- it also sets up a UDS socket for istio-cni to send logs to this container.
    - based on config, it may run the 'repair' controller that detects pods where istio setup fails and restarts them, or created in corner cases.
    - if ambient is enabled, also runs an ambient controller, watching Pod, Namespace

- `istio-cni`
    - CNI plugin executable copied to `/opt/cni/bin`
    - currently implemented for k8s only

			}
		}
	} else {
		// If there is a specific pod name in ResourceName, use the agent in the pod.
		if len(dr.ResourceNames) != 1 {
			return nil, fmt.Errorf("`ResourceNames` must have one element when `all` flag is turned on")
		}
		slice := strings.SplitN(dr.ResourceNames[0], ".", 2)
		if len(slice) != 2 {
			return nil, fmt.Errorf("invalid resource name format: %v", slice)
		}
		podName := slice[0]
		ns := slice[1]

	// This should be fine, as the host binaries are all we have to work with here anyway,
	// as we are running within a privileged container - and we don't want to take the time to
	// redetect for each pod anyway.
	//
	// Extreme corner case:
	// If for some reason your host had both binaries, and you were injecting out-of-band
	// iptables rules within a pod context into `legacy` tables, but your host context preferred

does not impact the `Cluster` type, so we can skip generating `Cluster` in this case. Warning: Envoy currently has a bug that *requires* `Endpoints` to be pushed any time the corresponding `Cluster` is pushed, so this optimization is intentionally turned off in this specific case.

Finally, we determine which subset of the type we need to generate. XDS has two modes - "State of the World (SotW)" and "Delta". In SotW, we generally need to generate all resources of the type, even if only one...