What happened
Fairwinds published a write-up on running n8n in production on Amazon EKS via the CNCF blog. They ran it in queue mode for about a year on a community Helm chart, then moved to n8n’s official chart. They kept the database, queue, object storage, IAM role and secrets, and recreated everything else.
Why it matters
n8n is easy to start with a single container. Production is different: workers, webhook processors, PostgreSQL, Redis, object storage, external secrets and ingress all need to work together. Small teams often adopt n8n for internal automation and webhooks, and it is now also used for AI workloads such as agents and RAG pipelines, which can put more load on workers and secrets.
The article’s main lesson is about state. The two charts used different workload models for the main pod. The old one used a StatefulSet with a volume per replica. In the authors’ setup the new one used a Deployment with one shared volume, and their gp3 EBS volume could not attach to several main pods across nodes. An in-place upgrade did not fit, so they did a full cutover. Before that, they checked what lived on the main pod’s disk. Nothing needed to persist, so they turned main persistence off.
If your automation platform runs on community charts with local patches, that maintenance cost grows quietly until a migration forces the issue.
What to do this week
- List where your n8n (or any similar tool) keeps state: database, queue, object storage, secrets and local disk. Confirm what, if anything, must survive a pod restart.
- Check which Helm chart you use, who maintains it, and how many local patches you carry.
- Keep secrets such as the encryption key outside the cluster, for example in a managed secrets store, and sync them in.
- Rehearse any migration in non-production using a restored database snapshot, and test rollback before touching live traffic.
- Change one thing at a time. The authors left the n8n image version alone during cutover so a database migration would not be part of it.
Need a hand?
If you run workflow tools or other stateful apps on Kubernetes and a chart change or upgrade is coming, our team can review the design and plan a safe cutover with you. Have a look at our Kubernetes consulting page. The first call is free and it is with an engineer, not a sales person.
Source: Deploying, maintaining, and migrating production n8n on EKS
Related service: Kubernetes Consulting