deploy_monitoring

Add a /health endpoint returning status and version for liveness, and a /ready endpoint that verifies dependencies like database connectivity before reporting readiness. Kubernetes uses these for liveness and readiness probes.

Define a Histogram metric such as http_request_duration_seconds with labels for method, route, and status. Record each request's duration so you can compute latency SLIs and error rates in Grafana dashboards.

Create an alert using an expression like rate(http_requests_total{status=~"5.."}[5m]) > 0.05 with a 5-minute duration and a critical severity label. Route critical alerts to PagerDuty and informational ones to Slack.

Run kubectl rollout undo deployment/app to revert to the previous stable revision. For Vercel-hosted apps, use vercel rollback to restore the prior deployment.

An error budget quantifies how much unreliability your SLO permits, such as allowed downtime or failed requests. When the budget is nearly exhausted, deployments should pause until reliability recovers.