Limit data URLs in css-variables

[tunetheweb]

As mentioned in HTTPArchive/data-pipeline#262 (comment)

---

Test websites:

• https://www.morgenpost.de/politik/article407241629/game-changer-fuer-die-ukraine-bahnt-sich-an-putin-droht.html"

[max-ostapenko]

It seems a summary part of the object is a duplicate (same content, different structure) of data available in parsed_css metric.

I'm able to find the variables/selectors/values there:

SELECT
  url, css
FROM `httparchive.all.parsed_css`
WHERE date = "2024-09-01"
AND client = 'mobile'
AND NOT is_root_page
AND rank <= 10000
and page = 'https://www.morgenpost.de/politik/article407241629/game-changer-fuer-die-ukraine-bahnt-sich-an-putin-droht.html'
AND css LIKE "%--playkit-icon-quality-HD-active-url%"

I can't give a guarantee, but it should be possible to reuse it for various metrics calculation.

The other part of the object - custom_metrics.css-variables.computed seems more unique.

So a quick optimization could be something as follows:

1. double-check if anybody is using custom_metrics.css-variables.summary (last 90 days of logs don't show any queries),
2. drop it in favor of parsed_css,
3. BTW URLs are not shortened in parsed_css either. Apply this change there instead.

[tunetheweb]

@rviscomi do you know the difference between these two custom metrics and whether we need both?

[rviscomi]

css-variables gives us runtime data about custom property usage

parsed_css gives us the raw stylesheet AST

They're similar in that they're both used for Web Almanac CSS analysis, but both are needed

[tunetheweb]

OK, then I'm less worried about triming parsed_css data URLs since it's in it's own table and massive anyway. Plus I'm not 100% confident of what change to make and where.

I do think there's an argument to move both css_variables and parsed_css into the one table (and thus reduce the size of custom_metrics column in the main all.pages table some more) but let's leave that until the after the Web Almanac is finished.

So in meantime I think this is good to merge?

[github-actions]

Custom metrics for https://almanac.httparchive.org/en/2022/

WPT test run results: http://webpagetest.httparchive.org/results.php?test=240925_03_1W
Changed custom metrics values:

{
  "_css-variables": {
    "summary": {}
  }
}

Webpage test results that are too big for a comment are available as the action's artifact.

[max-ostapenko]

We definitely need to split css_variables due to the size, added to the list.

And yeah, good to merge.