Files
marqdo-cli/docs/performance.md
T
cflmy d2267e912a
CI / fmt + clippy + test (Linux x86_64) (push) Waiting to run
CI / schema instance validation + consumers (push) Waiting to run
CI / runtime contract (macos-latest) (push) Waiting to run
CI / runtime contract (windows-latest) (push) Waiting to run
Linux strict security lane (G-C) / strict isolation escape matrix (push) Waiting to run
ci/woodpecker/push/woodpecker Pipeline was canceled
perf: add correctness-verified marqdo.perf.v2 sampler
Replace the three-run baseline profile with tools/perf/runner.py: one fixed
release binary, per-child rusage via posix_spawn/wait4, 5 warmups and 20
samples by default, SHA-derived interleaved order, p50/p95/MAD, CPU user/sys
and peak RSS. Every warmup output is normalized and compared with a committed
wire digest before timing; a mismatch is CORRECTNESS_FAIL, never a fast
sample. Short or unapproved-host runs are SMOKE, and baselines from another
platform_id/profile are refused.

Add workloads.json, expected digests for nine existing workloads, a dedicated
fixed-node workflow (manual, self-hosted, requires --cpu-list), and
docs/performance.md. bench.py stays as a compatibility shim. This is step 5 of
the CI/security RFC (B08).
2026-11-21 17:28:05 +08:00

41 lines
1.8 KiB
Markdown

# Performance measurements (G4)
Status: the sampler and correctness corpus are implemented; **no release
performance baseline exists yet**. Do not cite a development-container or
TCG/QEMU number as a release gate.
## Three separate measurements
| Class | Goal | Host |
|---|---|---|
| Language core throughput | interpreter, expressions, memory/containers/stdlib algorithms | fixed G4 Linux node |
| CLI experience | cold `check`/`run`/`inspect`, parsing, stdlib loading | fixed G4 Linux node |
| Strict overhead | probe, re-exec, verification and strict run time vs pure | separate isolated performance node |
The G3 security VM can report correctness, not release performance: TCG time is
dominated by emulation and is not comparable with native or KVM.
## Tool
`tools/perf/runner.py` (schema `marqdo.perf.v2`) builds one release binary,
records its SHA256/Cargo.lock/rustc/source SHA, verifies every workload's exact
wire digest **before** timing, then reports warmups/samples, p50, p95, MAD,
CPU user/sys, peak RSS and the environment manifest. See
[`tools/perf/README.md`](../tools/perf/README.md).
Rules enforced by the tool:
* a wrong workload output is `CORRECTNESS_FAIL`; its timing is never reported;
* fewer than 5 warmups / 20 samples is `SMOKE`, not release evidence;
* `--approved-node` is required for `MEASURED`;
* a baseline from another `platform_id` or profile is refused;
* p50 +15% / RSS +20% are investigation thresholds, not automatic rollbacks.
## Current evidence
* Development-container smoke run (2026-11-21, `--warmups 1 --samples 3
--allow-smoke`): all nine workloads `VERIFIED`/`SMOKE`, no correctness
failures. Stored as a smoke artifact, **not** a release baseline.
* Remote G4: **UNVERIFIED** until a fixed self-hosted node runs
`.github/workflows/performance-linux.yml` with a fixed `--cpu-list`.