astonishing-exabyte-93491
03/08/2023, 7:46 PMaws.eks.Cluster
and aws.eks.NodeGroup
resource providers. I’ve done this in the past on GCP, so the strategy was pretty straight-forward:
1. Provide a new NodePool with autoscalling
enabled.
2. Cordon, then drain nodes using kubectl
.
3. Decommission the nodepool with an old kubernetes version, which equates here to remove the resource from the pulumi stack.
Again, my goal is to prevent worker nodes from being unreachable while upgrading to a more recent version.
I’m operating under the following assumptions:
1. control plane upgrades do not compromise data plane workloads in any way, with the exception that the api server will not be reachable for a few minutes.
2. There can not be more than two minor versions delta between master and worker nodes.
Your insight will be very much appreciated.
Many thanks,