Issue #21560 has been updated by luke-gru (Luke Gruber). Assignee set to ractor Thanks for the benchmarks, they're very helpful. We (Shopify) have been working on Ractor issues and performance a lot for 3.5.0, but we haven't had time to dig into the scheduler much yet. I'll try to take a look soon. ---------------------------------------- Bug #21560: RUBY_MN_THREADS=1 causes large performance regression in Puma 7 https://bugs.ruby-lang.org/issues/21560#change-114511 * Author: schneems (Richard Schneeman) * Status: Open * Assignee: ractor * Backport: 3.2: UNKNOWN, 3.3: UNKNOWN, 3.4: UNKNOWN ---------------------------------------- ## Expected Running a webserver with `RUBY_MN_THREADS=1` will increase performance or keep it the same. ## Actual - Before: 22919.85 Requests/sec - After: 2287.13 Requests/sec ## Reproduction Tracking this at https://github.com/puma/puma/issues/3720. I would like to get a smaller reproduction, but until then here's how we're able to induce the failure: Prepare puma: ``` $ git clone https://github.com/puma/puma $ cd puma $ git checkout v7.0.0.pre1 $ rake compile ``` Boot the server to deliver a "hello world" response: ``` $ bundle exec ruby -Ilib bin/puma -w 1 -t1 --preload test/rackup/hello.ru ``` Exercise the server with `wrk` (brew install wrk): ``` $ wrk -H 'Host: tfb-server' -H 'Accept: text/plain,text/html;q=0.9,application/xhtml+xml;q=0.9,application/xml;q=0.8,*/*;q=0.7' -H 'Connection: keep-alive' --latency -d 15 -c 16 --timeout 8 -t 12 http://localhost:9292 ``` ### Results Before: ``` $ bundle exec ruby -Ilib bin/puma -w 1 -t1 --preload test/rackup/hello.ru ``` ``` $ wrk -H 'Host: tfb-server' -H 'Accept: text/plain,text/html;q=0.9,application/xhtml+xml;q=0.9,application/xml;q=0.8,*/*;q=0.7' -H 'Connection: keep-alive' --latency -d 15 -c 16 --timeout 8 -t 12 http://localhost:9292 Running 15s test @ http://localhost:9292 12 threads and 16 connections Thread Stats Avg Stdev Max +/- Stdev Latency 786.74us 1.45ms 45.57ms 90.62% Req/Sec 1.92k 188.55 4.20k 83.22% Latency Distribution 50% 386.00us 75% 523.00us 90% 2.11ms 99% 4.63ms 344638 requests in 15.04s, 25.23MB read Requests/sec: 22919.85 Transfer/sec: 1.68MB ``` After: ``` $ env RUBY_MN_THREADS=1 bundle exec ruby -Ilib bin/puma -w 1 -t1 --preload test/rackup/hello.ru ``` ``` $ wrk -H 'Host: tfb-server' -H 'Accept: text/plain,text/html;q=0.9,application/xhtml+xml;q=0.9,application/xml;q=0.8,*/*;q=0.7' -H 'Connection: keep-alive' --latency -d 15 -c 16 --timeout 8 -t 12 http://localhost:9292 Running 15s test @ http://localhost:9292 12 threads and 16 connections Thread Stats Avg Stdev Max +/- Stdev Latency 28.89ms 40.82ms 129.37ms 78.94% Req/Sec 207.34 64.09 252.00 70.70% Latency Distribution 50% 212.00us 75% 56.77ms 90% 99.48ms 99% 127.01ms 34365 requests in 15.03s, 2.52MB read Requests/sec: 2287.13 Transfer/sec: 171.45KB ``` (Notice the dramatic drop in requests per second). -- https://bugs.ruby-lang.org/