fix code about watermark of cpu/mem usage percent

kaiyuechen · kaiyuechen · commit 042d169545fa · 2022-10-13T20:06:42.000+08:00
diff --git a/examples/ensurance/evict-on-cpu-usage-percent/elastic-pod-qos.yaml b/examples/ensurance/evict-on-cpu-usage-percent/elastic-pod-qos.yaml
@@ -5,9 +5,6 @@ metadata:
 spec:
   allowedActions:
     - eviction
-  resourceQOS:
-    cpuQOS:
-      cpuPriority: 7
   labelSelector:
     matchLabels:
       preemptible_job: "true"
diff --git a/examples/ensurance/evict-on-cpu-usage-percent/pod.yaml b/examples/ensurance/evict-on-cpu-usage-percent/pod.yaml
@@ -8,12 +8,12 @@ metadata:
 spec:
   containers:
     - command:
-        - stress-nga
+        - stress-ng
         - -c
         - "2"
         - --cpu-method
         - cpuid
-      image: ccr.ccs.tencentyun.com/tkeimages/stress-ng:v0.12.09.ln
+      image: docker.io/gocrane/stress-ng:v0.12.09
       imagePullPolicy: IfNotPresent
       name: low
       resources:
diff --git a/examples/ensurance/evict-on-cpu-usage-total/be-rules.yaml b/examples/ensurance/evict-on-cpu-usage-total/be-rules.yaml
@@ -5,11 +5,6 @@ metadata:
 spec:
   allowedActions:
     - eviction
-  resourceQOS:
-    cpuQOS:
-      cpuPriority: 7
-      htIsolation:
-        enable: false
   scopeSelector:
     matchExpressions:
       - operator: In
diff --git a/examples/ensurance/evict-on-mem-usage-percent/elastic-pod-qos.yaml b/examples/ensurance/evict-on-mem-usage-percent/elastic-pod-qos.yaml
@@ -5,9 +5,6 @@ metadata:
 spec:
   allowedActions:
     - eviction
-  resourceQOS:
-    cpuQOS:
-      cpuPriority: 7
   labelSelector:
     matchLabels:
       preemptible_job: "true"
diff --git a/examples/ensurance/evict-on-mem-usage-percent/pod.yaml b/examples/ensurance/evict-on-mem-usage-percent/pod.yaml
@@ -1,13 +1,13 @@
 apiVersion: v1
 kind: Pod
 metadata:
-  name: low-pi6
+  name: low-pi1
   labels:
     k8s-app: low
     preemptible_job: "true"
 spec:
   containers:
-    - image: ccr.ccs.tencentyun.com/tkeimages/stress-ng:v0.12.09
+    - image: docker.io/gocrane/stress-ng:v0.12.09
       command:
         - stress-ng
         - --vm-hang
@@ -17,14 +17,6 @@ spec:
         - --vm-bytes
         - "2G"
       name: stress
-      volumeMounts:
-        - mountPath: /data
-          name: data
-  volumes:
-    - hostPath:
-        path: /data/dd
-        type: DirectoryOrCreate
-      name: data
 ---
 apiVersion: v1
 kind: Pod
@@ -35,7 +27,7 @@ metadata:
     preemptible_job: "true"
 spec:
   containers:
-    - image: ccr.ccs.tencentyun.com/tkeimages/stress-ng:v0.12.09
+    - image: docker.io/gocrane/stress-ng:v0.12.09
       command:
         - stress-ng
         - --vm-hang
@@ -45,12 +37,4 @@ spec:
         - --vm-bytes
         - "3.5G"
       name: stress
-      volumeMounts:
-        - mountPath: /data
-          name: data
-  volumes:
-    - hostPath:
-        path: /data/dd
-        type: DirectoryOrCreate
-      name: data
 
