// not as optimal but at least always safe to do.
				runtime.UnlockOSThread()
			}
		}()

		return toRun()
	}

	var wg sync.WaitGroup
	wg.Add(1)

	// Start the callback in a new green thread so that if we later fail
	// to switch the namespace back to the original one, we can safely
	// leave the thread locked to die without a risk of the current thread

			// for the pod. If this is a pod added via informer, the pod should have already gone thru
			// the CNI plugin chain, and have a PodIP.
			//
			// If PodIPs exists, it is preferred, otherwise fallback to PodIP.
			//
			// If we get to this point and have a pod that really and truly has no IP in either of those,
			// it's not routable at this point and something is wrong/we should discard this event.

		return nil
	}
	return err
}

// Get any IPs currently assigned to the Pod.
//
// If 'PodIPs' exists, it is preferred (and should be guaranteed to contain the address in 'PodIP'),
// otherwise fallback to 'PodIP'.
//
// Note that very early in the pod's lifecycle (before all the node CNI plugin invocations finish)
// K8S may not have received the pod IPs yet, and may not report the pod as having any.

the certificates are persisted to disk, rather than kept in memory like in the standard Kubernetes deployment.

## Certificate Rotation

The agent also handles rotating certificates near expiration. It does so by triggering a callback from the `SecretManager` to the SDS server
when a certificate is near expiration (configurable by `SECRET_GRACE_PERIOD_RATIO`, defaulting to half of the expiration). If the SDS server