koordlet: revise noderesource and podresource when device collector disabled

Signed-off-by: saintube <saintube@foxmail.com>

Author: saintube

pkg/koordlet/metricsadvisor/collectors/noderesource/node_resource_collector.go

@@ -126,8 +126,15 @@ func (n *nodeResourceCollector) collectNodeResUsed() {
 	}
 	nodeMetrics = append(nodeMetrics, cpuUsageMetrics)
 
-	for _, deviceCollector := range n.deviceCollectors {
-		if metric, _ := deviceCollector.GetNodeMetric(); metric != nil {
+	for name, deviceCollector := range n.deviceCollectors {
+		if !deviceCollector.Enabled() {
+			klog.V(6).Infof("skip node metrics from the disabled device collector %s", name)
+			continue
+		}
+
+		if metric, err := deviceCollector.GetNodeMetric(); err != nil {
+			klog.Warningf("get node metrics from the device collector %s failed, err: %s", name, err)
+		} else {
 			nodeMetrics = append(nodeMetrics, metric...)
 		}
 		if info := deviceCollector.Infos(); info != nil {

pkg/koordlet/metricsadvisor/collectors/noderesource/node_resource_collector_test.go

@@ -134,14 +134,15 @@ DirectMap1G:           0 kB`)
 		CPUTick:   0,
 		Timestamp: testLastCPUStatTime,
 	}
+	testDeviceCollector := &fakeDeviceCollector{isEnabled: true}
 
 	c := &nodeResourceCollector{
 		started:         atomic.NewBool(false),
 		appendableDB:    metricCache,
 		metricDB:        metricCache,
 		lastNodeCPUStat: testLastCPUStat,
 		deviceCollectors: map[string]framework.DeviceCollector{
-			"TestDeviceCollector": &fakeDeviceCollector{},
+			"TestDeviceCollector": testDeviceCollector,
 		},
 		sharedState: framework.NewSharedState(),
 	}
@@ -167,6 +168,24 @@ DirectMap1G:           0 kB`)
 	assert.Equal(t, wantCPU, nodeCPU.Value)
 	assert.Equal(t, wantMemory, nodeMemory.Value)
 
+	// test collect without device collector
+	testDeviceCollector.isEnabled = false
+	testDeviceCollector.getNodeMetric = func() ([]metriccache.MetricSample, error) {
+		panic("should not be called")
+	}
+	assert.NotPanics(t, func() {
+		c.collectNodeResUsed()
+	})
+	assert.True(t, c.Started())
+	// validate collected values
+	// assert collected time is less than 10s
+	got, err = testGetNodeMetrics(t, c.metricDB, testNow, 5*time.Second)
+	wantCPU = testCPUUsage
+	assert.Equal(t, wantCPU, float64(got.Cpu().MilliValue()/1000))
+	// MemTotal - MemAvailable
+	wantMemory = float64(524288 * 1024)
+	assert.Equal(t, wantMemory, float64(got.Memory().Value()))
+
 	// test first cpu collection
 	c.lastNodeCPUStat = nil
 	assert.NotPanics(t, func() {
@@ -185,9 +204,18 @@ DirectMap1G:           0 kB`)
 
 type fakeDeviceCollector struct {
 	framework.DeviceCollector
+	isEnabled     bool
+	getNodeMetric func() ([]metriccache.MetricSample, error)
+}
+
+func (f *fakeDeviceCollector) Enabled() bool {
+	return f.isEnabled
 }
 
 func (f *fakeDeviceCollector) GetNodeMetric() ([]metriccache.MetricSample, error) {
+	if f.getNodeMetric != nil {
+		return f.getNodeMetric()
+	}
 	return nil, nil
 }
 

pkg/koordlet/metricsadvisor/collectors/podresource/pod_resource_collector.go

@@ -164,6 +164,11 @@ func (p *podResourceCollector) collectPodResUsed() {
 
 		metrics = append(metrics, cpuUsageMetric, memUsageMetric)
 		for deviceName, deviceCollector := range p.deviceCollectors {
+			if !deviceCollector.Enabled() {
+				klog.V(6).Infof("skip pod metrics from the disabled device collector %s, pod %s", deviceName, podKey)
+				continue
+			}
+
 			if deviceMetrics, err := deviceCollector.GetPodMetric(uid, meta.CgroupDir, pod.Status.ContainerStatuses); err != nil {
 				klog.V(4).Infof("get pod %s device usage failed for %v, error: %v", podKey, deviceName, err)
 			} else if len(metrics) > 0 {
@@ -262,6 +267,11 @@ func (p *podResourceCollector) collectContainerResUsed(meta *statesinformer.PodM
 		containerMetrics = append(containerMetrics, cpuUsageMetric, memUsageMetric)
 
 		for deviceName, deviceCollector := range p.deviceCollectors {
+			if !deviceCollector.Enabled() {
+				klog.V(6).Infof("skip container metrics from the disabled device collector %s, container %s", deviceName, containerKey)
+				continue
+			}
+
 			if metrics, err := deviceCollector.GetContainerMetric(containerStat.ContainerID, meta.CgroupDir, containerStat); err != nil {
 				klog.Warningf("get container %s device usage failed for %v, error: %v", containerKey, deviceName, err)
 			} else {

pkg/koordlet/metricsadvisor/collectors/podresource/pod_resource_collector_test.go

