We're using KRO, ACK, and Argo CD on a moderately large platform and are evaluating whether EKS Auto Mode would fit into the setup. I'm especially interested in real-world experience with Auto Mode's managed compute and node provisioning alongside KRO-managed resources, including reconciliation issues or drift. Have you seen conflicts with AWS resources provisioned through ACK? How does Argo CD sync ordering behave with KRO ResourceGraphDefinitions when Auto Mode is enabled? Would you recommend Auto Mode, or is Karpenter with managed or self-managed node groups still the better option? We'd like to understand the operational and cost trade-offs before committing.
3 Answers
Before making KRO central to the platform, check its maturity and API stability. The ResourceGraphDefinition API has been described as subject to change and may not be ready for every production use case. Also consider whether Crossplane is a better fit than ACK if you want a more established, broadly adopted control-plane approach.
We tested this combination and removed Auto Mode fairly quickly. The reconciliation delays were enough to produce recurring drift warnings in Argo CD when KRO resources were involved. It may work for simpler deployments, but we found the timing and debugging overhead harder to justify than continuing with our existing node provisioning approach.
KRO and ACK can run with Auto Mode because the controllers operate in the managed control plane rather than on worker nodes, so they generally aren't affected by Auto Mode's immutable node model. The bigger concerns are Auto Mode's documented limitations, the premium for managed operations, and whether on-demand-only capacity works for your workload. If you're currently using Spot capacity with Karpenter, compare the management fee and reduced flexibility against the engineering time saved on patching and scaling.
Related Questions
Can't Load PhpMyadmin On After Server Update
Redirect www to non-www in Apache Conf
How To Check If Your SSL Cert Is SHA 1
Windows TrackPad Gestures