Apply a CartsStack instance

Pre-bind the Pod Identity association

Pod Identity is an EKS API, not a Kubernetes one, so it can't live in the RGD. Create the association first (EKS allows binding a ServiceAccount that doesn't exist yet), so the carts Pod boots with credentials already wired in. The role is the same wildcard-scoped -carts-dynamo role from the ACK lab, so it already covers the kro table:

~$aws eks create-pod-identity-association --cluster-name ${EKS_CLUSTER_AUTO_NAME} \

--role-arn arn:aws:iam::${AWS_ACCOUNT_ID}:role/${EKS_CLUSTER_AUTO_NAME}-carts-dynamo \

--namespace carts-kro --service-account carts | jq '.association.associationId'

"a-..."

Apply the CartsStack instance

The instance manifest is small:

apiVersion: kro.run/v1alpha1
kind: CartsStack
metadata:
  name: carts-kro
  namespace: default
spec:
  tableName: ${EKS_CLUSTER_AUTO_NAME}-carts-kro
  namespace: carts-kro

Apply it:

~$kubectl kustomize ~/environment/eks-workshop/modules/fastpaths/eks-capabilities/kro/instance \

| envsubst | kubectl apply -f -

cartsstack.kro.run/carts-kro created

ヒント

Compare with the ACK lab: the same end state took three separate kubectl apply steps plus an aws eks create-pod-identity-association call plus a kubectl rollout restart. Here, one CartsStack resource drives the whole graph, and a platform team only writes the RGD once for everyone who needs a carts-shaped stack.

kro reconciles the six child resources in dependency order: it creates the Namespace first, then the Table (which the ACK DynamoDB controller starts provisioning in AWS), then the ConfigMap and ServiceAccount once the Namespace exists, and finally the Deployment and Service once the SA and ConfigMap they reference exist. The carts Pod boots with IAM credentials already injected by the Pod Identity association we created in the previous step.

Wait for the instance to reach ACTIVE:

~$kubectl wait cartsstack carts-kro --for=jsonpath='{.status.state}'=ACTIVE --timeout=10m

cartsstack.kro.run/carts-kro condition met

注記

This can take a few minutes. The instance only reports ACTIVE once every child is healthy, and the slowest child is the real DynamoDB table being created in AWS.

Inspect what kro created

Because the instance reached ACTIVE, kro has already created and health-checked every child. List them in the new namespace to see the whole graph from one apply:

~$kubectl -n carts-kro get table,configmap,serviceaccount,deployment,service

NAME                                    SYNCED   AGE

table.dynamodb.services.k8s.aws/items   True     ...

NAME                     DATA   AGE

configmap/carts          2      ...

configmap/kube-root-ca.crt   1  ...

NAME                   SECRETS   AGE

serviceaccount/carts   0         ...

serviceaccount/default 0         ...

NAME                    READY   UP-TO-DATE   AVAILABLE   AGE

deployment.apps/carts   1/1     1            1           ...

NAME            TYPE        CLUSTER-IP   PORT(S)   AGE

service/carts   ClusterIP   ...          80/TCP    ...

The Table shows SYNCED=True, meaning the ACK controller finished creating the real AWS table. Confirm it exists on the AWS side:

~$aws dynamodb describe-table \

--table-name "$EKS_CAP_DDB_TABLE_KRO" \

--query 'Table.TableStatus' --output text

ACTIVE

One CartsStack apply produced a full namespace of Kubernetes resources plus a real AWS DynamoDB table, with kro handling the ordering and health checks for you.

Verify the Pod can reach DynamoDB

The carts Pod has a ServiceAccount, a ConfigMap pointing at the new table, IAM credentials via Pod Identity, and a network path to DynamoDB. Wait for the Deployment to be fully rolled out (Amazon EKS Auto Mode may still be scaling a node for the Pod), then confirm Pod Identity injected the role's credentials at Pod boot:

~$kubectl rollout status -n carts-kro deployment/carts --timeout=150s

deployment "carts" successfully rolled out

~$kubectl exec -n carts-kro deployment/carts -- env \

| grep AWS_CONTAINER_CREDENTIALS_FULL_URI

AWS_CONTAINER_CREDENTIALS_FULL_URI=http://...

The AWS_CONTAINER_CREDENTIALS_FULL_URI env var being present confirms Pod Identity injected the role's credentials when the Pod was scheduled. The Pod is wired to the AWS DynamoDB table via:

• ServiceAccount → aws eks create-pod-identity-association
• → IAM role ${EKS_CLUSTER_AUTO_NAME}-carts-dynamo
• → AWS DynamoDB table ${EKS_CLUSTER_AUTO_NAME}-carts-kro

That's the kro lab done. You've defined a higher-level Kubernetes API (CartsStack) that composes ACK and native Kubernetes resources into a single resource, and a single instance produced a complete, running, IAM-bound carts service.

Optional: see the kro lab's carts in the retail store UI

By default, the retail store's ui Pod talks to carts.carts.svc, which is the ACK lab's namespace (set by RETAIL_UI_ENDPOINTS_CARTS in the base UI ConfigMap). Repoint that same variable at the kro-managed carts service in carts-kro and restart the UI:

~$kubectl -n ui set env deployment/ui RETAIL_UI_ENDPOINTS_CARTS=http://carts.carts-kro:80

~$kubectl -n ui rollout status deployment/ui --timeout=60s

deployment "ui" successfully rolled out

Expose the UI with an Ingress. Amazon EKS Auto Mode includes the AWS Load Balancer Controller, so applying an Ingress provisions a public Application Load Balancer automatically:

~$envsubst < ~/environment/eks-workshop/modules/fastpaths/eks-capabilities/ui-ingress/ingressclass.yaml \

| kubectl apply -f -

~$kubectl apply -f ~/environment/eks-workshop/modules/fastpaths/eks-capabilities/ui-ingress/ingress.yaml

Wait for the load balancer to finish provisioning (this takes a few minutes), then print its URL:

~$curl --head -X GET --retry 30 --retry-all-errors --retry-delay 15 \

--connect-timeout 30 --max-time 60 -k -s -o /dev/null \

$(kubectl get ingress -n ui ui-auto -o jsonpath="{.status.loadBalancer.ingress[*].hostname}")

~$echo "http://$(kubectl get ingress -n ui ui-auto -o jsonpath='{.status.loadBalancer.ingress[*].hostname}')"

http://k8s-ui-uiauto-....us-west-2.elb.amazonaws.com

Open that URL in your browser. Because the ALB serves the app at the domain root, you can navigate the full store, including Explore. Add a couple of items to a cart; they'll land in the kro-managed ${EKS_CLUSTER_AUTO_NAME}-carts-kro DynamoDB table:

~$aws dynamodb scan --table-name "$EKS_CAP_DDB_TABLE_KRO" \

--query 'Count' --output text

When you're done, remove the Ingress and revert the UI to the ACK lab's carts namespace:

~$kubectl delete -n ui ingress ui-auto --ignore-not-found

~$kubectl -n ui set env deployment/ui RETAIL_UI_ENDPOINTS_CARTS=http://carts.carts.svc:80

~$kubectl -n ui rollout status deployment/ui --timeout=60s