m := podsRepaired.With(typeLabel.Value(repairType))
	log := repairLog.WithLabels("pod", pod.Namespace+"/"+pod.Name)
	key := types.NamespacedName{Name: pod.Name, Namespace: pod.Namespace}
	// We will get an event every time the pod changes. The repair is not instantaneous, though -- it will only recover
	// once the pod restarts (in CrashLoopBackoff), which can take some time.

  // implementation will use to handle pods of this class. The possible values
  // are specific to the node & CRI configuration.  It is assumed that all
  // handlers are available on every node, and handlers of the same name are
  // equivalent on every node.
  // For example, a handler called "runc" might specify that the runc OCI
  // runtime (using native Linux containers) will be used to run the containers
  // in a pod.

due to design choices in Helm.
The underlying Kubernetes resources we configure have 1000s of fields; given enough users and bespoke use cases,
eventually someone will want to customize every one of those fields.
If all fields are exposed in `values.yaml`, we end up with an massive API that is also likely worse than just using the Kubernetes API directly.

      # and doesn't spam the readiness endpoint too much
      #
      # If the probe is disabled, it is recommended to have delay=1s,period=2s,failureThreshold=30.
      # This ensures the startup is reasonable fast (polling every 2s). 1s delay is used since the startup is not often ready instantly.
      startupProbe:
        enabled: true
        failureThreshold: 600 # 10 minutes

      # Resources for the sidecar.
      resources:

  // when this priority class should be used.
  // +optional
  optional string description = 4;

  // preemptionPolicy is the Policy for preempting pods with lower priority.
  // One of Never, PreemptLowerPriority.
  // Defaults to PreemptLowerPriority if unset.
  // +optional
  optional string preemptionPolicy = 5;
}

// PriorityClassList is a collection of priority classes.
message PriorityClassList {

            "lineInterpolation": "linear",
            "lineWidth": 1,
            "pointSize": 5,
            "scaleDistribution": {
              "type": "linear"
            },
            "showPoints": "never",
            "spanNulls": false,
            "stacking": {
              "group": "A",
              "mode": "none"
            },
            "thresholdsStyle": {
              "mode": "off"
            }

  // An object that describes the pod that will be created.
  // The DaemonSet will create exactly one copy of this pod on every node
  // that matches the template's node selector (or on every node if no node
  // selector is specified).
  // The only allowed template.spec.restartPolicy value is "Always".

	}
	for _, family := range families {
		// Equiv:
		// ip rule add fwmark 0x111/0xfff pref 32764 lookup 100
		//
		// Adds in-pod rules for marking packets with the istio-specific TPROXY mark.
		// A very similar mechanism is used for sidecar TPROXY.
		//
		// TODO largely identical/copied from tools/istio-iptables/pkg/capture/run_linux.go
		inpodMarkRule := netlink.NewRule()
		inpodMarkRule.Family = family

						t.Fatal(err)
					}
					assert.Equal(t, isReady.Load(), false)
				case <-time.After(5 * time.Second):
					t.Fatal("timed out waiting for invalid configuration to be detected")
				}

				// Revert valid SA
				if err := file.AtomicCopy(filepath.Join("testdata", c.saFilename), tempDir, c.saFilename); err != nil {
					t.Fatal(err)
				}

				// Run sleepWatchInstall

This component accomplishes that in the following ways:

1. By installing a separate, very basic "CNI plugin" binary onto the node to forward low-level pod lifecycle events (CmdAdd/CmdDel/etc) from whatever node-level CNI subsystem is in use to this node agent for processing via socket.
1. By running as a node-level daemonset that: