Patrik Ragnarsson

So we should avoid [the `sleep`](https://github.com/puma/puma/blob/89a448e51b9ba22d46a7a6108fb4787623290dc2/lib/puma/server.rb#L389-L395) for a single worker

> I've not validated that removing that sleep fixes the RUBY_MN_THREADS=1 perf issue. From https://bugs.ruby-lang.org/issues/21560 it sounds like the issue goes beyond `sleep` (all scheduling) I don't think Puma is...

This looks pretty neat/simple to me.

> This feature trades a risk of increased latency for more throughput. Is that a good tradeoff? It think that's exactly what's being asked for? https://github.com/puma/puma/issues/3777#issuecomment-3337234298, https://github.com/puma/puma/issues/3777#issuecomment-3448936586 > Should we...

If Rails would take advantage how this, how would Rails detect that it's supported by the running web server?

Can and should the number of io threads be included in [the stats](https://github.com/puma/puma/blob/731b97d2a5c7838c9d736462e259e629655b00a1/docs/stats.md)? https://github.com/puma/puma/blob/731b97d2a5c7838c9d736462e259e629655b00a1/lib/puma/thread_pool.rb#L91-L104

I don't think there will be any more 3.x releases, I'm afraid. Is there a problem with using Sinatra v4?

You can always use your own fork. The beauty of open source.

Yeah this doesn't look too good, this is a very common pattern with Sinatra. I see now that my own app is broken with Sinatra v4.2.0: https://github.com/Starkast/wikimum/actions/runs/18363209493/job/52310792793?pr=810#step:5:31 `config.ru` ```ruby #...

@pilaf I reverted https://github.com/sinatra/sinatra/pull/2114 in v4.2.1, do you mind trying it?