Stefan Eissing

FYI: I increased the rate limit counter by a factor if 10 and re-enabled the falling tests in our CI. An interim fix, but CI seems happy. https://github.com/curl/curl/pull/19274

[6a5e1f05d4b1e353b26411b4f87cae6c38ab740d.sqlog.txt](https://github.com/user-attachments/files/24052685/6a5e1f05d4b1e353b26411b4f87cae6c38ab740d.sqlog.txt) Attached a qlog where stream 72 gives `FINISHED` event and the qlog lists: ``` [{"frame_type":"reset_stream","stream_id":72,"error_code":258,"final_size":13992},{"frame_type":"padding","payload_length":1399}] ```

Well, you'd need the whole curl test environment for that. A httpd, an nghttpx with h3, pytest, etc. If you want I can point you to the details, but it...

Confirming the fix makes the issue no longer reproduce on my system. Thanks.

There are no "hidden" tuning options for mod_proxy_http2. As you probably noticed, there is only ever one request on the backend connection at a time. That is due to the...

I'll check this out tomorrow.

This works excellent. I added some check defines where we can adapt the MPM magic once we have the value for the backport: ```diff diff --git a/modules/http2/h2.h b/modules/http2/h2.h index f496a6dcb2..ae0e845704...

> Hm, something (related?) broke in the H2 tests (https://github.com/apache/httpd/actions/runs/9283938584/job/25545210350).. Hmm, is this a return to the MPM during a graceful shutdown? Does it then close a connection in PROCESS...

In case this is too complicated to "simply" fix in the mpm, I can make mod_http2 refrain from returning in case the server is stopping. There'd still be an edge...

@ylavic if mpm_event cannot handle it, how about this: ```diff diff --git a/modules/http2/h2_session.c b/modules/http2/h2_session.c index e5809d59ac..7618b64f77 100644 --- a/modules/http2/h2_session.c +++ b/modules/http2/h2_session.c @@ -1536,6 +1536,7 @@ static void h2_session_ev_ngh2_done(h2_session *session, int...