Container deletion from the workload

When deleting pods that use the resources of a given power profile,
the workload may be incorrectly updated, removing all existing
container entries. This error can be observed whenever containers
with the same name, but different pods, share resources of the
same workload.

This commit changes the container identification of a workload,
searching for its UID instead of its name, as well as including
workload information when admitting a powerpod.

Signed-off-by: Eduardo Juliano Alberti <[email protected]>

Author: ejalberti

controllers/powerpod_controller.go

@@ -92,10 +92,10 @@ func (r *PowerPodReconciler) Reconcile(c context.Context, req ctrl.Request) (ctr
 	if !pod.ObjectMeta.DeletionTimestamp.IsZero() || pod.Status.Phase == corev1.PodSucceeded {
 		// If the pod's deletion timestamp is not zero, then the pod has been deleted
 
-		powerPodState := r.State.GetPodFromState(pod.GetName(), pod.GetNamespace())
+		powerPodState := r.State.GetPodFromStateUID(string(pod.GetUID()))
 
-		logger.V(5).Info("removing the pod from the internal state")
-		if err = r.State.DeletePodFromState(pod.GetName(), pod.GetNamespace()); err != nil {
+		logger.V(5).Info("Removing Pod from internal state", "Pod Name", pod.GetName(), "UID", pod.GetUID())
+		if err = r.State.DeletePodFromStateUID(string(pod.GetUID())); err != nil {
 			logger.Error(err, "error removing the pod from the internal state")
 			return ctrl.Result{}, err
 		}
@@ -105,14 +105,15 @@ func (r *PowerPodReconciler) Reconcile(c context.Context, req ctrl.Request) (ctr
 		for _, container := range powerPodState.Containers {
 			workload := container.Workload
 			cpus := container.ExclusiveCPUs
+			logger.V(5).Info("Removing", "Workload", workload, "CPUs", cpus)
 			if _, exists := workloadToCPUsRemoved[workload]; exists {
 				workloadToCPUsRemoved[workload] = append(workloadToCPUsRemoved[workload], cpus...)
 			} else {
 				workloadToCPUsRemoved[workload] = cpus
 			}
 		}
 		for workloadName, cpus := range workloadToCPUsRemoved {
-			logger.V(5).Info("retrieving the workload instance %s", workloadName)
+			logger.V(5).Info("retrieving the workload instance", "Workload Name", workloadName)
 			workload := &powerv1.PowerWorkload{}
 			err = r.Get(context.TODO(), client.ObjectKey{
 				Namespace: IntelPowerNamespace,
@@ -209,6 +210,7 @@ func (r *PowerPodReconciler) Reconcile(c context.Context, req ctrl.Request) (ctr
 
 			workloadContainer := container
 			workloadContainer.Pod = pod.Name
+			workloadContainer.Workload = workloadName
 			containerList = append(containerList, workloadContainer)
 		}
 		for i, newContainer := range containerList {
@@ -280,9 +282,11 @@ func (r *PowerPodReconciler) getPowerProfileRequestsFromContainers(containers []
 		containerID := getContainerID(pod, container.Name)
 		coreIDs, err := r.PodResourcesClient.GetContainerCPUs(pod.GetName(), container.Name)
 		if err != nil {
+			logger.V(5).Info("Error getting CoreIDs.", "ContainerID", containerID)
 			return map[string][]uint{}, []powerv1.Container{}, err
 		}
 		cleanCoreList := getCleanCoreList(coreIDs)
+		logger.V(5).Info("Reserving cores to container.", "ContainerID", containerID, "Cores", cleanCoreList)
 
 		logger.V(5).Info("creating the power container")
 		powerContainer := &powerv1.Container{}
@@ -347,7 +351,7 @@ func getNewWorkloadContainerList(nodeContainers []powerv1.Container, podStateCon
 
 	logger.V(5).Info("checking if there are new containers for the workload")
 	for _, container := range nodeContainers {
-		if !isContainerInList(container.Name, podStateContainers, logger) {
+		if !isContainerInList(container.Name, container.Id, podStateContainers, logger) {
 			newNodeContainers = append(newNodeContainers, container)
 		}
 	}
@@ -356,9 +360,9 @@ func getNewWorkloadContainerList(nodeContainers []powerv1.Container, podStateCon
 }
 
 // Helper function - if container is in a list of containers
-func isContainerInList(name string, containers []powerv1.Container, logger *logr.Logger) bool {
+func isContainerInList(name string, uid string, containers []powerv1.Container, logger *logr.Logger) bool {
 	for _, container := range containers {
-		if container.Name == name {
+		if container.Name == name && container.Id == uid {
 			return true
 		}
 	}

pkg/podstate/podstate.go

@@ -38,6 +38,16 @@ func (s *State) GetPodFromState(podName string, podNamespace string) powerv1.Gua
 	return powerv1.GuaranteedPod{}
 }
 
+func (s *State) GetPodFromStateUID(podUID string) powerv1.GuaranteedPod {
+	for _, existingPod := range s.GuaranteedPods {
+		if existingPod.UID == podUID {
+			return existingPod
+		}
+	}
+
+	return powerv1.GuaranteedPod{}
+}
+
 func (s *State) GetCPUsFromPodState(podState powerv1.GuaranteedPod) []uint {
 	cpus := make([]uint, 0)
 	for _, container := range podState.Containers {
@@ -47,6 +57,16 @@ func (s *State) GetCPUsFromPodState(podState powerv1.GuaranteedPod) []uint {
 	return cpus
 }
 
+func (s *State) DeletePodFromStateUID(podUID string) error {
+	for i, pod := range s.GuaranteedPods {
+		if pod.UID == podUID {
+			s.GuaranteedPods = append(s.GuaranteedPods[:i], s.GuaranteedPods[i+1:]...)
+		}
+	}
+
+	return nil
+}
+
 func (s *State) DeletePodFromState(podName string, podNamespace string) error {
 	for i, pod := range s.GuaranteedPods {
 		if pod.Name == podName && pod.Namespace == podNamespace {