Patrik Ragnarsson
Patrik Ragnarsson
Should worker threads be eagerly reading requests before handing the connection over to the reactor?
@joshuay03 Have you seen https://github.com/puma/puma/pull/3506? What's your take on that? In think both patches solves the same problem?
What makes `exe/` more modern over `bin/`?
I mean https://github.com/rails/rails/blob/main/railties/exe/rails hasn't changed in 8 years
I'll repeat my comment https://github.com/puma/puma/issues/3715#issuecomment-3247114128 because I think it is relevant > `echo 'silence_single_worker_warning; app { [200, {}, ["OK"]] }' | RUBY_MN_THREADS=1 ruby -Ilib bin/puma --config /dev/stdin --worker 1 --threads...
I can try that in a moment, but I'll share a thing first. `hey -disable-keepalive -n 5000 -c 100 http://127.0.0.1:7777` is my `hey` command. Against 1 worker 1 thread `echo...
I'm not sure what you are after with that change @MSP-Greg , I did it, started Puma with `--worker 1 --threads 1:1` and ran the `hey` command, it took ~54s.
I made this change to Puma ```diff $ git diff diff --git a/lib/puma/server.rb b/lib/puma/server.rb index a6c6260a..e999714e 100644 --- a/lib/puma/server.rb +++ b/lib/puma/server.rb @@ -388,7 +388,10 @@ module Puma if pool.busy_threads >=...
> You mentioned 'if I remove this sleep'. The change should have the same effect? Seems like it doesn't have https://github.com/puma/puma/issues/3720#issuecomment-3250879752
I didn't remove the else, sorry :)
Anyway, it reproduces with multiple workers too https://github.com/puma/puma/issues/3720#issuecomment-3250871753