Stream reset issues with h2c after updating to Jetty 12.1.6

[ghjm]

Jetty version(s)
12.1.6

Jetty Environment
I'm using Jetty via Dropwizard 5.0.1 with type: simple and connector type: h2c. Dropwizard enables gzip by default.

HTTP version
http/2 (h2c)

Java version/vendor (use: java -version)
OpenJDK Runtime Environment Zulu21.48+15-CA (build 21.0.10+7-LTS)
OpenJDK 64-Bit Server VM Zulu21.48+15-CA (build 21.0.10+7-LTS, mixed mode, sharing)

OS type/version
Alpine Linux v3.20

Description
I have an app deployed to Google Cloud Run, with the "Use HTTP/2 end-to-end" setting enabled, so all traffic to the Jetty service is http/2. After Jetty was updated to version 12.1.6, the app became inaccessible with the following error message appearing to all users:

upstream connect error or disconnect/reset before headers. reset reason: remote refused stream reset

Reverting to Jetty 12.1.5 solved the problem. I used git bisect and found that the following commit seems to be the one causing my issue: #14260 (Improve GZIP Performance). Perhaps the handler callback changes are somehow causing premature termination of the stream, or something like that.

[sbordet]

It is unlikely that the fix for #14260 causes HTTP/2 to break.
Unfortunately, you posted too little information to troubleshoot this issue.

Do you have a stack trace?
How is your GzipHandler configured?
Does this happen for all requests, or only some?
Do your requests have content, and is it compressed?
Can you always reproduce the issue?

Do you have a reproducer that we can try ourselves?

[ghjm]

Thanks for the quick reply. From what I know now:

• No stack traces are showing up in the logs as far as I can see.
• There's no GzipHandler in my app's code. I think this is default configuration in Dropwizard.
• The problem happens for all requests.
• The problem happens on just a straight GET / for the main page. The html/css/js responses all have Content-Encoding: gzip.
• Yes, the issue is 100% reproducible in my environment.

I will work on getting a minimal reproducer that you can look at. FWIW, the same problem has also come up in another one of our apps.

[sbordet]

@ghjm you have news?