UPSTREAM: <carry>: kubelet/cm: move CPU reading from cm to cm/cpumanager

Adapted from kubernetes@77d03e4 to avoid merge conflict
This can be dropped in the 4.19 rebase

Authored-by: Francesco Romani <fromani@redhat.com>
Signed-off-by: Peter Hunt <pehunt@redhat.com>

Author: haircommander

pkg/kubelet/cm/container_manager_linux.go

@@ -34,7 +34,6 @@ import (
 	"github.com/opencontainers/runc/libcontainer/configs"
 	"k8s.io/klog/v2"
 	"k8s.io/mount-utils"
-	"k8s.io/utils/cpuset"
 	utilpath "k8s.io/utils/path"
 
 	libcontaineruserns "github.com/opencontainers/runc/libcontainer/userns"
@@ -132,10 +131,6 @@ type containerManagerImpl struct {
 	topologyManager topologymanager.Manager
 	// Interface for Dynamic Resource Allocation management.
 	draManager dra.Manager
-	// The full set of CPUs on the node. This field is set lazily, and is used to make sure
-	// the `cpuset` cgroup hierarchy is created on cgroup v2 when cpumanager is using a
-	// None policy.
-	allCPUs cpuset.CPUSet
 }
 
 type features struct {

pkg/kubelet/cm/cpumanager/cpu_manager.go

@@ -94,6 +94,10 @@ type Manager interface {
 	// GetCPUAffinity returns cpuset which includes cpus from shared pools
 	// as well as exclusively allocated cpus
 	GetCPUAffinity(podUID, containerName string) cpuset.CPUSet
+
+	// GetAllCPUs returns all the CPUs known by cpumanager, as reported by the
+	// hardware discovery. Maps to the CPU capacity.
+	GetAllCPUs() cpuset.CPUSet
 }
 
 type manager struct {
@@ -137,7 +141,11 @@ type manager struct {
 	// stateFileDirectory holds the directory where the state file for checkpoints is held.
 	stateFileDirectory string
 
-	// allocatableCPUs is the set of online CPUs as reported by the system
+	// allCPUs is the set of online CPUs as reported by the system
+	allCPUs cpuset.CPUSet
+
+	// allocatableCPUs is the set of online CPUs as reported by the system,
+	// and available for allocation, minus the reserved set
 	allocatableCPUs cpuset.CPUSet
 
 	// pendingAdmissionPod contain the pod during the admission phase
@@ -157,6 +165,11 @@ func NewManager(cpuPolicyName string, cpuPolicyOptions map[string]string, reconc
 	var policy Policy
 	var err error
 
+	topo, err = topology.Discover(machineInfo)
+	if err != nil {
+		return nil, err
+	}
+
 	switch policyName(cpuPolicyName) {
 
 	case PolicyNone:
@@ -166,10 +179,6 @@ func NewManager(cpuPolicyName string, cpuPolicyOptions map[string]string, reconc
 		}
 
 	case PolicyStatic:
-		topo, err = topology.Discover(machineInfo)
-		if err != nil {
-			return nil, err
-		}
 		klog.InfoS("Detected CPU topology", "topology", topo)
 
 		reservedCPUs, ok := nodeAllocatableReservation[v1.ResourceCPU]
@@ -206,6 +215,7 @@ func NewManager(cpuPolicyName string, cpuPolicyOptions map[string]string, reconc
 		topology:                   topo,
 		nodeAllocatableReservation: nodeAllocatableReservation,
 		stateFileDirectory:         stateFileDirectory,
+		allCPUs:                    topo.CPUDetails.CPUs(),
 	}
 	manager.sourcesReady = &sourcesReadyStub{}
 	return manager, nil
@@ -340,6 +350,10 @@ func (m *manager) GetAllocatableCPUs() cpuset.CPUSet {
 	return m.allocatableCPUs.Clone()
 }
 
