matomo-org / matomo

Empowering People Ethically with the leading open source alternative to Google Analytics that gives you full control over your data. Matomo lets you easily collect data from websites & apps and visualise this data and extract insights. Privacy is built-in. Liberating Web Analytics. Star us on Github? +1. And we love Pull Requests!
https://matomo.org/
GNU General Public License v3.0
19.66k stars 2.62k forks source link

Problems with assets files in installs with a separate webserver #1527

Closed anonymous-matomo-user closed 13 years ago

anonymous-matomo-user commented 14 years ago

For scalability I'm running the webserver on a separate server from the php backends. That worked well so far as I'm rsyncing every piwik update to the webserver too. No problems with static JS and CSS files.

But the new assets are compiled and built on the php backends as static files which the webserver tries to deliver on his own but he doesn't have the compiled assets yet. For now I've solved this situation by another rsync which transfers the tmp folder too but I would like some other solution which at least allows me to have the assets delivered by php too (even if it only does a passthrough of its compiled assets) so that the php-backends come into play again.

julienmoumne commented 14 years ago

Attachment: proxy.patch

mattab commented 14 years ago

But the new assets are compiled and built on the php backends as static files which the webserver tries to deliver on his own but he doesn't have the compiled assets yet.

When a server doesn't have the assets, it should generate them. It is important to ensure config files are synched, as assets files depend on which plugins are enabled.

Why are assets not created automatically on all your servers?

anonymous-matomo-user commented 14 years ago

Because the webserver is only delivering static files and dynamic requests are answered from php backends through fastcgi on other servers. The problem is the mixture of static and dynamic files here. Static files are generated from the dynamic requests but static files are not handled by the servers returning the php output.

robocoder commented 14 years ago

It's doable. Instead of the page linking to the static files, the page could link to .php files that would serve the minified content, e.g., /tmp/assets/{hashcode}.js.php.

(I contemplated something similar, to handle the permissions issue on /tmp/assets.)

mattab commented 14 years ago

Is there still a permission issue with assets with the new fixes?

Rather than serving the file from a .php script (we should then ensure that caching works for all etc. which will add complexity), how does the system work to decide which requests go where? is it based on the filename?

robocoder commented 14 years ago

The permission problems should now be fixed.

Making the asset cacheable won't be too hard ... I can lift code that I already wrote in js/index.php.

anonymous-matomo-user commented 14 years ago

Thanks vipsoft for taking this one!

Matt: yes, its filename based but just forwarding the js to the php-backends won't help. They are raw fastcgi php listeners, not full webservers.

mattab commented 14 years ago

why don't we call the static CSS file [hash].php ? same for JS? filename extension doesn't matter for browsers, and this would maybe work?

anonymous-matomo-user commented 14 years ago

It will be parsed as php by all webservers then. Isn't php always sending some headers which avoid caching in the browser and last modified headers and similar won't be sent for dynamic content? (wouldn't matter for me but just wanted to throw this in)

robocoder commented 14 years ago

Proposal:

Refactor AssetManager.php into a base class + factory method, and introduce a single config option:

; if blank, disables merging of assets; css and js files are served individually
; merge - js and css assets are minified and merged; by default, content is served directly
; proxy - js and css assets are minified and merged; content is served via a php proxy script
asset_manager = merge

Files:

For the PHP proxy, lift code from js/index.php which sends the Last-Modified header, handles If-Modified-Since (in the request), and Accept-Encoding (on-the-fly compression).

julienmoumne commented 14 years ago

Is there really a need to have two mechanisms for handling assets inclusion ?

Wouldn't the php proxy method be sufficient or is there any advantages I don't see in allowing standard includes of merged assets ?

robocoder commented 14 years ago

I don't mind if the php proxy serves all use cases. (I wasn't that keen on serving content directly out of tmp.)

Where should the proxy be located? CoreHome? e.g.,

<link rel="stylesheet" type="text/css" href="index.php?module=CoreHome&action=getCss" />
<script type="text/javascript" src="index.php?module=CoreHome&action=getJs"></script>
mattab commented 14 years ago

+1 for keeping one mechanism only. Proxy proposal in CoreHome looks good.

julienmoumne commented 14 years ago

I implemented the proxy solution proposed by vipsoft.

I had to factorize the code from /js/index.php (cache and compression management) in core/Piwik.php. The content-type is provided as a string, there is always the possibility of using a magic mime database.

anonymous-matomo-user commented 14 years ago

I've applied the patch, thanks a lot for it! I had one problem though ... All my systems are running with zlib.output_compression = On which leads to garbled (probably double compressed) output of the css and js files. As a workaround I added ini_set('zlib.output_compression',0); here if output compression is enabled in core/Piwik.php.

IMO it would be best if either thats set (dunno if that works everywhere) or additional output compression by piwik is disabled when that variable is detected.

robocoder commented 14 years ago

re: zlib.output_compression; yeah, we should test for that; if enabled, let php handle the Accept-Encoding header using its built-in support.

Thanks again Julien.

mattab commented 14 years ago

(In [2876]) Gzipped files should be written to /js/ refs #1527

julienmoumne commented 14 years ago

Test for zlib.output_compression added.

Compressed files moved to /tmp/.

Vary: Accept-Encoding header added.

gzdeflate used instead of gzcompress because of MSIE misinterpretation of compression RFCs.

robocoder commented 14 years ago

Except for some typos (e.g., "satic" and "futur") and excess blank lines, it looks good. Go ahead and check in.

Oh...js/index.php might be missing a closing curly brace. (Hard to tell from the patch.)

anonymous-matomo-user commented 14 years ago

Any chance to include that in 1.0? Whats the issue why it was moved scheduled to 1.1?

robocoder commented 14 years ago

matt and I haven't had time to code review.

you're welcome to apply the patch against 0.9.9 and test it out.

anonymous-matomo-user commented 14 years ago

The patch works fine in my environment, applied against 0.9.9. I got some rejects but these were just for the file headers.

julienmoumne commented 14 years ago

(In [3048]) fixes #1527 - merged assets are now returned to browser using a custom php proxy script, refs #660 - asset manager updated in two ways: - inclusion directive via proxy, - hash management dropped

robocoder commented 14 years ago

(In [3054]) refs #1527 - add check for finfo_open, in case PECL extension not installed and not php 5.3.x

robocoder commented 14 years ago

(In [3140]) refs #1507, refs #1527 - revert hack from #1507, and fix unit test on Windows per Julien's email

julienmoumne commented 14 years ago

(In [3147]) refs #1527 - bypassing a unit test on windows, as far as I know, it is not possible to remove reading rights on a windows file using PHP

robocoder commented 13 years ago

(In [3365]) refs #1527 - minor fix detecting when zlib.output_compression is enabled

robocoder commented 13 years ago

(In [3976]) refs #1527 - refactoring test for php output compression