Azure AKS insights

AWS Instance Profile Role Removed from aws-auth Breaks Node Registrationcritical

When all EKS-managed node groups using a shared instance profile are deleted, AWS removes the instance profile role from aws-auth ConfigMap, breaking Cast AI-managed nodes that rely on the same role for kubelet authentication to the API server.

Node NotReady from Memory/CPU Saturationcritical

AKS nodes enter NotReady state when memory or CPU saturation causes kubelet and containerd to become unresponsive. This occurs when resource limits are exceeded or PSI metrics indicate sustained pressure.

Pod CrashLoop from OOMKillcritical

Pods enter CrashLoopBackOff when containers are repeatedly killed by OOM (Out of Memory) killer, indicating memory limits are too restrictive or memory leaks exist.

CPU Throttling Despite Available Capacitywarning

Containers experience CFS throttling (container_cpu_cfs_throttled_seconds_total increasing) even when node CPU utilization is moderate, indicating overly restrictive CPU limits are causing performance degradation.

Kubecost Pod Insufficient Replicas Degraded Statecritical

When Kubecost addon shows 'InsufficientNumberOfReplicas' error, cost monitoring is degraded because required pod replicas are not running, often due to missing compute nodes or scheduling failures.

ETCD Database Size Approaching Limitcritical

ETCD database size (etcd_mvcc_db_total_size_in_use_in_bytes) exceeding 2GB indicates control plane stress, which can cause API server slowness and cluster instability.

API Server Rate Limiting Activewarning

API server returning 429 (Too Many Requests) errors indicates clients are being throttled, which can cause controller reconciliation delays and operational failures.

Pod Ready State Degradationwarning

KubePodReadyStateLow alert indicates unhealthy pods unable to serve traffic, which correlates with application availability issues and can precede cascading failures.