Interestingly Pulumi's own blog post (https://www.pulumi.com/blog/day-2-kubernetes-migrating-eks-nodegroups-with-zero-downtime/) shows exact problem with current infrastructure as a code tools,but doesn't admit it. No matter what you do, you fall back to bash 😞

In principle, everything that was done in Bash could have been done in Pulumi; it’s just a couple of

kubectl

commands and a loop. Pulumi’s k8s provider uses the client-go library to make API calls, and the same approach could be used to replicate the

kubectl

functionality without too much trouble. We were trying to keep the example as simple as possible for the blog post, but I agree that it’s a little jarring to switch back and forth between

pulumi

and

kubectl

. We’ll consider revising the post to show how it could be done without Bash /

kubectl

.

In principle, everything that was done in Bash could have been done in Pulumi

I can't see how. lets say we are in the state "before". can I make single commit to a CI changing only pulimi files which would do the transition in a single pulumi run?