n8n is great as a prototyping playground, but production demands a completely different discipline. After 30 deployments we know what separates a workflow that ships from one that wakes you up at 3 AM.
Five things we always do
- Self-hosted from day one. n8n Cloud is great for an MVP, but a production deployment with SLA must run on your infrastructure - control over upgrades, audit logs, GDPR.
- Every workflow has an error workflow. We never leave an execution without a catch-all - errors go to Slack or your ticketing system.
- Retries with exponential backoff. Three attempts: 5s / 30s / 120s. 90% of errors are transient timeouts.
- Webhook auth always. Even for “internal” endpoints. HMAC signature or JWT.
- Versioning through git. Export JSON to repo + CI that auto-imports to staging.