键,但需要时间。 该项目将帮助您快速识别哪个域名占用了大部分 CPU。
Install
npm install 3rd-party-cpu-abuser
使用 Node 6.3.0
Usage from CLI
- From the Chrome dev tools "Performance" tab, record a profiling session, desktop or mobile
- End, then "Save Profile", to have the JSON export
- run
node cli.js path/to/the.json
- Read the output
输出(来自报纸网站上的 android 平板电脑,有很多第 3 方):
Analyzing samples/VA-article.json
┌──────────────────────────┬──────────┐
│ Total CPU busy time (ms) │ 10107.07 │
├──────────────────────────┼──────────┤
│ Total number of domains │ 57 │
├──────────────────────────┼──────────┤
│ Number of big offenders │ 40 │
└──────────────────────────┴──────────┘
┌───────────────┬────────────────────────────────┐
│ CPU Time (ms) │ domain name │
├───────────────┼────────────────────────────────┤
│ 1311.39 │ www.vogue.fr │
├───────────────┼────────────────────────────────┤
│ 865.91 │ z.moatads.com │
├───────────────┼────────────────────────────────┤
│ 835.78 │ pagead2.googlesyndication.com │
├───────────────┼────────────────────────────────┤
│ 788.86 │ securepubads.g.doubleclick.net │
├───────────────┼────────────────────────────────┤
│ ……………… │ …………………………………………………… │
├───────────────┼────────────────────────────────┤
│ 565.78 │ data05.adlooxtracking.com │
├───────────────┼────────────────────────────────┤
│ 559.52 │ tpc.googlesyndication.com │
└───────────────┴────────────────────────────────┘
如何阅读: “总 CPU 繁忙时间”是分析器记录的所有内容
“大罪犯”是总专用 CPU 时间超过 150 毫秒的域的数量(您可以使用 -t
标志更改此参数)
时间 => ; domain :时间以毫秒为单位,domain 是托管 JS 的地方。 探查器计算在每个函数中花费的时间,并通过深入到调用堆栈,它知道它是在哪个文件中定义的,在哪个域中。 我们将所有内容加起来,它可以让您很好地了解哪个第 3 方让您的页面滞后
JSON export
或者您可以访问与 JSON 相同的数据: node cli.js path/to/the.json --json
或者你可以编写自己的 nodeJS 模块并使用它:
const statsPerDomain = require('3rd-party-cpu-abuser')
console.log(
statsPerDomain.data({
file: 'path/to/the.json'
})
)
Start and End Times
默认情况下,整个跟踪中的所有脚本执行都被计算在内,但如果你想只计算命名事件之前或之后的执行时间然后 -s
(--startMark
) 和 -e
(--endMark
) 执行此操作。
例如
node cli.js path/to/the.json -s firstMeaningfulPaint
或
node cli.js path/to/the.json -e loadEventEnd
或
node cli.js path/to/the.json -s loadEventStart -e loadEventEnd
可能在跟踪中的一些示例事件包括:
- domLoading
- firstLayout
- firstPaint
- firstContentfulPaint
- firstTextPaint
- firstImagePaint
- firstMeaningfulPaintCandidate
- firstMeaningfulPaint
- domInteractive
- domContentLoadedEventStart
- domContentLoadedEventEnd
- domComplete
- loadEventStart
- loadEventEnd
用于子集跟踪的时间戳是根帧上事件的时间戳(通常加载的页面)。 如果事件不存在,将抛出错误,如果事件多于第一个,将使用。
Grouping
默认情况下,CPU 时间按子域分组,但可以按无
、类别
、子域
、域
分组、URL
或 EventName
(在 Chrome 的性能选项卡中使用的类别)使用 g
(--groupBy
)
标志示例
node cli.js path/to/the.json -g URL
How
它使用 Paul Irish 的 devtools- timeline-model 从原始时间线数据 JSON 文件中提取有意义的数据。 它只是计算每个域的统计数据并很好地显示它。
Future
如果你想帮助我,欢迎你:
- add Alias for domains (eg : cdn.adnxs.com is appNexus, 2mdn.net is Doubleclick…)
- add score for each domain :
- is CPU consumed before / after DOM ready, before / after onload
- how many times the CPU runs at 100% for more than 150 ms (so, blocks the UI)
- how many forced reflows (like here)
- compare 2 different runs side by side
- add tests
Why
Lot of sites are bloated with ads and other 3rd parties, all of them trying to abuse the CPU, leading the user to leave because the page is so laggy. JS Profiling is often the key to start a negotiation with your 3rd party (or remove them…) but takes time. This project will help you identify quickly wich domain name eats most of the CPU.
