log.Info("CNI ambient server kubeclient started")
	pods := s.handlers.GetAmbientPods()
	err := s.dataplane.ConstructInitialSnapshot(pods)
	if err != nil {
		log.Warnf("failed to construct initial snapshot: %v", err)
	}

	log.Info("CNI ambient server marking ready")
	s.Ready()
	s.dataplane.Start(s.ctx)
	s.handlers.Start()
}

func (s *Server) Stop() {

		consErr = append(consErr, err)
	}

	podsByUID := slices.GroupUnique(ambientPods, (*corev1.Pod).GetUID)
	if err := s.buildZtunnelSnapshot(podsByUID); err != nil {
		log.Warnf("failed to construct initial ztunnel snapshot: %v", err)
		consErr = append(consErr, err)
	}

	return errors.Join(consErr...)
}

func (s *NetServer) buildZtunnelSnapshot(ambientPodUIDs map[types.UID]*corev1.Pod) error {

	mt := monitortest.New(t)
	setupLogging()
	ctx, cancel := context.WithCancel(context.Background())
	defer cancel()

	fixture := connect(ctx)
	ztunClient := fixture.ztunClient
	uid := fixture.uid
	// read initial pod add
	readRequest(t, ztunClient)
	sendAck(ztunClient)
	// read snapshot sent
	m, fds := readRequest(t, ztunClient)
	assert.Equal(t, len(fds), 0)
	sent := m.Payload.(*zdsapi.WorkloadRequest_SnapshotSent).SnapshotSent

		// We have no good way to distinguish between these two cases from here. But we don't need to!
		// Existing pods will be handled by the dataplane using `GetAmbientPods`,
		// and the initial enqueueNamespace, and new pods will be handled by the CNI.

	case controllers.EventUpdate:
		// For update, we just need to handle opt outs
		newPod := event.New.(*corev1.Pod)
		oldPod := event.Old.(*corev1.Pod)

				gvr := action.GetResource()
				ns := action.GetNamespace()
				watch, err := fc.Tracker().Watch(gvr, ns)
				if err != nil {
					return false, nil, err
				}
				// Kubernetes Fake watches do not add initial state, but real server does
				// https://kubernetes.io/docs/reference/using-api/api-concepts/#semantics-for-watch
				// Tracking in https://github.com/kubernetes/kubernetes/issues/123109

			return err
		}
		installLog.Info("Detected changes to the node-level CNI setup, checking to see if configs or binaries need redeploying")
		// We don't support (or want) to silently (re)deploy any binaries that were not in the initial "snapshot"
		// so we intentionally discard/do not update the list of installedBins on redeploys.
		if _, err := in.installAll(ctx); err != nil {
			return err
		}

ConfigMap, Deployment, etc. definitions for a complex application. When in
doubt re-run istioctl kube-inject on deployments to get the most up-to-date changes.

It's best to do kube-inject when the resource is initially created.
`,
		Example: `  # Update resources on the fly before applying.
  kubectl apply -f <(istioctl kube-inject -f <resource.yaml>)

  # Create a persistent version of the deployment with Istio sidecar injected.

			cniConfDirOrderedFiles: []string{"minikube_cni.conf"},
		},
		{
			name:                  "confFile env var point to missing file initially and ignore different conf",
			chainedCNIPlugin:      true,
			preConfFile:           "missing_initially.conf",
			resultFileName:        "missing_initially.conflist",
			delayedConfFile:       testDataDir + "/pre/calico.conflist",

change which control plane revision handles injection, it's possible to create a revision tag "prod" and label our
namespace "istio.io/rev=prod". The "prod" revision tag could point to "1-7-6" initially and then be changed to point to "1-8-1"
at some later point.

This allows operators to change which Istio control plane revision should handle injection for a namespace or set of namespaces