Release notes: tier5_http bench harness (bench-harness)¶

Branch: cursor/httpd-plan-continue

Summary¶

TOML-driven tier-5 HTTP benchmark harness: defaults.toml, suite.toml, per-scenario bench.toml, and Python drivers that resolve scenarios from suite config only (no hardcoded scenario lists).

Verification (2026-05-23)¶

• ./scripts/check-tier5-http-harness.sh — OK
• HTTPD_BENCH_SKIP_TIMING=1 ./scripts/httpd-plan-gates.sh — OK (includes harness smoke)

Agent continuation¶

1. Done: nginx-src-audit, exploit-harness, m1-nginx-bench-parity on plan branch.
2. Benchmarks repo: refresh HTTP matrix via ../benchmarks when nginx+wrk timing rows are captured (SKIP_BENCH=1 for docs-only).

Changed¶

• benchmarks/tier5_http/ — defaults.toml, suite.toml, scenarios/*/bench.toml, templates/nginx.conf.in, README.md
• benchmarks/harness/http_bench_toml.py, bench_http.py, verify_http.py, plot_http.py (stub)
• scripts/check-tier5-http-harness.sh, scripts/httpd-plan-gates.sh
• docs/benchmarks-http.md
• Plan todo bench-harness → completed

Not changed¶

• exploit_http.py / audit_nginx_src.py (later todos)
• Live Pages bench refresh (no new timing CSV this slice)
• Full li-httpd wrk parity runs (nginx+wrk optional; CI uses stub verify)

Breaking / Security / Performance / Downstream¶