Post-Deploy Verification

After every deploy, verify the application is healthy. The deploy job itself does not run an explicit health check -- it relies on Docker's built-in health check on the web container. You should confirm the deploy manually, especially for changes that touch auth, database queries, or external service integrations.

Quick Verification (Remote)

Health Endpoint

curl -s https://trovella.ai/api/health | jq

Expected response when all services are healthy:

{
  "status": "healthy",
  "checks": {
    "database": { "ok": true },
    "redis": { "ok": true },
    "typesense": { "ok": true }
  },
  "timestamp": "2026-..."
}

The health endpoint checks three backing services:

Check	What It Tests
database	Cloud SQL connection via the Cloud SQL Auth Proxy container
redis	Upstash Redis connection
typesense	Typesense container connection

A "degraded" response (HTTP 200) means at least one service is down but the app is partially functional. An "unhealthy" response (HTTP 503) means all services are down.

Watch the Pipeline

# See the latest run on main
gh run list --branch main --limit 1

# Live tail the current run
gh run watch

VM-Level Verification

SSH into the VM to inspect container status and logs directly.

Container Status

gcloud compute ssh trovella-prod-vm \
  --zone=us-central1-a --project=trovella-prod \
  --tunnel-through-iap

# Check all containers (all should show "Up" with "(healthy)" for web)
docker compose -f /opt/trovella/docker-compose.prod.yml ps

Expected output shows five containers: caddy, web, cloud-sql-proxy, typesense, and inngest. The web container should show (healthy) in its status. Other containers show Up without a health qualifier (they don't have Docker health checks configured).

Recent Logs

# All services (last 50 lines)
docker compose -f /opt/trovella/docker-compose.prod.yml logs --tail=50

# Specific services
docker compose -f /opt/trovella/docker-compose.prod.yml logs web --tail=50
docker compose -f /opt/trovella/docker-compose.prod.yml logs caddy --tail=50
docker compose -f /opt/trovella/docker-compose.prod.yml logs inngest --tail=20
docker compose -f /opt/trovella/docker-compose.prod.yml logs typesense --tail=20

Look for:

• web: Ready on http://0.0.0.0:3000 confirms Next.js started
• caddy: No TLS errors, successful reverse proxy connections
• inngest: Functions registered confirms background job handlers connected
• cloud-sql-proxy: No connection errors to Cloud SQL

Docker Health Check Details

The web container health check is defined in docker-compose.prod.yml:

healthcheck:
  test:
    [
      "CMD",
      "node",
      "-e",
      "fetch('http://localhost:3000/api/health').then(r => { if (!r.ok) process.exit(1) }).catch(() => process.exit(1))",
    ]
  interval: 10s
  timeout: 5s
  retries: 3
  start_period: 15s

• Start period: 15 seconds grace time for Next.js to boot
• Interval: checks every 10 seconds
• Retries: 3 consecutive failures before marking unhealthy
• Timeout: each check must respond within 5 seconds

Monitoring Alerts

Cloud Monitoring provides automated alerting for four conditions:

Alert	Threshold	Response
CPU	> 85% for 5 minutes	Email notification
Memory	> 90% for 5 minutes	Email notification
Disk	> 80%	Email notification (immediate)
HTTPS Uptime	Health check fails	Email, auto-close after 30 min recovery

The uptime check hits https://trovella.ai/api/health every 5 minutes from Google's global probes. If it fails, you receive an email alert.

View and manage alerts in the Cloud Monitoring console. Alert policies are defined in Terraform at infra/modules/compute-vm/monitoring.tf.

Error Tracking

Sentry

Check Sentry for new errors after deploy. Sentry captures:

• Unhandled exceptions
• Error boundary catches
• Request errors (via onRequestError instrumentation)

Filter by release to see errors specific to the latest deploy.

Cloud Logging

Structured application logs flow to Cloud Logging via Pino. Useful filters after a deploy:

# All errors since the deploy
severity >= ERROR
timestamp >= "2026-04-08T12:00:00Z"

# Slow operations
jsonPayload.durationMs > 5000

Admin Dashboards (via SSH Tunnel)

Two internal dashboards are accessible via SSH tunnel for deeper post-deploy inspection:

Inngest Dashboard

gcloud compute ssh trovella-prod-vm \
  --zone=us-central1-a --project=trovella-prod \
  --tunnel-through-iap -- -L 8288:localhost:8288

Open http://localhost:8288 to inspect background job execution, failed functions, and queue depth.

Drizzle Studio

gcloud compute ssh trovella-prod-vm \
  --zone=us-central1-a --project=trovella-prod \
  --tunnel-through-iap -- -L 4983:localhost:4983

Open https://local.drizzle.studio to browse the production database schema and data.