Merge pull request #129477 from felipeagger/feat/improve-alloc-res-ckpt

[FG:InPlacePodVerticalScaling] Improve allocated resources checkpointing

pkg/kubelet/status/state/state.go

@@ -47,6 +47,7 @@ type Reader interface {
 
 type writer interface {
 	SetContainerResourceAllocation(podUID string, containerName string, alloc v1.ResourceRequirements) error
+	SetPodResourceAllocation(podUID string, alloc map[string]v1.ResourceRequirements) error
 	SetPodResizeStatus(podUID string, resizeStatus v1.PodResizeStatus)
 	Delete(podUID string, containerName string) error
 }

pkg/kubelet/status/state/state_checkpoint.go

@@ -127,6 +127,17 @@ func (sc *stateCheckpoint) SetContainerResourceAllocation(podUID string, contain
 	return sc.storeState()
 }
 
+// SetPodResourceAllocation sets pod resource allocation
+func (sc *stateCheckpoint) SetPodResourceAllocation(podUID string, alloc map[string]v1.ResourceRequirements) error {
+	sc.mux.Lock()
+	defer sc.mux.Unlock()
+	err := sc.cache.SetPodResourceAllocation(podUID, alloc)
+	if err != nil {
+		return err
+	}
+	return sc.storeState()
+}
+
 // SetPodResizeStatus sets the last resize decision for a pod
 func (sc *stateCheckpoint) SetPodResizeStatus(podUID string, resizeStatus v1.PodResizeStatus) {
 	sc.mux.Lock()
@@ -165,6 +176,10 @@ func (sc *noopStateCheckpoint) SetContainerResourceAllocation(_ string, _ string
 	return nil
 }
 
+func (sc *noopStateCheckpoint) SetPodResourceAllocation(_ string, _ map[string]v1.ResourceRequirements) error {
+	return nil
+}
+
 func (sc *noopStateCheckpoint) SetPodResizeStatus(_ string, _ v1.PodResizeStatus) {}
 
 func (sc *noopStateCheckpoint) Delete(_ string, _ string) error {

pkg/kubelet/status/state/state_checkpoint_test.go

@@ -110,11 +110,9 @@ func Test_stateCheckpoint_storeState(t *testing.T) {
 			originalSC, err := NewStateCheckpoint(testDir, testCheckpoint)
 			require.NoError(t, err)
 
-			for podUID, containerAlloc := range tt.args.podResourceAllocation {
+			for podUID, alloc := range tt.args.podResourceAllocation {
-				for containerName, alloc := range containerAlloc {
+				err = originalSC.SetPodResourceAllocation(podUID, alloc)
-					err = originalSC.SetContainerResourceAllocation(podUID, containerName, alloc)
+				require.NoError(t, err)
-					require.NoError(t, err)
-				}
 			}
 
 			actual := originalSC.GetPodResourceAllocation()

pkg/kubelet/status/state/state_mem.go

@@ -77,6 +77,15 @@ func (s *stateMemory) SetContainerResourceAllocation(podUID string, containerNam
 	return nil
 }
 
+func (s *stateMemory) SetPodResourceAllocation(podUID string, alloc map[string]v1.ResourceRequirements) error {
+	s.Lock()
+	defer s.Unlock()
+
+	s.podAllocation[podUID] = alloc
+	klog.V(3).InfoS("Updated pod resource allocation", "podUID", podUID, "allocation", alloc)
+	return nil
+}
+
 func (s *stateMemory) SetPodResizeStatus(podUID string, resizeStatus v1.PodResizeStatus) {
 	s.Lock()
 	defer s.Unlock()

pkg/kubelet/status/status_manager.go

@@ -310,25 +310,23 @@ func (m *manager) GetPodResizeStatus(podUID types.UID) v1.PodResizeStatus {
 func (m *manager) SetPodAllocation(pod *v1.Pod) error {
 	m.podStatusesLock.RLock()
 	defer m.podStatusesLock.RUnlock()
+
+	podAlloc := make(map[string]v1.ResourceRequirements)
 	for _, container := range pod.Spec.Containers {
 		alloc := *container.Resources.DeepCopy()
-		if err := m.state.SetContainerResourceAllocation(string(pod.UID), container.Name, alloc); err != nil {
+		podAlloc[container.Name] = alloc
-			return err
-		}
 	}
 
 	if utilfeature.DefaultFeatureGate.Enabled(features.SidecarContainers) {
 		for _, container := range pod.Spec.InitContainers {
 			if podutil.IsRestartableInitContainer(&container) {
 				alloc := *container.Resources.DeepCopy()
-				if err := m.state.SetContainerResourceAllocation(string(pod.UID), container.Name, alloc); err != nil {
+				podAlloc[container.Name] = alloc
-					return err
-				}
 			}
 		}
 	}
 
-	return nil
+	return m.state.SetPodResourceAllocation(string(pod.UID), podAlloc)
 }
 
 // SetPodResizeStatus checkpoints the last resizing decision for the pod.