ssh to managed nodes via bastion host
check kubelet status
ssh -i vg-karp-ssh.pem [email protected]
kubectl get pods --all-namespaces -o wide | grep ip-10-1-4-117
drain node so that pod can be schedule on helathy node
pod capacity of server
17 pod per node for t3a.medium
for checking cpu and memory consuption
check cpu and memory via metric server
kubectl top node
hpa need prerequisites as metric server
eks uses containerd as CRI plugin
VPC CNI plugin
EBS CSI driver for storage
total pod running on eks
kubectl get pods --all-namespaces --no-headers | wc -l
pod limit because of vpc cni depend on ec2 type as t3a.medium have 17 pod limit
check nodes
aws eks list-nodegroups
aws eks describe-nodegroup --cluster-name VG-KARP-CLUSTER
eks/k8s event
kubectl get events --all-namespaces --sort-by=.metadata.creationTimestamp
Kubernetes events are ephemeral. By default:
Retention is very short (usually ~1 hour in EKS).
Older events get garbage-collected automatically.
kubectl get events only shows events currently stored in etcd, so anything older than that is gone.
dashboard to build for k8s
total nodes
nodes cpu, memory, disk usage
each node capacity totla cluster pod capacity
total used pods
karpenter events
kubelet status
check pod
kubectl get pods -A
show pod status - running/pending
check did it restarted
cluster access
aws eks update-kubeconfig --name VG-KARP-CLUSTER --region ap-southeast-1
make sure aws configure user have access to clusyter
StatefulSets are mostly immutable. You cannot change fields like:
volumeClaimTemplates (size, storageClass) after creation
Certain spec fields other than replicas, template, updateStrategy, minReadySeconds
Helm tries to apply your values.yaml → Kubernetes rejects it because the StatefulSet spec changed in an immutable way.
AllowVolumeExpansion: → behaves as false
your current ebs-csi StorageClass is missing
every pod have service account but it does not have cluster level permission
check what default pod service account have permission
