Under the hood: how Amazon EKS Auto Mode detects, repairs, and diagnoses node failures

This article explains how Amazon EKS Auto Mode automatically detects, repairs, and diagnoses node failures through the Node Monitoring Agent and Karpenter.

• Node Monitoring Agent detects faults and translates them into Kubernetes node conditions
• Karpenter consumes conditions and automatically replaces unhealthy nodes without manual intervention
• Condition-severity detections trigger node replacement; Event-severity detections remain informational
• Safety threshold prevents cascading failures by capping repairs at 20% of nodes in a NodePool
• Toleration windows (10 minutes for hardware, 30 minutes for other faults) allow transient issues to self-resolve
• kubectl ekslogs plugin enables diagnostics collection without SSH access to managed instances
• Detection and diagnosis are separate concerns: detection runs continuously for repair, diagnosis runs on-demand for investigation

EKS Auto Mode eliminates manual node repair workflows by automating detection, replacement, and diagnostics through integrated components and safety guardrails.