bump eks module version (#220)

* feat: Upgrade eks and addons to 1.21, improve k8s upgrade docs

* fix: Moved import statements in k8s cluster doc

* feat: Use new eks module which enables EKs prefix delegation, to allow more IPs per node

* docs: Added a note about upgrading addons

* docs: typo fix

* fix: Fixed eks and addon versions

* fix: Fixed eks and addon versions

Author: bmonkman

doc-site/docs/guides/kubernetes-cluster.md

@@ -210,7 +210,7 @@ aws eks describe-addon-versions \
 </Tabs>
 The process should be:
 
-- Update the addon versions versions in `<env>/main.tf` to match the correct versions for the new cluster version. See the tabs above or [read the AWS docs here](https://docs.aws.amazon.com/eks/latest/userguide/eks-add-ons.html)
+- Update the addons in `<env>/main.tf` to match the correct addon versions for the new cluster version. See the tabs above or [read the AWS docs here](https://docs.aws.amazon.com/eks/latest/userguide/eks-add-ons.html)
 - Update the `eks_cluster_version` in terraform
 - Run `terraform apply`
 - Update your nodes to the new version
@@ -222,6 +222,15 @@ The process should be:
 
 This will bring up new nodes, gracefully drain your workloads onto them while preventing new pods from being scheduled to the old ones, then take down the old nodes. If your workloads are set up with multiple replicas there should be no downtime during this process.
 
+:::note
+You can't upgrade the addons by more than one minor version at a time. So for example you couldn't change from "v1.7.10" to "v1.9.0". You would first need to go to "v1.8.0", apply, then update to "v1.9.0".
+:::
+
+:::tip
+Sometimes you may need to make changes to node groups that would be destructive to the nodes in your cluster, for example changing the type of instances in the group. In situations like this you can apply in stages where you would add a new node group with a unique name and the new instance types. You can apply to bring up this new group of instances, then remove the old group and apply again. This will make sure you always have sufficient nodes in your cluster to handle your workloads.
+:::
+
+
 ## More resources
 
 [https://kubernetes.io/docs/reference/kubectl/cheatsheet/](https://kubernetes.io/docs/reference/kubectl/cheatsheet/)

templates/kubernetes/terraform/modules/kubernetes/metrics_server.tf

@@ -9,12 +9,11 @@ resource "helm_release" "metrics_server" {
   # Switch away from this bitnami one when possible. https://github.com/kubernetes-sigs/metrics-server/issues/572
   repository = "https://charts.bitnami.com/bitnami"
   chart      = "metrics-server"
-  version    = "5.0.2"
+  version    = "5.9.2"
   namespace  = local.metrics_server_namespace
 
   set {
     name  = "apiService.create"
     value = true
   }
 }
-

templates/terraform/environments/prod/main.tf

@@ -56,12 +56,12 @@ module "prod" {
   ecr_repositories = [] # Should be created by the staging environment
 
   # EKS configuration
-  eks_cluster_version = "1.20"
+  eks_cluster_version = "1.21"
   # Cluster addons. These often need to be updated when upgrading the cluster version.
   # See: https://docs.aws.amazon.com/eks/latest/userguide/eks-add-ons.html
-  eks_addon_vpc_cni_version    = "v1.7.10-eksbuild.1"
-  eks_addon_kube_proxy_version = "v1.20.4-eksbuild.2"
-  eks_addon_coredns_version    = "v1.8.3-eksbuild.1"
+  eks_addon_vpc_cni_version    = "v1.9.0-eksbuild.1"
+  eks_addon_kube_proxy_version = "v1.21.2-eksbuild.2"
+  eks_addon_coredns_version    = "v1.8.4-eksbuild.1"
 
   eks_node_groups = {
     main = {
@@ -71,8 +71,6 @@ module "prod" {
       # Enable use of spot instances instead of on-demand.
       # This can provide significant cost savings and should be stable due to the use of the termination handler, but means that individuial nodes could be restarted at any time. May not be suitable for clusters with long-running workloads
       use_spot_instances = false
-      # This is the normal image. Other possibilities are AL2_x86_64_GPU for gpu instances or AL2_ARM_64 for ARM instances
-      ami_type           = "AL2_x86_64"
     }
   }
 

templates/terraform/environments/stage/main.tf

@@ -70,7 +70,7 @@ module "stage" {
   eks_cluster_version = "1.21"
   # Cluster addons. These often need to be updated when upgrading the cluster version.
   # See: https://docs.aws.amazon.com/eks/latest/userguide/eks-add-ons.html
-  eks_addon_vpc_cni_version    = "v1.7.10-eksbuild.1"
+  eks_addon_vpc_cni_version    = "v1.9.0-eksbuild.1"
   eks_addon_kube_proxy_version = "v1.21.2-eksbuild.2"
   eks_addon_coredns_version    = "v1.8.4-eksbuild.1"
 
@@ -82,8 +82,6 @@ module "stage" {
       # Enable use of spot instances instead of on-demand.
       # This can provide significant cost savings and should be stable due to the use of the termination handler, but means that individuial nodes could be restarted at any time. May not be suitable for clusters with long-running workloads
       use_spot_instances = true
-      # This is the normal image. Other possibilities are AL2_x86_64_GPU for gpu instances or AL2_ARM_64 for ARM instances
-      ami_type           = "AL2_x86_64"
     }
   }
 

templates/terraform/modules/environment/main.tf

@@ -56,7 +56,7 @@ data "aws_caller_identity" "current" {}
 # Provision the EKS cluster
 module "eks" {
   source  = "commitdev/zero/aws//modules/eks"
-  version = "0.4.5"
+  version = "0.5.0"
   providers = {
     aws = aws.for_eks
   }

templates/terraform/modules/environment/variables.tf

@@ -47,14 +47,8 @@ variable "eks_addon_coredns_version" {
 }
 
 variable "eks_node_groups" {
-  type = map(object({
-    instance_types     = list(string)
-    asg_min_size       = string
-    asg_max_size       = string
-    use_spot_instances = bool
-    ami_type           = string
-  }))
-  description = "Map of maps of eks node group config where keys are node group names"
+  type = any
+  description = "Map of maps of eks node group config where keys are node group names. See the EKS module documentation for details"
 }
 
 variable "hosted_domains" {