diff --git a/examples/ensurance/evict-on-mem-usage-percent/watermark.yaml b/examples/ensurance/evict-on-mem-usage-percent/watermark.yaml
@@ -13,6 +13,6 @@ spec:
       metricRule:
         name: memory_total_utilization
         value: 50
-      name: cpu-usage-percent
+      name: mem-usage-percent
       restoreThreshold: 2
       strategy: None
diff --git a/examples/ensurance/evict-on-mem-usage-total/elastic-pod-qos.yaml b/examples/ensurance/evict-on-mem-usage-total/elastic-pod-qos.yaml
@@ -5,9 +5,6 @@ metadata:
 spec:
   allowedActions:
     - eviction
-  resourceQOS:
-    cpuQOS:
-      cpuPriority: 7
   labelSelector:
     matchLabels:
       preemptible_job: "true"
diff --git a/examples/ensurance/evict-on-mem-usage-total/pod.yaml b/examples/ensurance/evict-on-mem-usage-total/pod.yaml
@@ -1,13 +1,13 @@
 apiVersion: v1
 kind: Pod
 metadata:
-  name: low-pi6
+  name: low-pi1
   labels:
     k8s-app: low
     preemptible_job: "true"
 spec:
   containers:
-    - image: ccr.ccs.tencentyun.com/tkeimages/stress-ng:v0.12.09
+    - image: docker.io/gocrane/stress-ng:v0.12.09
       command:
         - stress-ng
         - --vm-hang
@@ -17,14 +17,6 @@ spec:
         - --vm-bytes
         - "2G"
       name: stress
-      volumeMounts:
-        - mountPath: /data
-          name: data
-  volumes:
-    - hostPath:
-        path: /data/dd
-        type: DirectoryOrCreate
-      name: data
 ---
 apiVersion: v1
 kind: Pod
@@ -35,7 +27,7 @@ metadata:
     preemptible_job: "true"
 spec:
   containers:
-    - image: ccr.ccs.tencentyun.com/tkeimages/stress-ng:v0.12.09
+    - image: docker.io/gocrane/stress-ng:v0.12.09
       command:
         - stress-ng
         - --vm-hang
@@ -45,12 +37,4 @@ spec:
         - --vm-bytes
         - "3.5G"
       name: stress
-      volumeMounts:
-        - mountPath: /data
-          name: data
-  volumes:
-    - hostPath:
-        path: /data/dd
-        type: DirectoryOrCreate
-      name: data
 
diff --git a/examples/ensurance/evict-on-mem-usage-total/watermark.yaml b/examples/ensurance/evict-on-mem-usage-total/watermark.yaml
@@ -12,7 +12,7 @@ spec:
       avoidanceThreshold: 2
       metricRule:
         name: memory_total_usage
-        value: 5000000000
-      name: cpu-usage-percent
+        value: 5000000000 # means 5Gi
+      name: mem-usage
       restoreThreshold: 2
       strategy: None
diff --git a/pkg/ensurance/collector/cadvisor/cadvisor_linux.go b/pkg/ensurance/collector/cadvisor/cadvisor_linux.go
@@ -162,7 +162,8 @@ func (c *CadvisorCollector) Collect() (map[string][]common.TimeSeries, error) {
 				extResMemUse += float64(v.Stats[0].Memory.WorkingSet)
 			}
 
-			var containerLabels = GetContainerLabels(pod, containerId, containerName, hasExtCpuRes)
+			hasExtRes := hasExtCpuRes || hasExtMemRes
+			var containerLabels = GetContainerLabels(pod, containerId, containerName, hasExtRes)
 			addSampleToStateMap(types.MetricNameContainerMemTotalUsage, composeSample(containerLabels, float64(v.Stats[0].Memory.WorkingSet), now), stateMap)
 			klog.V(6).Infof("Pod: %s, containerName: %s, key %s, container_mem_total_usage %#v", klog.KObj(pod), containerName, key, float64(v.Stats[0].Memory.WorkingSet))
 
