million-hello-challenge

Benchmarking 1 million HTTP “Hello World” requests across nginx, Rust, Go, Kotlin, Node.js, and Python servers in GitHub Actions.

Consider this an upper bound on the performance of each language/framework, adding more code will make things slower.

Latest results:

• Formatted
• Raw

Servers in this repo:

• rust-server using axum
• go-server using builtin net/http
• kotlin-server using http4k with Helidon server using virtual threads, Java 25.
• node-server using builtin node:http server.
• python-server using tornado server, using pre-forking to use all available CPUs.
• nginx-server using nginx.

Test Setup:

• Use oha test tool to make 1 million HTTP requests
• Using HTTP 1.1 with varying number of connections.
• At oha client measure:
  • Success rate
  • Test duration
  • Requsts per Second (RPS)
  • Response time (P50, P99, P99.9 milliseconds)
• At server measure:
  • Total resident (RSS) memory usage
  • Total CPU time
  • Total threads created
  • Total processes created