GZIP compression and response size reporting issues

[macbre]

• allow responses to be gzipped
• get the correct bodySize

See ariya/phantomjs#10156 and ariya/phantomjs#10169

[macbre]

HTML:

{
"url":"http:https://elecena.pl/",
"method":"GET"
"bodySize":10734,
"Content-Encoding":"gzip"
}

$ curl "http:https://elecena.pl/" --compress | wc -c
10734

JS:

 {
"url":"http:https://cdn.macbre.net/elecena/r517/package/bootstrap,elecena,ui,suggest,ads,home.js",
"method":"GET",
"bodySize":988
}

$ curl "http:https://cdn.macbre.net/elecena/r517/package/bootstrap,elecena,ui,suggest,ads,home.js" --compress  | wc -c
29300

Image (the same domain):

{
"url":"http:https://beta.elecena/r517/skins/elecena/img/header.jpg",
"method":"GET",
"bodySize":9745
}

$ curl "http:https://beta.elecena/r517/skins/elecena/img/header.jpg" --compress  | wc -c
9745

Image (different domain):

{
"url":"http:https://cdn.macbre.net/elecena/r517/skins/elecena/img/header.jpg",
"method":"GET",
"bodySize":1044

[macbre]

Use HTTP proxy for running phantomas? https://github.com/nodejitsu/node-http-proxy/tree/caronte

[patramsey]

I like the idea of using the http proxy to get the correct bodySize when the response gzipped. Does the response object in the proxy have that information?

[macbre]

I don't think so. This would need to be a custom proxy that will add that kind of information.

[gurdenbatra]

Hey @macbre, just wanted to check in with the progress on this issue? Can I offer you my help?

[macbre]

No progress on this one, unfortunately. Using proxy in this case would be a great example of overengineering :) Given the fact that PhantomJS v2.0 is just around the corner.

@gurdenbatra, do you have a different solution on how to fix this issue?

[gurdenbatra]

@macbre Do you know if I can at least accurately measure the differential of bodySize? As in, not look at the accuracy of the value but just monitor the rate of change?

I do not have a solution in mind as the problem in hand is not completely clear to me. I'd love to help if you could point me in the right direction. Thanks.

[macbre]

@gurdenbatra, unfortunately, as my tests above show bodySize is not deterministic at all :)

[macbre]

https://github.com/ariya/phantomjs/issues?milestone=12&state=open - PhantomJS v2.0 is 15 days behind the deadline. I'm considering a workaround in phantomas... Something like a "light" HTTP proxy that will add X-Phantomas-... headers with required data (content encoding, content length, uncompressed size, etc.)

[macbre]

Will re-check when #313 is implemented

[macbre]

Still valid (v1.3.0):

phantomas "http:https://elecena.pl/" -v --block-domain googleapis.com,google-analytics.com,googletagservices.com --runs 5

.-----------------------------------------------------------------------------------------------------------.
| Report from 5 run(s) for <http:https://elecena.pl/> using phantomas v1.3.0                                      |
|-----------------------------------------------------------------------------------------------------------|
|             Metric             |     min      |     max      |   average    |    median    |    stddev    |
|--------------------------------|--------------|--------------|--------------|--------------|--------------|
| requests                       |           10 |           10 |           10 |           10 |            0 |
| gzipRequests                   |            1 |            1 |            1 |            1 |            0 |
| bodySize                       |        30313 |        44493 |      38537.2 |        41656 |      5259.88 |
| contentLength                  |      1430746 |      1433582 |    1432163.8 |      1432164 |       896.82 |
| ajaxRequests                   |            0 |            0 |            0 |            0 |            0 |
| htmlCount                      |            1 |            1 |            1 |            1 |            0 |
| htmlSize                       |        10907 |        10908 |      10907.8 |        10908 |          0.4 |
| cssCount                       |            1 |            1 |            1 |            1 |            0 |
| cssSize                        |         1023 |         3859 |       2724.6 |         2441 |      1061.13 |
| jsCount                        |            1 |            1 |            1 |            1 |            0 |
| jsSize                         |         2427 |         5263 |       3561.4 |         3845 |      1061.13 |
| jsonCount                      |            0 |            0 |            0 |            0 |            0 |
| jsonSize                       |            0 |            0 |            0 |            0 |            0 |
| imageCount                     |            7 |            7 |            7 |            7 |            0 |
| imageSize                      |      1414970 |      1414970 |      1414970 |      1414970 |            0 |

stddev for sizes should be zero

[cvan]

I have solved this on my own project, and oh boy it was a doozie. I will most certainly upstream a patch to phantomas.

