nns
|
cf760fab10
|
feat(s26): flaky-test detection + observability dashboards (8/8 ✓ 10/10 cert)
Auto-tag / Create date-tag (push) Waiting to run
CI / Backend (.NET 8) (push) Waiting to run
CI / Web (React + Vite) (push) Waiting to run
CI / POS (WPF, Windows) (push) Waiting to run
После 24 спринтов regress-suite разросся; нестабильность блокирует
доверие. Этот спринт: ловит flaky тесты, добавляет observability
(Grafana + Prometheus alerts + RUNBOOK), сертифицирует 10× cert-прогон.
1. tests/regression/find-flaky.sh — 10× прогон + JSON-агрегатор →
docs/flaky-tests.md (per-test pass/fail sequence + reproduce).
2. OrgFactory.signupWithRetry теперь honors Retry-After header
(api-client.ts:ApiError.retryAfterSec). Stage rate-limit поднят:
RATE_SIGNUP_HOUR=5000, RATE_PER_IP_MIN=5000 (~/food-market-stage/deploy/.env).
3. fullyParallel=true + workers=4 = тесты идут в недетерминированном
порядке; isolation работает (OrgFactory per-test).
4. workers=4 даёт **2.4× ускорение** (66.6s → 27.7s). Worker-scoped
фикстура lib/worker-org.ts добавлена как opt-in.
5. deploy/grafana/dashboards/quality-watchdog.json (10 панелей:
smoke success ratio 7d, incidents, multi-tenant violations,
current emoji, p95 by endpoint, step failures, RPS, DB p95,
docs posted, disk free) + dashboards/README.md.
quality-watchdog.sh пишет Prometheus textfile экспорт в
~/.fm-watchdog/textfile/quality_watchdog.prom для node_exporter.
6. deploy/prometheus/alerts.yml — 10 правил, 4 группы (uptime,
errors, database, quality-watchdog). MultiTenantViolation = P0.
deploy/prometheus/prometheus.yml — reference config.
7. docs/RUNBOOK.md +178 строк: action per alert (api-down,
rps-drop, http-errors-spike/growing, doc-posting-errors,
db-p95-high, disk-free-low, watchdog-red, multi-tenant-violation,
watchdog-incident). Junior-friendly с конкретными командами.
**Cert-прогон (10× workers=4):** 420/420 passed, 0 flaky, avg 30.1s/run,
total 300.6s (< 5min budget).
Изменения вне репо:
- ~/food-market-stage/deploy/.env — RATE_* limits bumped.
- ~/quality-watchdog.sh — добавлен .prom textfile экспорт.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
|
2026-06-08 14:44:19 +05:00 |
|