Adding pod_status_unknown as a Pod condition metric and removed the p…

…od_status_initialized metric (#66) * Adding pod_status_unknown as a Pod condition metric and removed the pod_status_initialized metric
pxaws · Aug 1, 2023 · 10a8fba · 10a8fba
1 parent e032a05
commit 10a8fba
Show file tree

Hide file tree

Showing 5 changed files with 235 additions and 77 deletions.
diff --git a/internal/aws/containerinsight/const.go b/internal/aws/containerinsight/const.go
@@ -114,7 +114,6 @@ const (
 	StatusUnknown                     = "status_unknown"
 	StatusReady                       = "status_ready"
 	StatusScheduled                   = "status_scheduled"
-	StatusInitialized                 = "status_initialized"
 
 	RunningPodCount       = "number_of_running_pods"
 	RunningContainerCount = "number_of_running_containers"
@@ -258,6 +257,9 @@ func init() {
 		StatusTerminated:           UnitCount,
 		StatusWaiting:              UnitCount,
 		StatusWaitingReasonCrashed: UnitCount,
+		StatusUnknown:              UnitCount,
+		StatusReady:                UnitCount,
+		StatusScheduled:            UnitCount,
 
 		// cluster metrics
 		NodeCount:       UnitCount,

diff --git a/receiver/awscontainerinsightreceiver/README.md b/receiver/awscontainerinsightreceiver/README.md
@@ -633,45 +633,48 @@ kubectl apply -f config.yaml
 <br/><br/> 
 
 ### Pod
-| Metric                                | Unit          |
-|---------------------------------------|---------------|
-| pod_cpu_limit                         | Millicore     |
-| pod_cpu_request                       | Millicore     |
-| pod_cpu_reserved_capacity             | Percent       |
-| pod_cpu_usage_system                  | Millicore     |
-| pod_cpu_usage_total                   | Millicore     |
-| pod_cpu_usage_user                    | Millicore     |
-| pod_cpu_utilization                   | Percent       |
-| pod_cpu_utilization_over_pod_limit    | Percent       |
-| pod_memory_cache                      | Bytes         |
-| pod_memory_failcnt                    | Count         |
-| pod_memory_hierarchical_pgfault       | Count/Second  |
-| pod_memory_hierarchical_pgmajfault    | Count/Second  |
-| pod_memory_limit                      | Bytes         |
-| pod_memory_mapped_file                | Bytes         |
-| pod_memory_max_usage                  | Bytes         |
-| pod_memory_pgfault                    | Count/Second  |
-| pod_memory_pgmajfault                 | Count/Second  |
-| pod_memory_request                    | Bytes         |
-| pod_memory_reserved_capacity          | Percent       |
-| pod_memory_rss                        | Bytes         |
-| pod_memory_swap                       | Bytes         |
-| pod_memory_usage                      | Bytes         |
-| pod_memory_utilization                | Percent       |
-| pod_memory_utilization_over_pod_limit | Percent       |
-| pod_memory_working_set                | Bytes         |
-| pod_network_rx_bytes                  | Bytes/Second  |
-| pod_network_rx_dropped                | Count/Second  |
-| pod_network_rx_errors                 | Count/Second  |
-| pod_network_rx_packets                | Count/Second  |
-| pod_network_total_bytes               | Bytes/Second  |
-| pod_network_tx_bytes                  | Bytes/Second  |
-| pod_network_tx_dropped                | Count/Second  |
-| pod_network_tx_errors                 | Count/Second  |
-| pod_network_tx_packets                | Count/Second  |
-| pod_number_of_container_restarts      | Count         | 
-| pod_number_of_containers              | Count         |   
-| pod_number_of_running_containers      | Count         |  
+| Metric                                | Unit         |
+|---------------------------------------|--------------|
+| pod_cpu_limit                         | Millicore    |
+| pod_cpu_request                       | Millicore    |
+| pod_cpu_reserved_capacity             | Percent      |
+| pod_cpu_usage_system                  | Millicore    |
+| pod_cpu_usage_total                   | Millicore    |
+| pod_cpu_usage_user                    | Millicore    |
+| pod_cpu_utilization                   | Percent      |
+| pod_cpu_utilization_over_pod_limit    | Percent      |
+| pod_memory_cache                      | Bytes        |
+| pod_memory_failcnt                    | Count        |
+| pod_memory_hierarchical_pgfault       | Count/Second |
+| pod_memory_hierarchical_pgmajfault    | Count/Second |
+| pod_memory_limit                      | Bytes        |
+| pod_memory_mapped_file                | Bytes        |
+| pod_memory_max_usage                  | Bytes        |
+| pod_memory_pgfault                    | Count/Second |
+| pod_memory_pgmajfault                 | Count/Second |
+| pod_memory_request                    | Bytes        |
+| pod_memory_reserved_capacity          | Percent      |
+| pod_memory_rss                        | Bytes        |
+| pod_memory_swap                       | Bytes        |
+| pod_memory_usage                      | Bytes        |
+| pod_memory_utilization                | Percent      |
+| pod_memory_utilization_over_pod_limit | Percent      |
+| pod_memory_working_set                | Bytes        |
+| pod_network_rx_bytes                  | Bytes/Second |
+| pod_network_rx_dropped                | Count/Second |
+| pod_network_rx_errors                 | Count/Second |
+| pod_network_rx_packets                | Count/Second |
+| pod_network_total_bytes               | Bytes/Second |
+| pod_network_tx_bytes                  | Bytes/Second |
+| pod_network_tx_dropped                | Count/Second |
+| pod_network_tx_errors                 | Count/Second |
+| pod_network_tx_packets                | Count/Second |
+| pod_number_of_container_restarts      | Count        | 
+| pod_number_of_containers              | Count        |   
+| pod_number_of_running_containers      | Count        |  
+| pod_status_ready                      | Count        |
+| pod_status_scheduled                  | Count        |
+| pod_status_unknown                    | Count        |
 