[macbre]

Great 👍

BTW, it works correctly under SlimerJS (introduced in phantomas v1.4.0)

[gmetais]

That's a great news!!

[macbre]

Unfortunately, not much can be done on phantomas side with this issue. A patch from @cvan is kind of hacky :)

Let's keep our fingers crossed for PhantomJS v2.0 (due by September 22, 2014) and use SlimerJS to run phantomas in the meantime.

[cvan]

In my testing, SlimerJS has the same issues

[jcleveley-zz]

Has anyone tested this issue on PhantomJS 2 (preview) ?
https://github.com/bprodoehl/phantomjs/releases

[macbre]

@jcleveley, good point. Binaries from GitHub repository you;ve mentioned do not run on my debian box:

./phantomjs: error while loading shared libraries: libicudata.so.50: cannot open shared object file: No such file or directory

Meanwhile I'm compiling the latest master from phantomjs' repository on droplet instance. Stay tuned :)

[macbre]

See #432 for the similar issue - content size reporting for gzipped content

[soulgalore]

hey @macbre , did phantomjs 2 work for you? I haven't tested yet for sitespeed.io, but planning checking it out this weekend or early next week.

[macbre]

@soulgalore, just tested phantomas with PhantomJS2 binaries compiled in Oct 2014 and there's no improvement.

For http:https://code.jquery.com/jquery-2.1.1.js I'm getting the following data:

• PhantomJS 2.0.0-dev

{"id":3,"url":"http:https://code.jquery.com/jquery-2.1.1.js","method":"GET","requestHeaders":{"Accept":"*/*","Referer":"http:https://localhost:8888/jquery-multiple.html","User-Agent":"phantomas/1.8.0 (PhantomJS/1.9.8; linux x64)"},"sendTime":"2014-12-06T11:04:01.175Z","bodySize":1695,"isBlocked":false,"protocol":"http","domain":"code.jquery.com","recvStartTime":"2014-12-06T11:04:01.494Z","timeToFirstByte":319,"recvEndTime":"2014-12-06T11:04:01.663Z","timeToLastByte":488,"receiveTime":169,"type":"js","headers":{"Date":"Sat, 06 Dec 2014 11:04:01 GMT","Content-Type":"application/x-javascript","Transfer-Encoding":"chunked","Connection":"keep-alive","Last-Modified":"Fri, 24 Oct 2014 00:16:07 GMT","Vary":"Accept-Encoding","ETag":"W/\"54499a47-3c637\"","Expires":"Thu, 31 Dec 2037 23:55:55 GMT","Cache-Control":"max-age=315360000, public","Server":"NetDNA-cache/2.2","X-Cache":"HIT","Content-Encoding":"gzip"},"contentType":"application/x-javascript","isJS":true,"gzip":true,"status":200,"statusText":"OK","contentLength":1695}

• PhantomJS 1.9.8

{"id":3,"url":"http:https://code.jquery.com/jquery-2.1.1.js","method":"GET","requestHeaders":{"User-Agent":"phantomas/1.8.0 (PhantomJS/1.9.8; linux x64)","Accept":"*/*","Referer":"http:https://localhost:8888/jquery-multiple.html"},"sendTime":"2014-12-06T11:05:40.010Z","bodySize":45018,"isBlocked":false,"protocol":"http","domain":"code.jquery.com","recvStartTime":"2014-12-06T11:05:40.241Z","timeToFirstByte":231,"recvEndTime":"2014-12-06T11:05:40.365Z","timeToLastByte":355,"receiveTime":124,"type":"js","headers":{"Date":"Sat, 06 Dec 2014 11:05:40 GMT","Content-Type":"application/x-javascript","Transfer-Encoding":"chunked","Connection":"keep-alive","Last-Modified":"Fri, 24 Oct 2014 00:16:07 GMT","Vary":"Accept-Encoding","ETag":"W/\"54499a47-3c637\"","Expires":"Thu, 31 Dec 2037 23:55:55 GMT","Cache-Control":"max-age=315360000, public","Server":"NetDNA-cache/2.2","X-Cache":"HIT","Content-Encoding":"gzip"},"contentType":"application/x-javascript","isJS":true,"gzip":true,"status":200,"statusText":"OK","contentLength":45018}

• SlimerJS 0.9.2 (correctly reports uncompressed size)

