vllry / cluster-reconciler Goto Github PK

Suggestion from @deads2k: look at runtime.RawExtension type.

I tried to deploy the reconciler to a cluster to do some work, but it's unclear what the intended development flow is. (e.g. make docker-build is broken, and when fixed, the resulting pod CrashLoops due to argument errors).

It would really help to have a few steps/commands briefly documented, outlining initial setup, and the build/test integration loop.

Document scenarios where (perceived) intent of the hub API is liable to clash with intended state in the cluster.

e.g. a user may want to ensure an object's labels aren't changed by hand, but also respect specific automation-managed labels.

The current model handles garbage collection by reconciling the complete desired state of the cluster with all resources in the cluster. This requires fetching a large amount of data, and requires that the controller know about the sum of all desired state at once.

Tracking objects and garbage collection for individual manifest bundles would allow for explicit garbage collection of individual objects (e.g. "this object used to be in the desired state, now it's not"), with no dynamic discovery. It would also allow the reconciler to reconcile without knowing the full desired state. In turn, this allows operations to be split up, and would allow the reconciler to safely pull workload data from multiple hubs.

This could be implemented on the hub API side (as status fields?), or on the reconciler-side.