BENCHMARKS.md and the benchmarks.yml workflow document a k6 load-test harness and a regression check that are not present in the tree:
scripts/load-test.sh (referenced for smoke/average/stress/spike/soak/breakpoint scenarios) — does not exist. benchmarks.yml calls scripts/load-test.sh "$SCENARIO" || true, so the load test never runs and the || true masks the missing script.scripts/bench-regression.sh (baseline/compare, ">15% regression" gate) — does not exist.
As a result, the throughput/latency figures (Docker pull p95, npm p95, req/s) and the under-load memory figures (RAM at 100/1000 concurrent) in BENCHMARKS.md were never measured.
Scope
Implement the real harness so those figures become measured:
scripts/load-test.sh — k6 scenarios against a running NORA (the request distribution already sketched in BENCHMARKS.md § Throughput).scripts/bench-regression.sh — baseline/compare with a configurable threshold.- Wire both into
benchmarks.yml and publish the JSON reports as release artifacts.
Until then, the docs and CI have been scoped to only the metrics CI actually measures today (binary size, startup time, idle memory, Criterion micro-benchmarks).
BENCHMARKS.mdand thebenchmarks.ymlworkflow document a k6 load-test harness and a regression check that are not present in the tree:scripts/load-test.sh(referenced forsmoke/average/stress/spike/soak/breakpointscenarios) — does not exist.benchmarks.ymlcallsscripts/load-test.sh "$SCENARIO" || true, so the load test never runs and the|| truemasks the missing script.scripts/bench-regression.sh(baseline/compare, ">15% regression" gate) — does not exist.As a result, the throughput/latency figures (Docker pull p95, npm p95, req/s) and the under-load memory figures (RAM at 100/1000 concurrent) in
BENCHMARKS.mdwere never measured.Scope
Implement the real harness so those figures become measured:
scripts/load-test.sh— k6 scenarios against a running NORA (the request distribution already sketched inBENCHMARKS.md§ Throughput).scripts/bench-regression.sh— baseline/compare with a configurable threshold.benchmarks.ymland publish the JSON reports as release artifacts.Until then, the docs and CI have been scoped to only the metrics CI actually measures today (binary size, startup time, idle memory, Criterion micro-benchmarks).