node: disable resource managers when pod-level resources are enabled

When pod-level resources are detected, the cpu and memory manages
cannot engage because the feature is not yet compatible,
one of the main reasons being the managers only work at container level.

So, the managers has to detect if pod level resources are in use,
and turn themselves to no-operation skipping resource allocation
should that be the case.

We add an intentional loud log to inform the user, because
pods with pod-level resources landing on a node which cannot
actuate the desired spec is likely to be undesirable.

Signed-off-by: Francesco Romani <[email protected]>

Author: ffromani

pkg/kubelet/cm/cpumanager/policy_static.go

@@ -22,6 +22,7 @@ import (
 
 	v1 "k8s.io/api/core/v1"
 	utilfeature "k8s.io/apiserver/pkg/util/feature"
+	resourcehelper "k8s.io/component-helpers/resource"
 	"k8s.io/klog/v2"
 	podutil "k8s.io/kubernetes/pkg/api/v1/pod"
 	v1qos "k8s.io/kubernetes/pkg/apis/core/v1/helper/qos"
@@ -320,6 +321,11 @@ func (p *staticPolicy) Allocate(s state.State, pod *v1.Pod, container *v1.Contai
 		return nil
 	}
 
+	if utilfeature.DefaultFeatureGate.Enabled(features.PodLevelResources) && resourcehelper.IsPodLevelResourcesSet(pod) {
+		klog.V(2).InfoS("CPU Manager allocation skipped, pod is using pod-level resources which are not supported by the static CPU manager policy", "pod", klog.KObj(pod), "podUID", pod.UID)
+		return nil
+	}
+
 	klog.InfoS("Static policy: Allocate", "pod", klog.KObj(pod), "containerName", container.Name)
 	// container belongs in an exclusively allocated pool
 	metrics.CPUManagerPinningRequestsTotal.Inc()
@@ -557,6 +563,11 @@ func (p *staticPolicy) GetTopologyHints(s state.State, pod *v1.Pod, container *v
 		return nil
 	}
 
+	if utilfeature.DefaultFeatureGate.Enabled(features.PodLevelResources) && resourcehelper.IsPodLevelResourcesSet(pod) {
+		klog.V(3).InfoS("CPU Manager hint generation skipped, pod is using pod-level resources which are not supported by the static CPU manager policy", "pod", klog.KObj(pod), "podUID", pod.UID)
+		return nil
+	}
+
 	// Short circuit to regenerate the same hints if there are already
 	// guaranteed CPUs allocated to the Container. This might happen after a
 	// kubelet restart, for example.
@@ -604,6 +615,11 @@ func (p *staticPolicy) GetPodTopologyHints(s state.State, pod *v1.Pod) map[strin
 		return nil
 	}
 
+	if utilfeature.DefaultFeatureGate.Enabled(features.PodLevelResources) && resourcehelper.IsPodLevelResourcesSet(pod) {
+		klog.V(3).InfoS("CPU Manager pod hint generation skipped, pod is using pod-level resources which are not supported by the static CPU manager policy", "pod", klog.KObj(pod), "podUID", pod.UID)
+		return nil
+	}
+
 	assignedCPUs := cpuset.New()
 	for _, container := range append(pod.Spec.InitContainers, pod.Spec.Containers...) {
 		requestedByContainer := p.guaranteedCPUs(pod, &container)

pkg/kubelet/cm/memorymanager/policy_static.go

@@ -27,6 +27,7 @@ import (
 	v1 "k8s.io/api/core/v1"
 	"k8s.io/apimachinery/pkg/api/resource"
 	utilfeature "k8s.io/apiserver/pkg/util/feature"
+	resourcehelper "k8s.io/component-helpers/resource"
 	"k8s.io/klog/v2"
 	podutil "k8s.io/kubernetes/pkg/api/v1/pod"
 	corehelper "k8s.io/kubernetes/pkg/apis/core/v1/helper"
@@ -108,6 +109,11 @@ func (p *staticPolicy) Allocate(ctx context.Context, s state.State, pod *v1.Pod,
 	}
 
 	podUID := string(pod.UID)
+	if utilfeature.DefaultFeatureGate.Enabled(features.PodLevelResources) && resourcehelper.IsPodLevelResourcesSet(pod) {
+		logger.V(2).Info("Allocation skipped, pod is using pod-level resources which are not supported by the static Memory manager policy", "podUID", podUID)
+		return nil
+	}
+
 	logger.Info("Allocate")
 	// container belongs in an exclusively allocated pool
 	metrics.MemoryManagerPinningRequestTotal.Inc()
@@ -412,6 +418,11 @@ func (p *staticPolicy) GetPodTopologyHints(ctx context.Context, s state.State, p
 		return nil
 	}
 
+	if utilfeature.DefaultFeatureGate.Enabled(features.PodLevelResources) && resourcehelper.IsPodLevelResourcesSet(pod) {
+		logger.V(3).Info("Topology hints generation skipped, pod is using pod-level resources which are not supported by the static Memory manager policy", "podUID", pod.UID)
+		return nil
+	}
+
 	for _, ctn := range append(pod.Spec.InitContainers, pod.Spec.Containers...) {
 		containerBlocks := s.GetMemoryBlocks(string(pod.UID), ctn.Name)
 		// Short circuit to regenerate the same hints if there are already
@@ -442,6 +453,11 @@ func (p *staticPolicy) GetTopologyHints(ctx context.Context, s state.State, pod
 		return nil
 	}
 
+	if utilfeature.DefaultFeatureGate.Enabled(features.PodLevelResources) && resourcehelper.IsPodLevelResourcesSet(pod) {
+		logger.V(3).Info("Topology hints generation skipped, pod is using pod-level resources which are not supported by the static Memory manager policy", "podUID", pod.UID)
+		return nil
+	}
+
 	containerBlocks := s.GetMemoryBlocks(string(pod.UID), container.Name)
 	// Short circuit to regenerate the same hints if there are already
 	// memory allocated for the container. This might happen after a

pkg/kubelet/cm/memorymanager/policy_static_test.go

@@ -2074,7 +2074,7 @@ func TestStaticPolicyAllocate(t *testing.T) {
 			pod:                      getPodWithPodLevelResources("pod1", podLevelRequirementsGuaranteed, "container1", requirementsGuaranteed),
 			expectedTopologyHints:    nil,
 			topologyHint:             &topologymanager.TopologyHint{},
-			expectedError:            fmt.Errorf("Memory Manager static policy does not support pod-level resources"),
+			expectedError:            nil,
 			podLevelResourcesEnabled: true,
 		},
 	}