changing proxies in mgmt cluster triggers a rolling update in workload clusters

While discussing with @jonathang we realized that !5549 (merged) has a very problematic side-effect: changing proxies in mgmt cluster triggers a rolling update in workload clusters.

We cannot set mgmt_cluster_state_values.proxies as done in !5549 (merged) because everything under mgmt_cluster_state_values is merged into values of all workload cluster values, at top-level, and propagated by workload cluster sylva-units into thec cluster unit settings, where it is injected into node-specific configuration (e.g. RKE2 or containerd proxy settings).

!5549 (merged) had the intent of propagating proxy settings of the mgmt cluster into only workload-cluster kube-job units, to address #2886 (closed)... but the implementation does not actually do that:

  • it propagates mgmt cluster proxy settings in all units of workload clusters (see explanation above)
  • I in fact think that this MR does not actually address #2886 (closed), because the modified kube-job units now read .Values.mgmt_cluster_state_values.proxies, which in workload cluster context does not contain anything coming from management cluster sylva-units values, because (see explanation above) the content of mgmt_cluster_state_values in mgmt cluster values is propagate at the top-level of workload cluster values not under mgmt_cluster_state_values

/cc @mederic.deverdilhac @priyagoyal