Filter terminated pods from node request metrics. (amazon-contributin…

…g#104)
lisguo · Oct 20, 2023 · 328502e · 328502e
1 parent 546d245
commit 328502e
Show file tree

Hide file tree

Showing 2 changed files with 60 additions and 4 deletions.
diff --git a/receiver/awscontainerinsightreceiver/internal/stores/podstore.go b/receiver/awscontainerinsightreceiver/internal/stores/podstore.go
@@ -303,10 +303,13 @@ func (p *PodStore) refreshInternal(now time.Time, podList []corev1.Pod) {
 			p.logger.Warn(fmt.Sprintf("podKey is unavailable, refresh pod store for pod %s", pod.Name))
 			continue
 		}
-		tmpCPUReq, _ := getResourceSettingForPod(&pod, p.nodeInfo.getCPUCapacity(), cpuKey, getRequestForContainer)
-		cpuRequest += tmpCPUReq
-		tmpMemReq, _ := getResourceSettingForPod(&pod, p.nodeInfo.getMemCapacity(), memoryKey, getRequestForContainer)
-		memRequest += tmpMemReq
+		// filter out terminated pods
+		if pod.Status.Phase != corev1.PodSucceeded && pod.Status.Phase != corev1.PodFailed {
+			tmpCPUReq, _ := getResourceSettingForPod(&pod, p.nodeInfo.getCPUCapacity(), cpuKey, getRequestForContainer)
+			cpuRequest += tmpCPUReq
+			tmpMemReq, _ := getResourceSettingForPod(&pod, p.nodeInfo.getMemCapacity(), memoryKey, getRequestForContainer)
+			memRequest += tmpMemReq
+		}
 		if pod.Status.Phase == corev1.PodRunning {
 			podCount++
 		}

diff --git a/receiver/awscontainerinsightreceiver/internal/stores/podstore_test.go b/receiver/awscontainerinsightreceiver/internal/stores/podstore_test.go
@@ -1141,6 +1141,59 @@ func TestPodStore_decorateNode(t *testing.T) {
 	assert.Equal(t, uint64(15), metric.GetField("node_status_allocatable_pods").(uint64))
 }
 
+func TestPodStore_decorateNode_multiplePodStates(t *testing.T) {
+	podStore := getPodStore()
+	defer require.NoError(t, podStore.Shutdown())
+
+	tags := map[string]string{ci.MetricType: ci.TypeNode}
+	fields := map[string]interface{}{
+		ci.MetricName(ci.TypeNode, ci.CPUTotal):      float64(100),
+		ci.MetricName(ci.TypeNode, ci.CPULimit):      uint64(4000),
+		ci.MetricName(ci.TypeNode, ci.MemWorkingset): float64(100 * 1024 * 1024),
+		ci.MetricName(ci.TypeNode, ci.MemLimit):      uint64(400 * 1024 * 1024),
+	}
+	metric := generateMetric(fields, tags)
+
+	// terminated pods should not contribute to requests
+	failedPod := generatePodInfo("./test_resources/pod_in_phase_failed.json")
+	succeededPod := generatePodInfo("./test_resources/pod_in_phase_succeeded.json")
+	podList := []corev1.Pod{*failedPod, *succeededPod}
+	podStore.refreshInternal(time.Now(), podList)
+	podStore.decorateNode(metric)
+
+	assert.Equal(t, uint64(0), metric.GetField("node_cpu_request").(uint64))
+	assert.Equal(t, uint64(4000), metric.GetField("node_cpu_limit").(uint64))
+	assert.Equal(t, float64(0), metric.GetField("node_cpu_reserved_capacity").(float64))
+	assert.Equal(t, float64(100), metric.GetField("node_cpu_usage_total").(float64))
+
+	assert.Equal(t, uint64(0), metric.GetField("node_memory_request").(uint64))
+	assert.Equal(t, uint64(400*1024*1024), metric.GetField("node_memory_limit").(uint64))
+	assert.Equal(t, float64(0), metric.GetField("node_memory_reserved_capacity").(float64))
+	assert.Equal(t, float64(100*1024*1024), metric.GetField("node_memory_working_set").(float64))
+
+	// non-terminated pods should contribute to requests
+	pendingPod := generatePodInfo("./test_resources/pod_in_phase_pending.json")
+	podList = append(podList, *pendingPod)
+	podStore.refreshInternal(time.Now(), podList)
+	podStore.decorateNode(metric)
+	assert.Equal(t, uint64(10), metric.GetField("node_cpu_request").(uint64))
+	assert.Equal(t, float64(0.25), metric.GetField("node_cpu_reserved_capacity").(float64))
+
+	assert.Equal(t, uint64(50*1024*1024), metric.GetField("node_memory_request").(uint64))
+	assert.Equal(t, float64(12.5), metric.GetField("node_memory_reserved_capacity").(float64))
+
+	runningPod := generatePodInfo("./test_resources/pod_in_phase_running.json")
+	podList = append(podList, *runningPod)
+	podStore.refreshInternal(time.Now(), podList)
+	podStore.decorateNode(metric)
+
+	assert.Equal(t, uint64(20), metric.GetField("node_cpu_request").(uint64))
+	assert.Equal(t, float64(0.5), metric.GetField("node_cpu_reserved_capacity").(float64))
+
+	assert.Equal(t, uint64(100*1024*1024), metric.GetField("node_memory_request").(uint64))
+	assert.Equal(t, float64(25), metric.GetField("node_memory_reserved_capacity").(float64))
+}
+
 func TestPodStore_Decorate(t *testing.T) {
 	// not the metrics for decoration
 	tags := map[string]string{}