search

kubernetes-sigs / cluster-api

Home for Cluster API, a subproject of sig-cluster-lifecycle
https://cluster-api.sigs.k8s.io
Apache License 2.0
3.6k stars 1.32k forks source link

BootstrapConfig should be optional for MachinePool based ClusterClass #10943

Open shyamradhakrishnan opened 4 months ago

shyamradhakrishnan commented 4 months ago

What steps did you take and what happened?

In managed clusters, while using ClusterClass, we are stuck to create MachinePools as it is forcing providing a BootstrapConfig. In non ClusterClass machinepools, BootstrapConfig is optional. For example https://github.com/oracle/cluster-api-provider-oci/blob/main/templates/cluster-template-managed.yaml#L50

What did you expect to happen?

BootstrapConfig should be optional in MachinePools based on ClusterClass. Specifically https://doc.crds.dev/github.com/kubernetes-sigs/cluster-api/cluster.x-k8s.io/ClusterClass/v1beta1@v1.7.4#spec-workers-machinePools-template-bootstrap-ref should be optional

Cluster API version

v1.7.0

Kubernetes version

No response

Anything else you would like to add?

No response

Label(s) to be applied

/kind bug One or more /area label. See https://github.com/kubernetes-sigs/cluster-api/labels?q=area for the list of labels.

sbueringer commented 4 months ago

We have to compare with Machines, but I wonder how "optional" bootstrap config is there. E.g. is it possible to just not set ref, or do folks also have to set an empty dataSecretName field (which seems more like a workaround then bootstrap config properly being optional, like it should be)

But agree with the general idea of this issue.

Would be great if other folks interested in MachinePools would also chime in

/area machinepool /triage accepted

fabriziopandini commented 3 months ago

As far as I remember either you define a bootstrap config ref, which generates a bootstrap data secret, or you provide a bootstrap data secret.

If we allow empty data secrets seems more a bug than a feature, but I have to dig into it a little bit more

shyamradhakrishnan commented 3 months ago

@fabriziopandini as long as we have ability to ignore bootstrap config, we should be oke. For example please see this https://github.com/kubernetes-sigs/cluster-api-provider-aws/blob/main/templates/cluster-template-rosa-machinepool.yaml#L65 , all providers do the same for managed machinepools are they dont require bootstrap.

fabriziopandini commented 3 months ago

@mboersma to chime in If we want first-class support for the use case where MP are used without bootstrap config, e.g. for managed machine pools, we probably should think to a better API UX (as well as impacts to MPM)

zihuaweng commented 3 months ago

Hi @fabriziopandini @mboersma we would like to use ClusterClass with managed machine pools and this is blocking us, do you have any updates?

fabriziopandini commented 3 months ago

No updates, I can only reiterate/expand on previous comments from an API design stand point:

cc @JoelSpeed for an opinion about API design

sbueringer commented 3 months ago

@fabriziopandini Just looked it up, and I saw that bootstrap.ref is indeed required for MachinePools and MachineDeployments. I assume for MachineDeployments it should definitely be optional? Is this something we should include in the v1beta2 umbrella issue?

JoelSpeed commented 3 months ago

I would expect the API to allow you to either specify a bootstrap template, or the data secret directly, and require that at least one of those be required.

Technically making a required field optional can be done without changing the API version, the main thing is to make sure consumers of the API understand the change. Would we say we control the consumers? The controllers? Or is there a wide enough blast radius (many providers) that we would rather be more cautious?

Looking at the MachineSpec used in MD and MP, as far as I can tell, the bootstrap field itself is required, but there are no required fields within it, that seems like an omission, a minProperties: 1 would have fixed this I think.

The data secret name probably also wants some validation, a minLength, maxLength, a regex since we know what the valid name format of a secret in K8s is.

In the ClusterClass MachinePoolTemplate, the shape of the API is different and is just a local object reference, which is curious, why don't we allow the option of a template or data secret there, that one does seem a little odd, I don't have the context

fabriziopandini commented 3 months ago

@sbueringer, Let's keep this discussion focused on MachinePool (validating API for MD) is tracked in the umbrella issue linked above.

@JoelSpeed thanks for all the hints!

I would expect the API to allow you to either specify a bootstrap template, or the data secret directly, and require that at least one of those be required. The data secret name probably also wants some validation...

Agreed, this is is the same point I'm making about the MP API, which todays lack of validation for data secret name. This allowed what IMO is a bug/a bad API design, that was exploited to address some legit use case.

But now that we are aware of this, and we are looking to graduate our APIs, I think that this should be properly addressed considering the whole story (e.g. the proposal about MPM).

In the ClusterClass MachinePoolTemplate, the shape of the API is different and is just a local object reference, which is curious, why don't we allow the option of a template or data secret there, that one does seem a little odd, I don't have the context

Based on my understanding, this is because it does not make sense to define a data secret with a cloud-init/ignition files that applies to all the clusters/MP originated by a CC. But this is is also where we need folks actually using and maintaining this feature to step in

k8s-triage-robot commented 19 hours ago

The Kubernetes project currently lacks enough contributors to adequately respond to all issues.

This bot triages un-triaged issues according to the following rules:

You can:

Please send feedback to sig-contributor-experience at kubernetes/community.

/lifecycle stale