filters/on_demand: fix first request dropped when body is present

[wdauchy]

Commit Message:

Track downstream_end_stream_ to support stream recreation with fully read request bodies. Previously, the filter rejected all requests with bodies by checking !callbacks_->decodingBuffer(), even when the body was complete.

Now properly distinguishes incomplete vs complete requests by tracking end_stream state through decoder methods, following the same pattern as the router filter's internal redirect support.

Additional Description:
Risk Level:
Testing: updated tests
Docs Changes:
Release Notes:
Platform Specific Features:
[Optional Runtime guard:]
[Optional Fixes #Issue]

Fixes #17891

[Optional Fixes commit #PR or SHA]
[Optional Deprecated:]
[Optional API Considerations:]

[repokitteh-read-only]

As a reminder, PRs marked as draft will not be automatically assigned reviewers,
or be handled by maintainer-oncall triage.

Please mark your PR as ready when you want it to be reviewed!

🐱

Caused by: #42158 was opened by wdauchy.

see: more, trace.

[botengyao]

Thanks, the fix lgtm, and could you also add a change log here?

/wait

[repokitteh-read-only]

CC @envoyproxy/runtime-guard-changes: FYI only for changes made to (source/common/runtime/runtime_features.cc).

🐱

Caused by: #42158 was synchronize by wdauchy.

see: more, trace.

[botengyao]

One small improvement otherwise lgtm

/wait

[botengyao]

thanks!

[adisuissa]

Does this addresses the comment mentioned #17891 (comment) for both H/1 and H/2?

[wdauchy]

Does this addresses the comment mentioned #17891 (comment) for both H/1 and H/2?

I believe the protocol level end_stream flag remains accurate regardless of buffering. The check !callbacks_->decodingBuffer() would fail for HTTP/2, since HTTP/2 may still hold a buffer even when the entire body has already been read.

I will double check this in practice.

[adisuissa]

I believe the protocol level end_stream flag remains accurate regardless of buffering. The check !callbacks_->decodingBuffer() would fail for HTTP/2, since HTTP/2 may still hold a buffer even when the entire body has already been read.

I will double check this in practice.

Will it make sense to add integration tests?

[wdauchy]

I believe the protocol level end_stream flag remains accurate regardless of buffering. The check !callbacks_->decodingBuffer() would fail for HTTP/2, since HTTP/2 may still hold a buffer even when the entire body has already been read.
I will double check this in practice.

Will it make sense to add integration tests?

good point, I will see if I can test this scenario

[wdauchy]

I believe the protocol level end_stream flag remains accurate regardless of buffering. The check !callbacks_->decodingBuffer() would fail for HTTP/2, since HTTP/2 may still hold a buffer even when the entire body has already been read.
I will double check this in practice.

Will it make sense to add integration tests?

I tried something in #42248
let me know what you think