cannaiq/backend/src/routes/health.ts

/**
 * Health Check Routes
 *
 * Comprehensive health endpoints for monitoring API, DB, Redis, Workers, Crawls, and Analytics.
 *
 * Endpoints:
 *   GET /api/health           - Quick API health check
 *   GET /api/health/db        - Postgres health
 *   GET /api/health/redis     - Redis health
 *   GET /api/health/workers   - Queue and worker status
 *   GET /api/health/crawls    - Crawl activity summary
 *   GET /api/health/analytics - Analytics/aggregates status
 *   GET /api/health/full      - Aggregated view of all subsystems
 */

import { Router, Request, Response } from 'express';
import { pool } from '../db/pool';
import { getRedis } from '../lib/redis';
import * as fs from 'fs';
import * as path from 'path';

const router = Router();

// Read package version
let packageVersion = '1.0.0';
try {
  const packagePath = path.join(__dirname, '../../package.json');
  if (fs.existsSync(packagePath)) {
    const pkg = JSON.parse(fs.readFileSync(packagePath, 'utf8'));
    packageVersion = pkg.version || '1.0.0';
  }
} catch {
  // Ignore errors reading package.json
}

// Store server start time for uptime calculation
const serverStartTime = Date.now();

// Types
interface HealthStatus {
  status: 'ok' | 'degraded' | 'error' | 'stale';
}

interface ApiHealth extends HealthStatus {
  uptime: number;
  timestamp: string;
  version: string;
  build_sha: string | null;
  build_time: string | null;
}

interface DbHealth extends HealthStatus {
  connected: boolean;
  latency_ms: number;
  error?: string;
}

interface RedisHealth extends HealthStatus {
  connected: boolean;
  latency_ms: number;
  error?: string;
}

interface QueueInfo {
  name: string;
  waiting: number;
  active: number;
  completed: number;
  failed: number;
  paused: boolean;
}

interface WorkerInfo {
  id: string;
  queue: string;
  status: string;
  last_heartbeat?: string;
}

interface WorkersHealth extends HealthStatus {
  queues: QueueInfo[];
  workers: WorkerInfo[];
}

interface CrawlsHealth extends HealthStatus {
  last_run: string | null;
  runs_last_24h: number;
  stores_with_recent_crawl: number;
  stores_total: number;
  stale_stores: number;
}

interface AnalyticsHealth extends HealthStatus {
  last_aggregate: string | null;
  daily_runs_last_7d: number;
  missing_days: number;
}

interface FullHealth extends HealthStatus {
  api: ApiHealth;
  db: DbHealth;
  redis: RedisHealth;
  workers: WorkersHealth;
  crawls: CrawlsHealth;
  analytics: AnalyticsHealth;
}

// ============================================================
// Helper Functions
// ============================================================

async function getApiHealth(): Promise<ApiHealth> {
  return {
    status: 'ok',
    uptime: Math.floor((Date.now() - serverStartTime) / 1000),
    timestamp: new Date().toISOString(),
    version: packageVersion,
    build_sha: process.env.APP_GIT_SHA && process.env.APP_GIT_SHA !== 'unknown' ? process.env.APP_GIT_SHA : null,
    build_time: process.env.APP_BUILD_TIME && process.env.APP_BUILD_TIME !== 'unknown' ? process.env.APP_BUILD_TIME : null,
  };
}

async function getDbHealth(): Promise<DbHealth> {
  const start = Date.now();
  try {
    // pool imported from db/pool
    await pool.query('SELECT 1');
    return {
      status: 'ok',
      connected: true,
      latency_ms: Date.now() - start,
    };
  } catch (err: any) {
    return {
      status: 'error',
      connected: false,
      latency_ms: Date.now() - start,
      error: err.message || 'Database connection failed',
    };
  }
}

async function getRedisHealth(): Promise<RedisHealth> {
  const start = Date.now();
  const isLocal = process.env.NODE_ENV === 'development' || process.env.NODE_ENV === 'local' || !process.env.NODE_ENV;

  // Check if Redis is configured
  if (!process.env.REDIS_URL && !process.env.REDIS_HOST) {
    // Redis is optional in local dev, required in prod/staging
    return {
      status: isLocal ? 'ok' : 'error',
      connected: false,
      latency_ms: 0,
      error: isLocal ? 'Redis not configured (optional in local)' : 'Redis not configured (required in production)',
    };
  }

  try {
    const redis = getRedis();
    // Use a timeout to prevent hanging
    const pingPromise = redis.ping();
    const timeoutPromise = new Promise<never>((_, reject) =>
      setTimeout(() => reject(new Error('Redis ping timeout')), 3000)
    );

    await Promise.race([pingPromise, timeoutPromise]);
    return {
      status: 'ok',
      connected: true,
      latency_ms: Date.now() - start,
    };
  } catch (err: any) {
    return {
      status: 'degraded',
      connected: false,
      latency_ms: Date.now() - start,
      error: err.message || 'Redis ping failed',
    };
  }
}

async function getWorkersHealth(): Promise<WorkersHealth> {
  try {
    // pool imported from db/pool

