05/19/2023, 5:47 PM
🧵 Pulumi 🥊 toolchains like helm, manifests, argocd
I rebuilt a helm service in Pulumi last year. Was a massive shift and got close but it was shut down and folks used helm at the end. I have another project coming up where folks are interested in gitops approach and think of using argocd now. I’ve heard the Pulumi operator is used by some big orgs (say DoorDash) so it’s not like it’s unused. I know it can run helm, apply yaml, or apply Pulumi plans. Here’s my main questions for those who have more experience in both Pulumi with k8s AND.m tools like argocd. • assuming your applying helm charts and not full Pulumi sdk code, what would sell you on using argocd vs Pulumi to apply? • The biggest problem I faced was Pulumi tracks stage. It sorta doesn’t fit in the model of k8s which does expect a desired state but it’s more a live comparison on helm chart version, or manifest not a database style tracking like Pulumi. In fact the biggest source of friction on my pilot was the state file (can explain in more detail if asked) where certain resources conflicted and had to do a lot do refactoring to support multi cluster built name space scale deployments. • is there any substantial difference to Pulumi applying vs argocd applying gitops? I’ve not used either. I was deploying my k8s more like traditional infra jobs from an agent in a pipeline. • Caveat I can’t use Pulumi service directly at this stage for that as I’d have to store state in azure blob storage and then work on possible vendor onboarding to centralize in Pulumi cloud.