feat(lineage): Phase 4 — SLO metrics + bundle v1.3.0 pilot (warn mode)
dict-relationships-v2 epic, Phase 4 (final). Backend SLO instrumentation +
schema bundle pilot. Alert rules + runbook идут отдельным коммитом в
ordinis-infra.
Backend metrics (Micrometer / Prometheus):
- nsi_dependents_query_duration_seconds (Timer + p50/p95/p99 percentiles)
labels: mode={schema|record|*_error}, target_dict
- nsi_cascade_close_total (Counter)
labels: target_dict, outcome={success|blocked|cascade_required|too_large|timeout|error}
- nsi_cascade_close_duration_seconds (Timer + percentiles)
- nsi_lineage_mv_last_refresh_seconds_ago (Gauge, NaN when mv disabled)
- nsi_lineage_mv_row_count (Gauge)
- nsi_lineage_mv_last_duration_seconds (Gauge)
- nsi_lineage_mv_refresh_total{outcome=ok|error} (Counter)
Cardinality bound: target_dict tag — ≤40 dictionaries; outcome — 6 values.
Total time series ≈ 240 per app instance — приемлемо.
Optional MeterRegistry inject — null в test ctor'ах (back-compat).
Bundle v1.3.0 pilot:
- spacecraft.satellite_type_code → satellite_type.code теперь имеет
x-references-on-close: warn. Закрытие satellite_type больше не
блокируется через 1.3.0; spacecraft остаётся orphan; orphan scanner
поднимет metric. Data steward потом repoints.
- bundle version 1.2.1 → 1.3.0, spacecraft schema 1.0.0 → 1.1.0
(description обновлён с pilot context).
- WARN-mode выбран как safe pilot — validates новый code path без
destructive cascade. CASCADE-mode flip может быть отдельным PR
per-relationship после operator review.
Verify:
- mvn verify -pl '!ordinis-app': SUCCESS (5.7s).
- 106/106 tests still green (existing tests не задеты).
- helm lint в ordinis-infra: clean (alerts render correctly).
Backward-compat: REGRESSION test (ReferenceCloseRegressionTest) covers
v1.2.1-style schemas без x-references-on-close. WARN mode на one field
не меняет contract других — bundle import idempotent.
Roadmap: 4 phases ⚓ shipped. Cascade engine + lineage UI + materialized
view + SLO metrics live.
Zimin A.N.
a7cdd5df5c