Nic Cope

> I don't understand the difference between the two scenarios and why we would want to save a list of unsynced resources in the one case and requeue later, versus...

I just did a quick sanity check of the codepath that plumbs up the rate limiter and I can't immediately see why it'd be getting lost anywhere. Also, even if...

> which is returning metav1.StatusReasonUnknown for non-API errors Shouldn't this be an API error though? e.g.: 1. We attempt to SSA apply a partially complete (missing required fields) resource via...

Nice find @ulucinar. I wonder if we can have the reconciler re-use the random names it generated and saved to `resourceRefs` on the previous reconciler, rather than changing them each...

I'm still working on https://github.com/crossplane/crossplane/pull/6777 to guard against XR watch thrashing in general. However, we already updated the controller to use deterministic names for composed resources: * https://github.com/crossplane/crossplane/pull/6630 * https://github.com/crossplane/crossplane/pull/6705...

When we do this we may also want to update the XRD reconciler to reconcile the v2 type, not the v1 type. Per https://github.com/crossplane/crossplane/blob/main/contributing/guide-api-promotion.md I don't think we can make...

@turkenh suggested an approach where the package manager changes what kind of MRs a provider installs. A Provider package is an OCI image that contains all of its CRDs -...

Here's another idea that may work: * Providers introduce new namespaced MR variants at some arbitrary semantic version (e.g. `provider-aws-ec2:v1.30.0`) * Providers maintain support for cluster scoped MR variants for...

> The biggest downside with this approach is that it doubles the number of CRDs installed. For anyone already using Crossplane and upgrading to v2 I think this is unavoidable....

> by introducing a scope parameter similar to the ones you propose adding the others: I'd rather not do this. See the main issue body for the full details, but...