{"id":3,"url":"http:https://code.jquery.com/jquery-2.1.1.js","method":"GET","requestHeaders":{"Host":"code.jquery.com","User-Agent":"phantomas/1.8.0 (SlimerJS/0.9.2; linux x64)","Accept":"*/*","Accept-Language":"en-US,en;q=0.5","Accept-Encoding":"gzip, deflate","Referer":"http:https://localhost:8888/jquery-multiple.html"},"sendTime":"2014-12-06T11:15:24.026Z","bodySize":247351,"isBlocked":false,"protocol":"http","domain":"code.jquery.com","recvStartTime":"2014-12-06T11:15:24.201Z","timeToFirstByte":175,"recvEndTime":"2014-12-06T11:15:24.366Z","timeToLastByte":340,"receiveTime":165,"type":"js","headers":{"Date":"Sat, 06 Dec 2014 11:15:24 GMT","Content-Type":"application/x-javascript","Transfer-Encoding":"chunked","Connection":"keep-alive","Last-Modified":"Fri, 24 Oct 2014 00:16:07 GMT","Vary":"Accept-Encoding","Etag":"W/\"54499a47-3c637\"","Expires":"Thu, 31 Dec 2037 23:55:55 GMT","Cache-Control":"max-age=315360000, public","Server":"NetDNA-cache/2.2","X-Cache":"HIT","Content-Encoding":"gzip"},"contentType":"application/x-javascript","isJS":true,"gzip":true,"status":200,"statusText":"OK","contentLength":247351}

Chrome reports gzipped content to weight ~88 kB and 242 kB uncompressed. Numbers above are not even near the real values ;)

[soulgalore]

hmm ok, thanks for the bad news :)

[fancyoung]

In PhantomJS v2.0,
I use foreach to search Content-Length in headers to get size instead of bodySize.

Something like:

  function getRealSize(response) {
    var tmp, size;
    for(var i=0; i<response.headers.length; i++) {
      tmp = response.headers[i];
      if(tmp.name == 'Content-Length') {
        return parseInt(tmp.value, 10);
      }   
    }   
  }

  size = getRealSize(response) || response.bodySize;

[macbre]

@fancyoung, PhantomJS v1.x does not provide Content-Length for gzip encoded content. Is it more reliable in PhantomJS 2.0?

[fancyoung]

@macbre
I use

page.customHeaders = {
    'Accept-Encoding': 'gzip;q=0'
};

to disable gzip, and it works for me.

[djberriman]

See ariya/phantomjs#10156 patch to ensure Content-Length is correctly returned when it is present. Still have to disable gzip as Content-Length header appears to be lost by QT.

[macbre]

No good news here:

phantomas using PhantomJS/2.1.1:

15:07:47.826 contentLength missing: {"url":"http:https://cdn.macbre.net/elecena/r535/package/fonts,bootstrap,elecena,ui,suggest,home.css","bodySize":13463}

Offenders for gzipRequests (3):
 * http:https://elecena.pl/ (gzip: 7.15 kB / uncompressed: 7.15 kB)
 * http:https://cdn.macbre.net/elecena/r535/package/fonts,bootstrap,elecena,ui,suggest,home.css (gzip: 13.15 kB / uncompressed: 13.15 kB)
 * http:https://cdn.macbre.net/elecena/r535/package/bootstrap,elecena,ui,suggest,home.js (gzip: 4.81 kB / uncompressed: 4.81 kB)

curl:

$ curl 'http:https://cdn.macbre.net/elecena/r535/package/fonts,bootstrap,elecena,ui,suggest,home.css' --compress -svo /dev/null 2>&1 | grep Content
< Content-Encoding: gzip
< Content-Length: 30702

$ curl 'http:https://cdn.macbre.net/elecena/r535/package/fonts,bootstrap,elecena,ui,suggest,home.css' -svo /dev/null 2>&1 | grep Content
< Content-Length: 149586

PhantomJS stats are still not reliable.

However, when using SlimerJS we do get correct stats:

Offenders for gzipRequests (3):
 * http:https://elecena.pl/ (gzip: 11.61 kB / uncompressed: 11.61 kB)
 * http:https://cdn.macbre.net/elecena/r535/package/bootstrap,elecena,ui,suggest,home.js (gzip: 9.48 kB / uncompressed: 28.24 kB)
 * http:https://cdn.macbre.net/elecena/r535/package/fonts,bootstrap,elecena,ui,suggest,home.css (gzip: 29.98 kB / uncompressed: 146.08 kB)

[macbre]

See #614