https://bz.apache.org/bugzilla/show_bug.cgi?id=65402
--- Comment #51 from Stefan Eissing <[email protected]> --- > Yes that would be possible. Maybe we take that approach. The problem is that > we have one event-source which is transporting (potentially) a pretty big > amount of data. If we reduce the interval it would increase the load and the > traffic. We are going to send increments via this events in the future - so > this problem should also be fixed in the future. No knowledge what would work best in your setup. It could be possible to send a "no-change" chunk frequently in between. > For my understanding: Is one H2Worker is processing exactly one request and > while processing is not able to process other requests? If this is the case > we would have to set H2MaxWorkers to a pretty big value. Think of the h2 worker as a single HTTP/1.1 connection - only internally inside the server. mod_http2 bundles several of these onto the main connection to the client. If you disable h2, your browser would make separate h1 connections and, if it closes them, the standard mpm_event worker would hang for 30 seconds as well, before the server can re-use this worker for sth else. The real fix, and the one we started discussing on the dev mailing list, is that mod_proxy notices a h2 RST_STREAM immediately and abandon the request. But that is a fundamental change and may take a while. One other mitigation that comes before that is that the h2 workers become dynamic again. I will make a release with that on the github repro mod mod_h2 probably today, so people can do early testing. This would mean that the server only starts H2MinWorkers and goes up to H2MaxWorkers on load, going down again when this subsides again. -- You are receiving this mail because: You are the assignee for the bug. --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