Install
npm install 3rd-party-cpu-abuser
Working from Node 6.3.0
Usage from CLI
- From the Chrome dev tools "Performance" tab, record a profiling session, desktop or mobile
- End, then "Save Profile", to have the JSON export
- run
node cli.js path/to/the.json
- Read the output
Output (from an android tablet on a newspaper website with lots of 3rd parties) :
Analyzing samples/VA-article.json
┌──────────────────────────┬──────────┐
│ Total CPU busy time (ms) │ 10107.07 │
├──────────────────────────┼──────────┤
│ Total number of domains │ 57 │
├──────────────────────────┼──────────┤
│ Number of big offenders │ 40 │
└──────────────────────────┴──────────┘
┌───────────────┬────────────────────────────────┐
│ CPU Time (ms) │ domain name │
├───────────────┼────────────────────────────────┤
│ 1311.39 │ www.vogue.fr │
├───────────────┼────────────────────────────────┤
│ 865.91 │ z.moatads.com │
├───────────────┼────────────────────────────────┤
│ 835.78 │ pagead2.googlesyndication.com │
├───────────────┼────────────────────────────────┤
│ 788.86 │ securepubads.g.doubleclick.net │
├───────────────┼────────────────────────────────┤
│ ……………… │ …………………………………………………… │
├───────────────┼────────────────────────────────┤
│ 565.78 │ data05.adlooxtracking.com │
├───────────────┼────────────────────────────────┤
│ 559.52 │ tpc.googlesyndication.com │
└───────────────┴────────────────────────────────┘
How to read it : "Total CPU busy time" is everything the profiler recorded
"big offenders" is the number of domain whose total dedicated CPU time exceeded 150 ms (you can change this parameter with a -t
flag)
time => domain : time is in milliseconds, domain is where the JS is hosted. The profiler counts the time spent in each function, and by getting down to the callstack, it knows in which file, on which domain it was defined. We add up everything and it gives you a good idea of which 3rd party let your page lag
JSON export
Alternatively you can access the same data as JSON : node cli.js path/to/the.json --json
Or you can write your own nodeJS module and use it :
const statsPerDomain = require('3rd-party-cpu-abuser')
console.log(
statsPerDomain.data({
file: 'path/to/the.json'
})
)
Start and End Times
By default, all script execution in the whole trace is counted but if you want to count just the execution time before or after a named event then -s
(--startMark
) and -e
(--endMark
) do this.
For example
node cli.js path/to/the.json -s firstMeaningfulPaint
or
node cli.js path/to/the.json -e loadEventEnd
or
node cli.js path/to/the.json -s loadEventStart -e loadEventEnd
Some example event that may be in a trace include:
- domLoading
- firstLayout
- firstPaint
- firstContentfulPaint
- firstTextPaint
- firstImagePaint
- firstMeaningfulPaintCandidate
- firstMeaningfulPaint
- domInteractive
- domContentLoadedEventStart
- domContentLoadedEventEnd
- domComplete
- loadEventStart
- loadEventEnd
The timestamp used to subset the trace is the timestamp of the event on the root frame (normally the page that's loaded). If the event doesn't exist an error will be thrown, if there's more than on the first will be used.
Grouping
By default, CPU time is grouped by Subdomain but it can be grouped by None
, Category
, Subdomain
, Domain
, URL
or EventName
(categories used in Chrome's performance tab) using the g
(--groupBy
) flag
For example
node cli.js path/to/the.json -g URL
How
It uses Paul Irish's devtools-timeline-model that extracts meaningful data from a raw Timeline Data JSON file. It just compute stats for each domain and display it nicely.
Future
If you want to help me with those, you're welcome :
- add Alias for domains (eg : cdn.adnxs.com is appNexus, 2mdn.net is Doubleclick…)
- add score for each domain :
- is CPU consumed before / after DOM ready, before / after onload
- how many times the CPU runs at 100% for more than 150 ms (so, blocks the UI)
- how many forced reflows (like here)
- compare 2 different runs side by side
- add tests