✨ WIP: Cluster provider and cluster-aware controllers

[embik]

This is a continuation of #2726 and #2207, a prototype for a multi-cluster support in controller-runtime. A design proposal is at #2746 which will need updating depending on the feedback for this PR. I've reorganized code changes into different commits than the previous attempts, but tried to keep authorship of changes faithful to the commits this work is based on.

The main change from the previous PR (#2726) is the update to typed generics support and adjustment to have a BYO request type. The implications of that is a) that you will need to bring your own typed EventHandler and b) that a wrapper function is needed to inject information (mostly the cluster name) when establishing a handler.

Apart from that, the core design of this is the cluster provider that can be plugged into a manager. This unlocks different ways to discover the Kubernetes "fleet" that is being reconciled.

[k8s-ci-robot]

[APPROVALNOTIFIER] This PR is NOT APPROVED

This pull-request has been approved by: embik
Once this PR has been reviewed and has the lgtm label, please assign joelanford for approval. For more information see the Kubernetes Code Review Process.

The full list of commands accepted by this bot can be found here.

Needs approval from an approver in each of these files:

• OWNERS

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

[sprsquish]

We're internally working on a mechanism for dealing with multiple clusters, too. It's been helpful for the cluster identifier to be more flexible than a simple string. We've made it comparable.

Something like this:

type Cluster[ID comparable] interface {
  Identifier() ID
  ...
}

I understand this suggestion would cause a much larger change requiring type parameters in many places. Future work could be done to migrate.

Mostly, I wanted to raise this as something to consider and that calling the method Identifier would provide room to migrate in the future.

[embik]

I'm definitely open to that, I think renaming it to Identifier() should be easy enough! Maybe some maintainers can leave a note if they consider this a future endeavour and this would be a worthwhile change to prepare that.

[sttts]

We were there before, had a logical cluster abstraction, and then went back to string as it just simplifies everything.

@sprsquish Am curious why a string as identifier and some lookup table on the cluster provider side does not work?

[sttts]

Also, adding Name() string to the interface has no impact on the consumers of the Cluster type. But adding a type parameter forces them to pass e.g. string or whatever everywhere including the manager. I don't think that's a good idea. IMO, we should not do it.

[sprsquish]

@sttts

Am curious why a string as identifier and some lookup table on the cluster provider side does not work?

Strings work fine.. We like to work with more structure in our identifiers. For instance, working with multiple clusters across cloud providers it's nice to have a structure like this:

type ClusterID struct {
  Provider string
  Region string
  Account string
  Name string
}

It can be marshaled into a string, of course, it's just nice not to have to.

I understand the need for a simpler approach and appreciate that this has been discussed already. I think there's a way to migrate over time that follows what's been happening with handlers by introducing TypedCluster[ID comparable] and an alias for the common use case type Cluster = TypedCluster[string] (Manager would need the same treatment).

The only change to the current proposal would be to use the more generic Identifier() rather than Name(). That leaves the door open for future discussions regarding generic identifiers without having to do all the work up front.

[sttts]

If I understand your use-case correctly, this pattern should help you elegantly: https://goplay.space/#wRWyxv-GUrj

Or in other words: the type only matters in code that is ClusterID aware. Turning the string back into the rich format is one type conversion away.

[k8s-ci-robot]

@embik: The following test failed, say /retest to rerun all failed tests or /retest-required to rerun all mandatory failed tests:

Test name	Commit	Details	Required	Rerun command
pull-controller-runtime-apidiff	5c285f6	link	false	/test pull-controller-runtime-apidiff

Full PR test history. Your PR dashboard. Please help us cut down on flakes by linking to an open issue when you hit one in your PR.

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes-sigs/prow repository. I understand the commands that are listed here.