feat: support custom HTTP filter ordering

[zhaohuabing]

This PR allows users to adjust HTTP filter orders to suit their specific use cases.
It also:

• adds a constraint limiting each EnvoyExtensionPolicy to have a maximum of 16 wasm/extproc extensions.
• Fixes the JSON tag in FilterPosition API

Note: this PR doesn't handle the ordering of individual wasm/extproc extensions. This will be addressed in a follow-up PR, as it involves API change and may need more discussion.

Related: #2571 #2993

[alexwo]

/retest

[codecov]

Codecov Report

Attention: Patch coverage is 94.66667% with 4 lines in your changes are missing coverage. Please review.

Project coverage is 66.94%. Comparing base (29946b0) to head (42c3626).
Report is 134 commits behind head on main.

❗ Current head 42c3626 differs from pull request most recent head e4f6a93. Consider uploading reports for the commit e4f6a93 to get more accurate results

Files	Patch %	Lines
internal/xds/translator/httpfilters.go	93.54%	2 Missing and 2 partials ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3273      +/-   ##
==========================================
+ Coverage   66.51%   66.94%   +0.42%     
==========================================
  Files         161      163       +2     
  Lines       22673    23370     +697     
==========================================
+ Hits        15080    15644     +564     
- Misses       6720     6816      +96     
- Partials      873      910      +37     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[guydc]

Is this a topological sorting problem? Can we use/reference an established algorithm (and possibly library) to create this sort? It would make it a bit easier to be convinced in the functional correctness.

[arkodg]

this logic could be simplified

1. init - ordered list
2. loop across filterOrder (overrideList)
3. find name, and delete it from list
4. find before or after and insert it before/after in list

Its still m (orig list)*n (override list) with both sizes being fairly small, but easier to comprehend

[zhaohuabing]

Is this a topological sorting problem? Can we use/reference an established algorithm (and possibly library) to create this sort? It would make it a bit easier to be convinced in the functional correctness.

I didn't find a good way to combine two DAGs: the original one and the custom one. Since we have checked the cycle dependencies in the Gateway API reconciler, would the current implementation be sufficient?

[guydc]

Do we need to validate that we don't have cycles like:

  spec:
    filterOrder:
      - name: envoy.filters.http.jwt_authn
        before: envoy.filters.http.cors
      - name: envoy.filters.http.cors
        before: envoy.filters.http.jwt_authn

[arkodg]

Do we need to validate that we don't have cycles like:

  spec:
    filterOrder:
      - name: envoy.filters.http.jwt_authn
        before: envoy.filters.http.cors
      - name: envoy.filters.http.cors
        before: envoy.filters.http.jwt_authn

imo if we apply in order, we wont hit this issue, right ?

[zhaohuabing]

Do we need to validate that we don't have cycles like:

  spec:
    filterOrder:
      - name: envoy.filters.http.jwt_authn
        before: envoy.filters.http.cors
      - name: envoy.filters.http.cors
        before: envoy.filters.http.jwt_authn

imo if we apply in order, we wont hit this issue, right ?

Added a check for a more friendly UI.

[arkodg]

LGTM thanks
thanks for adding those extra comments which will help the next dev looking at the code !