+func (m *manager) GetAllCPUs() cpuset.CPUSet {
+	return m.allCPUs.Clone()
+}
+
 type reconciledContainer struct {
 	podName       string
 	containerName string

pkg/kubelet/cm/cpumanager/cpu_manager_test.go

@@ -645,15 +645,8 @@ func TestCPUManagerGenerate(t *testing.T) {
 				if rawMgr.policy.Name() != testCase.expectedPolicy {
 					t.Errorf("Unexpected policy name. Have: %q wants %q", rawMgr.policy.Name(), testCase.expectedPolicy)
 				}
-				if rawMgr.policy.Name() == string(PolicyNone) {
-					if rawMgr.topology != nil {
-						t.Errorf("Expected topology to be nil for 'none' policy. Have: %q", rawMgr.topology)
-					}
-				}
-				if rawMgr.policy.Name() != string(PolicyNone) {
-					if rawMgr.topology == nil {
-						t.Errorf("Expected topology to be non-nil for policy '%v'. Have: %q", rawMgr.policy.Name(), rawMgr.topology)
-					}
+				if rawMgr.topology == nil {
+					t.Errorf("Expected topology to be non-nil for policy '%v'. Have: %q", rawMgr.policy.Name(), rawMgr.topology)
 				}
 			}
 		})

pkg/kubelet/cm/cpumanager/fake_cpu_manager.go

@@ -85,6 +85,11 @@ func (m *fakeManager) GetCPUAffinity(podUID, containerName string) cpuset.CPUSet
 	return cpuset.CPUSet{}
 }
 
+func (m *fakeManager) GetAllCPUs() cpuset.CPUSet {
+	klog.InfoS("GetAllCPUs")
+	return cpuset.CPUSet{}
+}
+
 // NewFakeManager creates empty/fake cpu manager
 func NewFakeManager() Manager {
 	return &fakeManager{

pkg/kubelet/cm/node_container_manager_linux.go

@@ -31,12 +31,9 @@ import (
 	utilfeature "k8s.io/apiserver/pkg/util/feature"
 	"k8s.io/klog/v2"
 	kubefeatures "k8s.io/kubernetes/pkg/features"
-	"k8s.io/kubernetes/pkg/kubelet/cm/cpumanager"
-	"k8s.io/kubernetes/pkg/kubelet/cm/cpumanager/topology"
 	"k8s.io/kubernetes/pkg/kubelet/events"
 	"k8s.io/kubernetes/pkg/kubelet/stats/pidlimit"
 	kubetypes "k8s.io/kubernetes/pkg/kubelet/types"
-	"k8s.io/utils/cpuset"
 )
 
 const (
@@ -199,30 +196,14 @@ func (cm *containerManagerImpl) getCgroupConfig(rl v1.ResourceList) *ResourceCon
 	// An alternative is to delegate the `cpuset` cgroup to the kubelet, but that would require some plumbing in libcontainer,
 	// and this is sufficient.
 	// Only do so on None policy, as Static policy will do its own updating of the cpuset.
-	if cm.NodeConfig.CPUManagerPolicy == string(cpumanager.PolicyNone) {
-		if cm.allCPUs.IsEmpty() {
-			cm.allCPUs = cm.getAllCPUs()
-		}
-		rc.CPUSet = cm.allCPUs
+	// Please see the comment on policy none's GetAllocatableCPUs
+	if cm.cpuManager.GetAllocatableCPUs().IsEmpty() {
+		rc.CPUSet = cm.cpuManager.GetAllCPUs()
 	}
 
 	return &rc
 }
 
-func (cm *containerManagerImpl) getAllCPUs() cpuset.CPUSet {
-	machineInfo, err := cm.cadvisorInterface.MachineInfo()
-	if err != nil {
-		klog.V(4).InfoS("Failed to get machine info to get default cpuset", "error", err)
-		return cpuset.CPUSet{}
-	}
-	topo, err := topology.Discover(machineInfo)
-	if err != nil {
-		klog.V(4).InfoS("Failed to get topology info to get default cpuset", "error", err)
-		return cpuset.CPUSet{}
-	}
-	return topo.CPUDetails.CPUs()
-}
-
 // GetNodeAllocatableAbsolute returns the absolute value of Node Allocatable which is primarily useful for enforcement.
 // Note that not all resources that are available on the node are included in the returned list of resources.
 // Returns a ResourceList.