I'm working on deploying mini-data centers aimed at reusing heat in challenging environments where conditions aren't ideal for typical IT setups—think dust, vibration, and unstable connectivity. We're focused on IIoT and edge computing in these non-IT-friendly locations. We primarily use K3s for orchestration, manage various data sources, and ensure local data action triggers while maintaining centralized cloud monitoring. Uptime for data collection is our highest priority. Given that we can't count on perfect infrastructure, I'm debating between two hardware options: a single, high-end industrial server that's ruggedized and space-efficient, or a 3-node cluster of cheaper industrial PCs designed for high availability within a lightweight Kubernetes setup. Essentially, I'm looking for the most reliable way to implement Kubernetes at the edge and would appreciate any insights or alternative suggestions!
3 Answers
It sounds like the 3-node cluster approach is definitely the way to go, especially for maintaining uptime. Chick-fil-a has implemented a similar setup for their POS systems, and they've had great success with it. If you're in a tough environment, a high-availability system should outperform a single server setup in terms of reliability.
I kind of went with just one server for a similar project, and it’s been a bit of a gamble. We faced downtime due to node dimm failures, so you might want to think about having a single node plus a cold spare to minimize risks. That way, you can prepare for hardware issues without needing frequent site visits.
That sounds like a cost-effective solution, especially since hardware failures are rare. How can you tell when you need to replace a failing part without traveling out there?
I agree with going for multiple cheaper nodes. If you have three control plane nodes, you can lose one and still keep everything running, albeit in a degraded state. It gives you time to address issues without a full shutdown. Just make sure to prioritize your critical workloads properly!
Totally makes sense. Are there any specific setups or hardware recommendations you would suggest?
I’ll check that out, it really sounds like a fitting use case!