aws_eks: add_manifest doesn't apply and always creates

[idomic]

Describe the bug

We have this code to launch a router:

 # Virtual Service
        virtual_service_manifest = {
            "apiVersion": "appmesh.k8s.aws/v1beta2",
            "kind": "VirtualService",
            "metadata": {"name": "apps-router", "namespace": namespace},
            "spec": {
                "awsName": "apps-router.apps.svc.cluster.local",
                "provider": {
                    "virtualRouter": {"virtualRouterRef": {"name": "apps-router"}}
                },
            },
        }
        self.cluster.add_manifest(
            "VirtualService", virtual_service_manifest
        ).node.add_dependency(apps_namespace)

Whenever there's a new change in the stack it deletes all of the routes within the router, I assume it happens since cdk defaults to kubectl create instead of kubectl apply.
I also tried calling the typescript module directly with the overwrite parameter, which didn't seem to work as expected.

Regression Issue

• Select this option if this issue appears to be a regression.

Last Known Working CDK Version

No response

Expected Behavior

I expected it to apply the manifest so it detects no changes and move on without deleting the routes.

Current Behavior

The routes are being deleted on any change in the cdk stack regardless to that resource.

Reproduction Steps

Deploy this resource and add a new one to the stack, deploy again.

Possible Solution

No response

Additional Information/Context

No response

CDK CLI Version

2.163.1

Framework Version

No response

Node.js Version

v20.8.1

OS

Macos

Language

Python

Language Version

No response

Other information

No response

[pahud]

Hi

aws-cdk/packages/@aws-cdk/custom-resource-handlers/lib/aws-eks/kubectl-handler/apply/__init__.py

Lines 55 to 74 in fe3af93

	if request_type == 'Create':
	# if "overwrite" is enabled, then we use "apply" for CREATE operations
	# which technically means we can determine the desired state of an
	# existing resource.
	if overwrite:
	kubectl('apply', manifest_file, *kubectl_opts)
	else:
	# --save-config will allow us to use "apply" later
	kubectl_opts.extend(['--save-config'])
	kubectl('create', manifest_file, *kubectl_opts)
	elif request_type == 'Update':
	if prune_label is not None:
	kubectl_opts.extend(['--prune', '-l', prune_label])
	kubectl('apply', manifest_file, *kubectl_opts)
	elif request_type == "Delete":
	try:
	kubectl('delete', manifest_file)
	except Exception as e:
	logger.info("delete error: %s" % e)

kubectl by default uses "apply" when request_type == 'Update'

You can consider to follow this debugging tips and see the full kubectl command executed by lambda
#32187 (comment)
And run that command manually using kubectl from your terminal to validate its behavior.

[github-actions]

This issue has not received a response in a while. If you want to keep this issue open, please leave a comment below and auto-close will be canceled.