integ/kubernetes/kubernetes-1.25.3/debian/deb_folder/patches/kubelet-cpumanager-infra-pods-use-system-reserved-CP.patch

From 17a648868f4e0c09d7dda5b157dc2286ff20383c Mon Sep 17 00:00:00 2001
From: Ramesh Kumar Sivanandam <rameshkumar.sivanandam@windriver.com>
Date: Mon, 7 Nov 2022 09:48:01 -0500
Subject: [PATCH] kubelet cpumanager infra pods use system reserved CPUs

This assigns system infrastructure pods to the "reserved" cpuset
to isolate them from the shared pool of CPUs.

Infrastructure pods include any pods that belong to the kube-system,
armada, cert-manager, vault, platform-deployment-manager, portieris,
notification, flux-helm or metrics-server namespaces.

The implementation is a bit simplistic, it is assumed that the
"reserved" cpuset is large enough to handle all infrastructure pods
CPU allocations.

This also prevents infrastucture pods from using Guaranteed resources.

Co-authored-by: Jim Gauld <james.gauld@windriver.com>
Signed-off-by: Gleb Aronsky <gleb.aronsky@windriver.com>
Signed-off-by: Thiago Miranda <ThiagoOliveira.Miranda@windriver.com>
Signed-off-by: Kaustubh Dhokte <kaustubh.dhokte@windriver.com>
Signed-off-by: Ramesh Kumar Sivanandam <rameshkumar.sivanandam@windriver.com>
---
 pkg/kubelet/cm/cpumanager/policy_static.go    | 38 +++++++++++++++++++
 .../cm/cpumanager/policy_static_test.go       | 19 +++++++++-
 2 files changed, 56 insertions(+), 1 deletion(-)

diff --git a/pkg/kubelet/cm/cpumanager/policy_static.go b/pkg/kubelet/cm/cpumanager/policy_static.go
index 14a6e7af9be..341e9f3dffe 100644
--- a/pkg/kubelet/cm/cpumanager/policy_static.go
+++ b/pkg/kubelet/cm/cpumanager/policy_static.go
@@ -53,6 +53,11 @@ func (e SMTAlignmentError) Type() string {
 	return ErrorSMTAlignment
 }
 
+// Define namespaces used by platform infrastructure pods
+var infraNamespaces = [...]string{
+	"kube-system", "armada", "cert-manager", "platform-deployment-manager", "portieris", "vault", "notification", "flux-helm", "metrics-server",
+}
+
 // staticPolicy is a CPU manager policy that does not change CPU
 // assignments for exclusively pinned guaranteed containers after the main
 // container process starts.
@@ -268,6 +273,25 @@ func (p *staticPolicy) updateCPUsToReuse(pod *v1.Pod, container *v1.Container, c
 }
 
 func (p *staticPolicy) Allocate(s state.State, pod *v1.Pod, container *v1.Container) error {
+	// Process infra pods before guaranteed pods
+	if isKubeInfra(pod) {
+		// Container belongs in reserved pool.
+		// We don't want to fall through to the p.guaranteedCPUs() clause below so return either nil or error.
+		if _, ok := s.GetCPUSet(string(pod.UID), container.Name); ok {
+			klog.Infof("[cpumanager] static policy: reserved container already present in state, skipping (namespace: %s, pod UID: %s, pod: %s, container: %s)", pod.Namespace, string(pod.UID), pod.Name, container.Name)
+			return nil
+		}
+
+		cpuset := p.reserved
+		if cpuset.IsEmpty() {
+			// If this happens then someone messed up.
+			return fmt.Errorf("[cpumanager] static policy: reserved container unable to allocate cpus (namespace: %s, pod UID: %s, pod: %s, container: %s); cpuset=%v, reserved:%v", pod.Namespace, string(pod.UID), pod.Name, container.Name, cpuset, p.reserved)
+		}
+		s.SetCPUSet(string(pod.UID), container.Name, cpuset)
+		klog.Infof("[cpumanager] static policy: reserved: AddContainer (namespace: %s, pod UID: %s, pod: %s, container: %s); cpuset=%v", pod.Namespace, string(pod.UID), pod.Name, container.Name, cpuset)
+		return nil
+	}
+
 	if numCPUs := p.guaranteedCPUs(pod, container); numCPUs != 0 {
 		klog.InfoS("Static policy: Allocate", "pod", klog.KObj(pod), "containerName", container.Name)
 		// container belongs in an exclusively allocated pool
@@ -384,6 +408,10 @@ func (p *staticPolicy) guaranteedCPUs(pod *v1.Pod, container *v1.Container) int
 	if cpuQuantity.Value()*1000 != cpuQuantity.MilliValue() {
 		return 0
 	}
+	// Infrastructure pods use reserved CPUs even if they're in the Guaranteed QoS class
+	if isKubeInfra(pod) {
+		return 0
+	}
 	// Safe downcast to do for all systems with < 2.1 billion CPUs.
 	// Per the language spec, `int` is guaranteed to be at least 32 bits wide.
 	// https://golang.org/ref/spec#Numeric_types
@@ -602,6 +630,16 @@ func (p *staticPolicy) generateCPUTopologyHints(availableCPUs cpuset.CPUSet, reu
 	return hints
 }
 
+// check if a given pod is in a platform infrastructure namespace
+func isKubeInfra(pod *v1.Pod) bool {
+	for _, namespace := range infraNamespaces {
+		if namespace == pod.Namespace {
+			return true
+		}
+	}
+	return false
+}
+
 // isHintSocketAligned function return true if numa nodes in hint are socket aligned.
 func (p *staticPolicy) isHintSocketAligned(hint topologymanager.TopologyHint, minAffinitySize int) bool {
 	numaNodesBitMask := hint.NUMANodeAffinity.GetBits()
diff --git a/pkg/kubelet/cm/cpumanager/policy_static_test.go b/pkg/kubelet/cm/cpumanager/policy_static_test.go
index 80a0c5a9e70..414e5ce144c 100644
--- a/pkg/kubelet/cm/cpumanager/policy_static_test.go
+++ b/pkg/kubelet/cm/cpumanager/policy_static_test.go
@@ -939,7 +939,8 @@ func TestStaticPolicyStartWithResvList(t *testing.T) {
 }
 
 func TestStaticPolicyAddWithResvList(t *testing.T) {
-
+	infraPod := makePod("fakePod", "fakeContainer2", "200m", "200m")
+	infraPod.Namespace = "kube-system"
 	testCases := []staticPolicyTestWithResvList{
 		{
 			description:     "GuPodSingleCore, SingleSocketHT, ExpectError",
@@ -981,6 +982,22 @@ func TestStaticPolicyAddWithResvList(t *testing.T) {
 			expCPUAlloc:     true,
 			expCSet:         cpuset.NewCPUSet(4, 5),
 		},
+		{
+			description:     "InfraPod, SingleSocketHT, ExpectAllocReserved",
+			topo:            topoSingleSocketHT,
+			numReservedCPUs: 2,
+			reserved:        cpuset.NewCPUSet(0, 1),
+			stAssignments: state.ContainerCPUAssignments{
+				"fakePod": map[string]cpuset.CPUSet{
+					"fakeContainer100": cpuset.NewCPUSet(2, 3, 6, 7),
+				},
+			},
+			stDefaultCPUSet: cpuset.NewCPUSet(4, 5),
+			pod:             infraPod,
+			expErr:          nil,
+			expCPUAlloc:     true,
+			expCSet:         cpuset.NewCPUSet(0, 1),
+		},
 	}
 
 	testExcl := true
-- 
2.25.1