Noticeable performance impact of change to handle peer disconnects

[rbjorklin]

Hello!

I tried porting an existing httpaf test to httpun for the TechEmpower web framework benchmark and posted about it here. It looks like the change that was introduced here actually carries a ~15% performance overhead. Not sure if there's a better way to deal with this but thought it was worth mentioning.

[anmonteiro]

how did you measure this? that's surprising to me

[rbjorklin]

• Clone https://github.com/rbjorklin/FrameworkBenchmarks/
• Checkout add-ocaml-httpun-eio
• Comment lines to use additional_domains
• cd frameworks/OCaml/httpun
• Build: dune build --release httpun_eio.exe
• Start server: ./_build/default/httpun_eio.exe
• Start client:

export HOST=<your local ip>`

docker run --rm -ti techempower/tfb.wrk /usr/bin/wrk \
  -H 'Host: tfb-server' \
  -H 'Accept: application/json,text/html;q=0.9,application/xhtml+xml;q=0.9,application/xml;q=0.8,*/*;q=0.7' \
  -H 'Connection: keep-alive' \
  --latency \
  --duration 2 \
  --connections 1 \
  --timeout 5 \
  --threads 1 \
  http://${HOST}:8080/json

• Remove the lines referenced in the first message then install local modification opam install ./gluten-eio.opam
• Re-run test from above

[anmonteiro]

I'm just a bit confused by the change you mentioned, since that PR just moved that around? Did you manage to find the actual source of the performance regression?

[anmonteiro]

Is this perhaps the change you're referring to? e528beb

Otherwise I'm either getting mega-confused, or that Fiber.first() had been there already.

[rbjorklin]

Oh yeah you're right. Sorry about that.