To give a bit more details:

- on Oracle, I was able to figure out by myself how to create all the things I needed (VCN, internet gateways, etc.) by reading their documentation, and it overall "made sense"

- on AWS... after spending hours trying to directly use the AWS Terraform provider, I gave up and used external modules

- on GCP, I eventually succeeded, but there was a lot of trial and error, and I kept hitting problems .../...

E.g.: out of the box, you get nodes with public IP addresses. But you can only use a handful of them per region; so you quickly run out. Fine, let's use nodes with private addresses. Oh but then you need to bring in routers and subnets and more, and oops you hit another quota limit.

My final goal was to deploy 10 small k8s clusters for a workshop; the only way to do it on GCP was to deploy them in 10 different zones.

(And of course the documentation was anything but stellar 😬)

@jpetazzo about Google Cloud, I would say the quota is the main issue. I've fixed it and have been able to run workshops without problems.

Quota is a security, so it makes sense to me to be opt out if you don't need it, no?

@davinkevin

You were lucky! :)

When I contacted them, they denied the request to increase the quotas that I needed.

Also, pain points (imho) include:

- no easy way to obtain the kubeconfig file (you have to assemble it with the API server addr, cert, auth plugin, etc.), while on other providers, you usually can do something like kubernetes_cluster._.kubeconfig

- fields that are not very well named or not very well explained, for instance the "minimal_version" .../...

... for the k8s control plane; it really is an "initial_version" (and the docs didn't explain clearly the reasoning, which is to work with Terraform semantics).

Or the fact that you can have node groups inside our outside the cluster, but the docs also don't explain clearly the difference, or what you should do if you want, say, scale-to-zero node groups.

So it's a lot of trial-and-error, and the fact that clusters and nodes take a long time to provision on GCP makes it super annoying 😴

@jpetazzo From my point of view, the platform is still one of the most "dev friendly" one, so it's super important to get this kind of feedbacks!

About the `kubeconfig`, I just use the `gcloud` command to generate it for me. It's possible for you?

About `minimal_version`, you are in terraform or the UI?

I'll try to get information within the #GDE program. If you want to discuss about this (better than toots), don't hesitate to ping me to plan this 😇.

/cc @glaforge