    // Get queue stats from v_queue_stats view or equivalent
    const queueStatsResult = await pool.query(`
      SELECT
        job_type as name,
        COUNT(*) FILTER (WHERE status = 'pending') as waiting,
        COUNT(*) FILTER (WHERE status = 'running') as active,
        COUNT(*) FILTER (WHERE status = 'success') as completed,
        COUNT(*) FILTER (WHERE status IN ('error', 'failed')) as failed,
        false as paused
      FROM dispensary_crawl_jobs
      WHERE created_at > NOW() - INTERVAL '7 days'
      GROUP BY job_type
    `);

    const queues: QueueInfo[] = queueStatsResult.rows.map((row: any) => ({
      name: row.name || 'unknown',
      waiting: parseInt(row.waiting) || 0,
      active: parseInt(row.active) || 0,
      completed: parseInt(row.completed) || 0,
      failed: parseInt(row.failed) || 0,
      paused: row.paused || false,
    }));

    // Get active workers from job_schedules or active heartbeats
    const workersResult = await pool.query(`
      SELECT
        COALESCE(job_config->>'worker_name', job_name) as id,
        job_name as queue,
        CASE WHEN enabled THEN 'connected' ELSE 'disconnected' END as status,
        last_run_at as last_heartbeat
      FROM job_schedules
      WHERE enabled = true
      ORDER BY last_run_at DESC NULLS LAST
      LIMIT 20
    `);

    const workers: WorkerInfo[] = workersResult.rows.map((row: any) => ({
      id: row.id,
      queue: row.queue,
      status: row.status,
      last_heartbeat: row.last_heartbeat?.toISOString() || undefined,
    }));

    // Determine overall status
    const hasActiveWorkers = workers.length > 0;
    const hasFailedJobs = queues.some((q) => q.failed > 0);
    const hasStuckJobs = queues.some((q) => q.active > 5); // Arbitrary threshold

    let status: 'ok' | 'degraded' | 'error' = 'ok';
    if (!hasActiveWorkers) {
      status = 'degraded';
    } else if (hasFailedJobs || hasStuckJobs) {
      status = 'degraded';
    }

    return {
      status,
      queues,
      workers,
    };
  } catch (err: any) {
    return {
      status: 'error',
      queues: [],
      workers: [],
    };
  }
}

async function getCrawlsHealth(): Promise<CrawlsHealth> {
  try {
    // pool imported from db/pool

    // Get crawl statistics
    const statsResult = await pool.query(`
      SELECT
        (SELECT MAX(completed_at) FROM dispensary_crawl_jobs WHERE status = 'success') as last_run,
        (SELECT COUNT(*) FROM dispensary_crawl_jobs WHERE status = 'success' AND completed_at > NOW() - INTERVAL '24 hours') as runs_24h,
        (SELECT COUNT(*) FROM dispensaries WHERE last_crawl_at > NOW() - INTERVAL '24 hours') as stores_recent,
        (SELECT COUNT(*) FROM dispensaries WHERE menu_type IS NOT NULL AND platform_dispensary_id IS NOT NULL) as stores_total,
        (SELECT COUNT(*) FROM dispensaries WHERE menu_type = 'dutchie' AND platform_dispensary_id IS NOT NULL AND (last_crawl_at IS NULL OR last_crawl_at < NOW() - INTERVAL '24 hours')) as stores_stale
    `);

    const stats = statsResult.rows[0] || {};
    const storesTotal = parseInt(stats.stores_total) || 0;
    const storesRecent = parseInt(stats.stores_recent) || 0;
    const staleStores = parseInt(stats.stores_stale) || 0;

    // Calculate freshness percentage
    const freshPercent = storesTotal > 0 ? (storesRecent / storesTotal) * 100 : 0;

    let status: 'ok' | 'degraded' | 'stale' | 'error' = 'ok';
    if (freshPercent >= 90) {
      status = 'ok';
    } else if (freshPercent >= 50) {
      status = 'degraded';
    } else {
      status = 'stale';
    }

    return {
      status,
      last_run: stats.last_run?.toISOString() || null,
      runs_last_24h: parseInt(stats.runs_24h) || 0,
      stores_with_recent_crawl: storesRecent,
      stores_total: storesTotal,
      stale_stores: staleStores,
    };
  } catch (err: any) {
    return {
      status: 'error',
      last_run: null,
      runs_last_24h: 0,
      stores_with_recent_crawl: 0,
      stores_total: 0,
      stale_stores: 0,
    };
  }
}

async function getAnalyticsHealth(): Promise<AnalyticsHealth> {
  try {
    // pool imported from db/pool

