This message was deleted.

Is the only option to shell out to kubectl?

It would be good to open an issue in pulumi-kubernetes with details of the use cases here. We did add a

kubectlReplace

function to the Kubernetes provider related to this - but I'm not sure it really completely solves for common scenarios - and would love to have an issue tracking these to make sure we think about what a great solution here would look like. https://github.com/pulumi/pulumi-kubernetes/pull/579

This is killing us a bit. We've written our own Python eks pulumi library that stands up the control plane/worker nodes, and we're flying with it, however we have a massive hole in our upgrade story as we're unable to nicely upgrade kube-proxy/coredns/aws-vpc-cni.

You can see how this is handled in the

@pulumi/eks

package for Node.js here: https://github.com/pulumi/pulumi-eks/blob/master/nodejs/eks/cni.ts#L160

Very intimately familiar, we've written our own dynamic provider code using a python kubernetes lib (instead of shelling out). just unfortunate that it has to be done that way.

FWIW - this is an issue tracking this - definitely feel free to add a note about your scenario and an upvote: https://github.com/pulumi/pulumi-kubernetes/issues/264