diff --git a/pkg/ensurance/executor/sort/mem_usage_sort_test.go b/pkg/ensurance/executor/sort/mem_usage_sort_test.go
@@ -0,0 +1,82 @@
+package sort
+
+import (
+	"testing"
+	"time"
+
+	v1 "k8s.io/api/core/v1"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+	"k8s.io/apimachinery/pkg/types"
+
+	"github.com/gocrane/crane/pkg/ensurance/executor/podinfo"
+)
+
+func TestMemUsageSorter(t *testing.T) {
+	now := metav1.NewTime(time.Unix(1000, 0).UTC())
+	later := metav1.NewTime(time.Unix(2000, 0).UTC())
+
+	pods := []podinfo.PodContext{
+		{
+			Key:             types.NamespacedName{Name: "elastic-mem-2"},
+			ElasticMemLimit: 2,
+			QOSClass:        v1.PodQOSBestEffort,
+		},
+		{
+			Key:             types.NamespacedName{Name: "elastic-mem-4"},
+			ElasticMemLimit: 4,
+			QOSClass:        v1.PodQOSBestEffort,
+		},
+		{
+			Key:         types.NamespacedName{Name: "mem-1"},
+			PodMemUsage: 1,
+			QOSClass:    v1.PodQOSGuaranteed,
+		},
+		{
+			Key:         types.NamespacedName{Name: "mem-2"},
+			PodMemUsage: 2,
+			QOSClass:    v1.PodQOSBurstable,
+		},
+		{
+			Key:         types.NamespacedName{Name: "guarantee-1"},
+			PodMemUsage: 1,
+			QOSClass:    v1.PodQOSGuaranteed,
+		},
+		{
+			Key:         types.NamespacedName{Name: "burstable-2"},
+			PodMemUsage: 1,
+			QOSClass:    v1.PodQOSBurstable,
+		},
+		{
+			Key:         types.NamespacedName{Name: "prioirty-2"},
+			Priority:    2,
+			PodMemUsage: 1,
+			QOSClass:    v1.PodQOSBurstable,
+		},
+		{
+			Key:         types.NamespacedName{Name: "prioirty-2-2"},
+			Priority:    2,
+			PodMemUsage: 2,
+			QOSClass:    v1.PodQOSBurstable,
+		},
+		{
+			Key:      types.NamespacedName{Name: "priority-1"},
+			Priority: 1,
+			QOSClass: v1.PodQOSBurstable,
+		},
+		{
+			Key:       types.NamespacedName{Name: "time-1"},
+			StartTime: &now,
+			QOSClass:  v1.PodQOSGuaranteed,
+		},
+		{
+			Key:       types.NamespacedName{Name: "time-2"},
+			StartTime: &later,
+			QOSClass:  v1.PodQOSGuaranteed,
+		},
+	}
+	MemUsageSort(pods)
+	t.Logf("sorted pods:")
+	for _, p := range pods {
+		t.Logf("key %s, useElasticMem %v, qosClass %s, priority %d, usage %f, elasticMemUsage %d, startTime %v", p.Key, (p.ElasticMemLimit != 0), p.QOSClass, p.Priority, p.PodMemUsage, p.ElasticMemLimit, p.StartTime)
+	}
+}
diff --git a/pkg/utils/pod.go b/pkg/utils/pod.go
@@ -205,8 +205,8 @@ func GetContainerNameFromPod(pod *v1.Pod, containerId string) string {
 		if len(strList) > 0 {
 			klog.V(6).Infof("cri-containerd is %s ", "cri-containerd-"+strList[len(strList)-1]+".scope")
 			klog.V(6).Infof("containerid is %s", containerId)
-			if "cri-containerd-"+strList[len(strList)-1]+".scope" == containerId {
-				klog.V(6).Infof("111111111")
+			containerIdFromPod := fmt.Sprintf("cri-containerd-%s.scope", strList[len(strList)-1])
+			if containerIdFromPod == containerId {
 				return v.Name
 			}
 		}

Original file line number	Diff line number	Diff line change
`@@ -162,7 +162,8 @@ func (c *CadvisorCollector) Collect() (map[string][]common.TimeSeries, error) {`
`162`	`162`	`extResMemUse += float64(v.Stats[0].Memory.WorkingSet)`
`163`	`163`	`}`
`164`	`164`
`165`		`- var containerLabels = GetContainerLabels(pod, containerId, containerName, hasExtCpuRes)`
	`165`	`+ hasExtRes := hasExtCpuRes \|\| hasExtMemRes`
	`166`	`+ var containerLabels = GetContainerLabels(pod, containerId, containerName, hasExtRes)`
`166`	`167`	`addSampleToStateMap(types.MetricNameContainerMemTotalUsage, composeSample(containerLabels, float64(v.Stats[0].Memory.WorkingSet), now), stateMap)`
`167`	`168`	`klog.V(6).Infof("Pod: %s, containerName: %s, key %s, container_mem_total_usage %#v", klog.KObj(pod), containerName, key, float64(v.Stats[0].Memory.WorkingSet))`
`168`	`169`