    // Check analytics/aggregate job runs
    const statsResult = await pool.query(`
      SELECT
        (SELECT MAX(completed_at) FROM job_run_logs WHERE job_name LIKE '%analytics%' AND status = 'success') as last_aggregate,
        (SELECT COUNT(DISTINCT DATE(started_at)) FROM job_run_logs WHERE job_name LIKE '%analytics%' AND status = 'success' AND started_at > NOW() - INTERVAL '7 days') as runs_7d
    `);

    const stats = statsResult.rows[0] || {};
    const runsLast7d = parseInt(stats.runs_7d) || 0;
    const missingDays = Math.max(0, 7 - runsLast7d);

    let status: 'ok' | 'degraded' | 'stale' | 'error' = 'ok';
    if (missingDays === 0) {
      status = 'ok';
    } else if (missingDays <= 2) {
      status = 'degraded';
    } else {
      status = 'stale';
    }

    return {
      status,
      last_aggregate: stats.last_aggregate?.toISOString() || null,
      daily_runs_last_7d: runsLast7d,
      missing_days: missingDays,
    };
  } catch (err: any) {
    return {
      status: 'error',
      last_aggregate: null,
      daily_runs_last_7d: 0,
      missing_days: 7,
    };
  }
}

function determineOverallStatus(
  api: ApiHealth,
  db: DbHealth,
  redis: RedisHealth,
  workers: WorkersHealth,
  crawls: CrawlsHealth,
  analytics: AnalyticsHealth
): 'ok' | 'degraded' | 'error' {
  const statuses = [api.status, db.status, redis.status, workers.status, crawls.status, analytics.status];

  if (statuses.includes('error')) {
    return 'error';
  }
  if (statuses.includes('degraded') || statuses.includes('stale')) {
    return 'degraded';
  }
  return 'ok';
}

// ============================================================
// Routes
// ============================================================

/**
 * GET /api/health - Quick API health check (no auth required)
 */
router.get('/', async (_req: Request, res: Response) => {
  const health = await getApiHealth();
  res.json(health);
});

/**
 * GET /api/health/db - Postgres health
 */
router.get('/db', async (_req: Request, res: Response) => {
  const health = await getDbHealth();
  const statusCode = health.status === 'ok' ? 200 : 503;
  res.status(statusCode).json(health);
});

/**
 * GET /api/health/redis - Redis health
 */
router.get('/redis', async (_req: Request, res: Response) => {
  const health = await getRedisHealth();
  const statusCode = health.status === 'ok' ? 200 : health.status === 'degraded' ? 200 : 503;
  res.status(statusCode).json(health);
});

/**
 * GET /api/health/workers - Queue and worker status
 */
router.get('/workers', async (_req: Request, res: Response) => {
  const health = await getWorkersHealth();
  const statusCode = health.status === 'ok' ? 200 : health.status === 'degraded' ? 200 : 503;
  res.status(statusCode).json(health);
});

/**
 * GET /api/health/crawls - Crawl activity summary
 */
router.get('/crawls', async (_req: Request, res: Response) => {
  const health = await getCrawlsHealth();
  const statusCode = health.status === 'ok' ? 200 : health.status === 'degraded' ? 200 : 503;
  res.status(statusCode).json(health);
});

/**
 * GET /api/health/analytics - Analytics/aggregates status
 */
router.get('/analytics', async (_req: Request, res: Response) => {
  const health = await getAnalyticsHealth();
  const statusCode = health.status === 'ok' ? 200 : health.status === 'degraded' ? 200 : 503;
  res.status(statusCode).json(health);
});

/**
 * GET /api/health/full - Aggregated view of all subsystems
 */
router.get('/full', async (_req: Request, res: Response) => {
  const [api, db, redis, workers, crawls, analytics] = await Promise.all([
    getApiHealth(),
    getDbHealth(),
    getRedisHealth(),
    getWorkersHealth(),
    getCrawlsHealth(),
    getAnalyticsHealth(),
  ]);

  const overallStatus = determineOverallStatus(api, db, redis, workers, crawls, analytics);

  const fullHealth: FullHealth = {
    status: overallStatus,
    api,
    db,
    redis,
    workers,
    crawls,
    analytics,
  };

  const statusCode = overallStatus === 'ok' ? 200 : overallStatus === 'degraded' ? 200 : 503;
  res.status(statusCode).json(fullHealth);
});

export default router;

// Export helper functions for reuse in other modules
export {
  getApiHealth,
  getDbHealth,
  getRedisHealth,
  getWorkersHealth,
  getCrawlsHealth,
  getAnalyticsHealth,
};