tps-bench

A single-binary HTTP load tester for model-serving endpoints (or any HTTP service). No dependencies beyond the Go standard library — drop it in CI to catch latency regressions before they ship.

Build & run

go build -o tps-bench .
./tps-bench -url http://localhost:8080/predict -n 5000 -c 50 -method POST -body '{"x":[1,2,3]}'

Output

requests:    5000
success:     5000 (100.0%)
elapsed:     2.143s
throughput:  2333.1 req/s
latency p50: 18.40 ms
latency p90: 31.27 ms
latency p99: 54.90 ms

Flags: -url -n -c -method -body -timeout. MIT licensed.