✨ Add index support to providers

[sttts]

A first attempt on Index support. Is this what we want?

See in particular:

1. GetFieldIndex() in the manager interface.
2. IndexField() in the provider interface.

Motivation:

Indexes must be created before cache start. They are usually uniform for multicluster controllers. Only the provider really knows how to set them up (some providers have dedicated clusters, others have one central cluster object and scoped down cluster shims on-top).

TODOs:

• add unit tests to the providers, in particular the namespace provider. This is not obvious.

[embik]

I think this makes sense, it's a relatively trivial function to implement on a provider that juggles clusters (as seen in the capi example provider), if we figure out we are missing something we can re-adjust.

[mjudeikis]

Just wondering, if somebody want to add dynamic indexes to only certain clusters?
In example:

1. I reconcile 100 clusters.
2. Only 20 clusters has Karpetner/Crossplane installed. Now my controller needs dynamically add indexes for these clusters for certain objects.

This would not be supported, right?

[sttts]

This would not be supported, right?

It depends on the provider. A cluster-api provider with real cluster.Cluster instances, there you can add indexes manually. But then this is not a uniform setup anymore. You would likely start multiple managers to separate.

But you have a point. Wondering whether we need something more flexible. I am also now calling the host cluster IndexField. And the host cluster most likely looks different to the other clusters.

[sttts]

I am also now calling the host cluster IndexField

I have changed that.

[sttts]

@mjudeikis I feel like such a custom logic is probably better suited to be put into the provider? It heavily depends on how you discover clusters, how you classify them through some means. I could imagine there to be some pre-engage and also engage-filter mechanism. Wdyt?

[mjudeikis]

I was thinking about uniform-cases here, where each cluster is separate.

But you have a point. Wondering whether we need something more flexible. I am also now calling the host cluster IndexField. And the host cluster most likely looks different to the other clusters.

For me, this was the biggest issue. How do we know these indexers will work on other clusters? Should these be per cluster?

need to read implementation more to understand how things are wired in