Add production health checks

backend/src/app.ts

@@ -12,8 +12,9 @@ import nodemailer, { type SendMailOptions } from 'nodemailer';
 import Stripe from 'stripe';
 import { z } from 'zod';
 
+import { db } from './db/client.js';
 import { ensureSchema } from './db/schema.js';
-import { adoptionReportQueueEvents, enqueueAdoptionReportJob } from './queues/adoptionReportQueue.js';
+import { adoptionReportQueueEvents, enqueueAdoptionReportJob, getAdoptionReportQueueCounts } from './queues/adoptionReportQueue.js';
 import { enqueueBirdMilestoneReminderJob, getBirdMilestoneReminderQueueCounts } from './queues/birdMilestoneReminderQueue.js';
 import { enqueueMedicationReminderJob, getMedicationReminderQueueCounts } from './queues/medicationReminderQueue.js';
 import {
@@ -2265,6 +2266,59 @@ const ensureBirdWritable = (bird: BirdRow, res: Response) => {
   return false;
 };
 
+type HealthCheckResult = {
+  ok: boolean;
+  latencyMs?: number;
+  error?: string;
+};
+
+const withHealthTimeout = async <T,>(operation: Promise<T>, timeoutMs = 2_000): Promise<T> => {
+  let timeout: NodeJS.Timeout | undefined;
+
+  try {
+    return await Promise.race([
+      operation,
+      new Promise<never>((_resolve, reject) => {
+        timeout = setTimeout(() => reject(new Error('Health check timed out')), timeoutMs);
+      }),
+    ]);
+  } finally {
+    if (timeout) {
+      clearTimeout(timeout);
+    }
+  }
+};
+
+const checkPostgresHealth = async (): Promise<HealthCheckResult> => {
+  const startedAt = Date.now();
+
+  try {
+    await withHealthTimeout(db.query('SELECT 1'));
+    return { ok: true, latencyMs: Date.now() - startedAt };
+  } catch (error) {
+    return {
+      ok: false,
+      latencyMs: Date.now() - startedAt,
+      error: error instanceof Error ? error.message : 'Postgres health check failed',
+    };
+  }
+};
+
+const checkRedisHealth = async (): Promise<HealthCheckResult> => {
+  const startedAt = Date.now();
+
+  try {
+    await withHealthTimeout(getBirdMilestoneReminderQueueCounts());
+    return { ok: true, latencyMs: Date.now() - startedAt };
+  } catch (error) {
+    return {
+      ok: false,
+      latencyMs: Date.now() - startedAt,
+      error: error instanceof Error ? error.message : 'Redis health check failed',
+    };
+  }
+};
+
 const writeAuditLog = async (
   auth: AuthContext,
   action: string,
@@ -2293,8 +2347,46 @@ const isBillingOnlyWorkspaceUpdate = (
   payload: z.infer<typeof workspaceSchema>,
 ) => workspace.workspace_type === 'standard' && payload.workspaceType === 'standard' && payload.name === workspace.name;
 
-app.get('/api/health', (_req: Request, res: Response) => {
-  res.json({ ok: true });
+app.get('/api/health/live', (_req: Request, res: Response) => {
+  res.json({
+    ok: true,
+    service: 'flockpal-backend',
+    status: 'live',
+    uptimeSeconds: Math.round(process.uptime()),
+    checkedAt: new Date().toISOString(),
+  });
+});
+
+app.get('/api/health/ready', async (_req: Request, res: Response) => {
+  const [postgres, redis] = await Promise.all([checkPostgresHealth(), checkRedisHealth()]);
+  const ok = postgres.ok && redis.ok;
+
+  res.status(ok ? 200 : 503).json({
+    ok,
+    service: 'flockpal-backend',
+    status: ok ? 'ready' : 'degraded',
+    checkedAt: new Date().toISOString(),
+    dependencies: {
+      postgres,
+      redis,
+    },
+  });
+});
+
+app.get('/api/health', async (_req: Request, res: Response) => {
+  const [postgres, redis] = await Promise.all([checkPostgresHealth(), checkRedisHealth()]);
+  const ok = postgres.ok && redis.ok;
+
+  res.status(ok ? 200 : 503).json({
+    ok,
+    service: 'flockpal-backend',
+    status: ok ? 'ready' : 'degraded',
+    checkedAt: new Date().toISOString(),
+    dependencies: {
+      postgres,
+      redis,
+    },
+  });
 });
 
 app.get('/api/metrics', requireAuth, requireAdmin, async (_req: Request, res: Response, next: NextFunction) => {
@@ -2326,6 +2418,7 @@ app.get('/api/metrics', requireAuth, requireAdmin, async (_req: Request, res: Re
       queues: {
         birdMilestoneReminders: birdMilestoneReminderQueueCounts,
         medicationReminders: medicationReminderQueueCounts,
+        adoptionReports: await getAdoptionReportQueueCounts(),
       },
     });
   } catch (error) {

backend/src/healthcheck.ts

@@ -0,0 +1,54 @@
+import { db } from './db/client.js';
+import { closeBirdMilestoneReminderQueue, getBirdMilestoneReminderQueueCounts } from './queues/birdMilestoneReminderQueue.js';
+
+const timeoutMs = Number(process.env.HEALTHCHECK_TIMEOUT_MS ?? 5_000);
+
+const withTimeout = async <T>(operation: Promise<T>, label: string): Promise<T> => {
+  let timeout: NodeJS.Timeout | undefined;
+
+  try {
+    return await Promise.race([
+      operation,
+      new Promise<never>((_resolve, reject) => {
+        timeout = setTimeout(() => reject(new Error(`${label} timed out`)), timeoutMs);
+      }),
+    ]);
+  } finally {
+    if (timeout) {
+      clearTimeout(timeout);
+    }
+  }
+};
+
+const checkHttp = async (path: string) => {
+  const port = process.env.PORT ?? '5000';
+  const response = await withTimeout(fetch(`http://127.0.0.1:${port}${path}`), path);
+
+  if (!response.ok) {
+    throw new Error(`${path} returned ${response.status}`);
+  }
+};
+
+const checkWorkerDependencies = async () => {
+  await withTimeout(db.query('SELECT 1'), 'postgres');
+  await withTimeout(getBirdMilestoneReminderQueueCounts(), 'redis');
+};
+
+const mode = process.argv[2] ?? 'api-ready';
+
+try {
+  if (mode === 'api-live') {
+    await checkHttp('/api/health/live');
+  } else if (mode === 'api-ready') {
+    await checkHttp('/api/health/ready');
+  } else if (mode === 'worker') {
+    await checkWorkerDependencies();
+  } else {
+    throw new Error(`Unknown healthcheck mode: ${mode}`);
+  }
+} catch (error) {
+  console.error(error instanceof Error ? error.message : error);
+  process.exitCode = 1;
+} finally {
+  await Promise.allSettled([closeBirdMilestoneReminderQueue(), db.close()]);
+}

backend/src/queues/adoptionReportQueue.ts

@@ -40,3 +40,5 @@ export const closeAdoptionReportQueue = async () => {
   await adoptionReportQueue.close();
   await adoptionReportQueueEvents.close();
 };
+
+export const getAdoptionReportQueueCounts = () => adoptionReportQueue.getJobCounts('waiting', 'active', 'delayed', 'completed', 'failed');