Scaffold dd0c/cost: Welford baseline, anomaly scorer, governance engine, tests
- Welford online algorithm for running mean/stddev baselines - Anomaly scorer: z-score → 0-100 mapping, property-based tests (10K runs, fast-check) - Governance engine: 14-day auto-promotion with FP rate gate, injectable Clock - Panic mode: defaults to active (safe) when Redis unreachable - Tests: 12 scorer cases (incl 2x 10K property-based), 9 governance cases, 3 panic mode cases - PostgreSQL schema with RLS: baselines (optimistic locking), anomalies, remediation_actions - Fly.io config, Dockerfile
This commit is contained in:
80
products/05-aws-cost-anomaly/src/governance/engine.ts
Normal file
80
products/05-aws-cost-anomaly/src/governance/engine.ts
Normal file
@@ -0,0 +1,80 @@
|
||||
import pino from 'pino';
|
||||
|
||||
const logger = pino({ name: 'governance' });
|
||||
|
||||
/**
|
||||
* Clock interface for deterministic governance tests (BMad must-have).
|
||||
* Inject FakeClock in tests, RealClock in production.
|
||||
*/
|
||||
export interface Clock {
|
||||
now(): number;
|
||||
}
|
||||
|
||||
export class RealClock implements Clock {
|
||||
now() { return Date.now(); }
|
||||
}
|
||||
|
||||
export class FakeClock implements Clock {
|
||||
private current: number;
|
||||
constructor(start = Date.now()) { this.current = start; }
|
||||
now() { return this.current; }
|
||||
advanceBy(ms: number) { this.current += ms; }
|
||||
}
|
||||
|
||||
export interface PromotionResult {
|
||||
promoted: boolean;
|
||||
newMode?: 'audit' | 'enforce';
|
||||
reason?: string;
|
||||
}
|
||||
|
||||
const PROMOTION_DAYS = 14;
|
||||
const MAX_FP_RATE = 0.10; // 10% false positive rate
|
||||
|
||||
/**
|
||||
* Governance engine — manages auto-promotion from shadow → audit → enforce.
|
||||
* Uses injectable Clock for deterministic testing.
|
||||
*/
|
||||
export class GovernanceEngine {
|
||||
private clock: Clock;
|
||||
|
||||
constructor(clock: Clock = new RealClock()) {
|
||||
this.clock = clock;
|
||||
}
|
||||
|
||||
/**
|
||||
* Evaluate whether a tenant should be promoted to the next governance mode.
|
||||
* Requires: 14+ days in current mode AND false positive rate < 10%.
|
||||
*/
|
||||
evaluatePromotion(
|
||||
tenantId: string,
|
||||
stats: { fpRate: number; daysInCurrentMode: number },
|
||||
): PromotionResult {
|
||||
if (stats.daysInCurrentMode < PROMOTION_DAYS) {
|
||||
return { promoted: false, reason: `Only ${stats.daysInCurrentMode} days — need ${PROMOTION_DAYS}` };
|
||||
}
|
||||
|
||||
if (stats.fpRate > MAX_FP_RATE) {
|
||||
return {
|
||||
promoted: false,
|
||||
reason: `false-positive rate ${(stats.fpRate * 100).toFixed(1)}% exceeds ${MAX_FP_RATE * 100}% threshold`,
|
||||
};
|
||||
}
|
||||
|
||||
return { promoted: true, newMode: 'audit' };
|
||||
}
|
||||
|
||||
/**
|
||||
* Check panic mode status.
|
||||
* Defaults to panic=active (safe) when Redis is unreachable (BMad must-have).
|
||||
*/
|
||||
async checkPanicMode(tenantId: string, redis: any): Promise<boolean> {
|
||||
try {
|
||||
const val = await redis.get(`panic:${tenantId}`);
|
||||
return val === '1';
|
||||
} catch (err) {
|
||||
logger.warn({ tenantId, error: (err as Error).message },
|
||||
'Redis unreachable — defaulting to panic=active');
|
||||
return true; // Safe default: panic IS active
|
||||
}
|
||||
}
|
||||
}
|
||||
Reference in New Issue
Block a user