@@ -78,6 +78,7 @@ func Test_collector_collectPodResUsed(t *testing.T) {
 	}
 	type fields struct {
 		podFilterOption       framework.PodFilter
+		deviceCollectors      map[string]framework.DeviceCollector
 		getPodMetas           []*statesinformer.PodMeta
 		initPodLastStat       func(lastState *gocache.Cache)
 		initContainerLastStat func(lastState *gocache.Cache)
@@ -96,6 +97,9 @@ func Test_collector_collectPodResUsed(t *testing.T) {
 			name: "cgroups v1",
 			fields: fields{
 				podFilterOption: framework.DefaultPodFilter,
+				deviceCollectors: map[string]framework.DeviceCollector{
+					"TestDeviceCollector": &fakeDeviceCollector{isEnabled: true},
+				},
 				getPodMetas: []*statesinformer.PodMeta{
 					{
 						CgroupDir: testPodMetaDir,
@@ -150,6 +154,9 @@ total_unevictable 0
 			name: "cgroups v2",
 			fields: fields{
 				podFilterOption: framework.DefaultPodFilter,
+				deviceCollectors: map[string]framework.DeviceCollector{
+					"TestDeviceCollector": &fakeDeviceCollector{isEnabled: true},
+				},
 				getPodMetas: []*statesinformer.PodMeta{
 					{
 						CgroupDir: testPodMetaDir,
@@ -212,9 +219,20 @@ unevictable 0
 			},
 		},
 		{
-			name: "cgroups v1, filter non-running pods",
+			name: "cgroups v1, filter non-running pods, skip disabled device collector",
 			fields: fields{
 				podFilterOption: &framework.TerminatedPodFilter{},
+				deviceCollectors: map[string]framework.DeviceCollector{
+					"TestDeviceCollector": &fakeDeviceCollector{
+						isEnabled: false,
+						getPodMetric: func(uid, podParentDir string, cs []corev1.ContainerStatus) ([]metriccache.MetricSample, error) {
+							panic("should not be called")
+						},
+						getContainerMetric: func(uid, podParentDir string, c *corev1.ContainerStatus) ([]metriccache.MetricSample, error) {
+							panic("should not be called")
+						},
+					},
+				},
 				getPodMetas: []*statesinformer.PodMeta{
 					{
 						CgroupDir: testPodMetaDir,
@@ -304,7 +322,8 @@ total_unevictable 0
 				},
 			})
 			collector.Setup(&framework.Context{
-				State: framework.NewSharedState(),
+				State:            framework.NewSharedState(),
+				DeviceCollectors: tt.fields.deviceCollectors,
 			})
 			c := collector.(*podResourceCollector)
 			tt.fields.initPodLastStat(c.lastPodCPUStat)
@@ -350,3 +369,28 @@ func Test_podResourceCollector_Run(t *testing.T) {
 		close(stopCh)
 	})
 }
+
+type fakeDeviceCollector struct {
+	framework.DeviceCollector
+	isEnabled          bool
+	getPodMetric       func(uid, podParentDir string, cs []corev1.ContainerStatus) ([]metriccache.MetricSample, error)
+	getContainerMetric func(uid, podParentDir string, c *corev1.ContainerStatus) ([]metriccache.MetricSample, error)
+}
+
+func (f *fakeDeviceCollector) Enabled() bool {
+	return f.isEnabled
+}
+
+func (f *fakeDeviceCollector) GetPodMetric(uid, podParentDir string, cs []corev1.ContainerStatus) ([]metriccache.MetricSample, error) {
+	if f.getPodMetric != nil {
+		return f.getPodMetric(uid, podParentDir, cs)
+	}
+	return nil, nil
+}
+
+func (f *fakeDeviceCollector) GetContainerMetric(containerID, podParentDir string, c *corev1.ContainerStatus) ([]metriccache.MetricSample, error) {
+	if f.getContainerMetric != nil {
+		return f.getContainerMetric(containerID, podParentDir, c)
+	}
+	return nil, nil
+}