✨ lightweight dependency injection

[pwittrock]

No description provided.

[k8s-ci-robot]

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: pwittrock

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

The pull request process is described here

Needs approval from an approver in each of these files:

• OWNERS [pwittrock]

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

[pwittrock]

@DirectXMan12 WDYT?

There are a couple use cases for this:

1. Injecting client-go generated clients and informers into Controllers for use with source.Informer
2. Injecting cluster-api actuators into Controllers instead of hard wiring them.

[pwittrock]

@droot thoughts?

[droot]

Took a look. In my mind, the primary use-case for dependency injection for a way to provide user-managed-objects like Reconcilers, Sources with the objects that only manager knows about. Injecting deps which belongs completely in the user-managed-space (which is second use-case you mentioned above), not super sure about the advantages. Discuss offline next week ?

[DirectXMan12]

/kind feature

[pwittrock]

Here is an example of the problem we are trying to solve:

https://github.com/kubernetes-sigs/cluster-api-provider-gcp/blob/master/cmd/manager/main.go#L78

Take a look at how the actuator is passed around.

[droot]

@pwittrock Took a look. our current scaffolding project structure isn't flexible enough for people to initialize newReconciler from the entry point main (the way stuff is wired using pkg init()) which is forcing us to solve this problem at the layer below (controller-runtime/manager) which I am not super sure about.

[ironcladlou]

I found it easier to implement inversion of control by just skipping the SDK boilerplate and using controller-runtime as a library (part of openshift/cluster-ingress-operator#70).

In this context, I feel like the underlying problem is not lack of a dependency injection system (i.e. more framework) but instead a more IoC-friendly architecture.

That said, I acknowledge that IoC might not be important to everyone and there are many ways to go about it.

[pwittrock]

@ironcladlou I mostly agree, though I think we still will need an answer for those who want to work within the framework as structured today.

[pwittrock]

PTAL