Self-Hosting n8n in Production: The Guide We Wish We Had

When self-hosting is the right call

Self-hosting wins on cost past roughly one hundred thousand executions per month, and on compliance the moment regulated data enters a workflow. Below that, the managed cloud is almost always cheaper once you price in an on-call rotation.

Queue mode is not optional

Single-process mode is fine for a homelab and dangerous in production. Split the main process, workers and webhook receivers, put Redis in front, and Postgres behind. Every serious incident we investigated started with someone skipping this step.

The observability stack

Structured logs, execution metrics scraped into Prometheus and a dashboard per workflow. Alert on failure rate and queue depth, not CPU. The interesting failures never touch CPU.