benchmarks/

Benchmarks

HyperServe has two comparison tools. Neither is a production capacity claim.

In-process benchmarks

The Go benchmarks in the root benchmark_test.go cover serial and concurrent request paths, middleware, static files, JSON, and MCP operations. The standalone limiter has its own middleware and entry-footprint benchmarks:

go test -run '^$' -bench . -benchmem .
go test -run '^$' -bench . -benchmem ./ratelimit

Use -count=5 when comparing revisions so one noisy sample does not drive a conclusion.

Loopback load profiles

The repository-native load command builds a maintained fixture and a small standard-library load tool, runs two profiles, records the environment, and then stops the fixture:

make benchmark-load

It requires go, git, and curl; no third-party load generator is needed. The defaults are deliberately short and can be changed without editing files:

For example:

BENCH_DURATION=30s BENCH_THREADS=8 BENCH_CONNECTIONS=100 make benchmark-load

The minimal profile uses a small handler with HyperServe's defaults. The middleware profile adds security headers and bearer authentication. Results go beneath benchmarks/results/, which Git ignores. Each run includes the exact commit, clean/dirty state, Go version, platform, workload, and profile details.

Compare runs only when their metadata and inputs match. See the performance guide for the interpretation boundary.