 | Resource Attribute   |
 |----------------------|

diff --git a/receiver/awscontainerinsightreceiver/internal/stores/podstore.go b/receiver/awscontainerinsightreceiver/internal/stores/podstore.go
@@ -53,14 +53,14 @@ var (
 		corev1.PodRunning:   ci.MetricName(ci.TypePod, ci.StatusRunning),
 		corev1.PodSucceeded: ci.MetricName(ci.TypePod, ci.StatusSucceeded),
 		corev1.PodFailed:    ci.MetricName(ci.TypePod, ci.StatusFailed),
-		corev1.PodUnknown:   ci.MetricName(ci.TypePod, ci.StatusUnknown),
 	}
 
 	PodConditionMetricNames = map[corev1.PodConditionType]string{
-		corev1.PodReady:       ci.MetricName(ci.TypePod, ci.StatusReady),
-		corev1.PodScheduled:   ci.MetricName(ci.TypePod, ci.StatusScheduled),
-		corev1.PodInitialized: ci.MetricName(ci.TypePod, ci.StatusInitialized),
+		corev1.PodReady:     ci.MetricName(ci.TypePod, ci.StatusReady),
+		corev1.PodScheduled: ci.MetricName(ci.TypePod, ci.StatusScheduled),
 	}
+
+	PodConditionUnknownMetric = ci.MetricName(ci.TypePod, ci.StatusUnknown)
 )
 
 type cachedEntry struct {
@@ -570,16 +570,21 @@ func (p *PodStore) addPodConditionMetrics(metric CIMetric, pod *corev1.Pod) {
 		metric.AddField(metricName, 0)
 	}
 
+	metric.AddField(PodConditionUnknownMetric, 0)
+
 	for _, condition := range pod.Status.Conditions {
-		if condition.Status != corev1.ConditionTrue {
-			continue
-		}
 
-		conditionKey := condition.Type
-		statusMetricName, conditionKeyValid := PodConditionMetricNames[conditionKey]
-		if conditionKeyValid {
-			metric.AddField(statusMetricName, 1)
+		switch condition.Status {
+		case corev1.ConditionTrue:
+			if statusMetricName, ok := PodConditionMetricNames[condition.Type]; ok {
+				metric.AddField(statusMetricName, 1)
+			}
+		case corev1.ConditionUnknown:
+			if _, ok := PodConditionMetricNames[condition.Type]; ok {
+				metric.AddField(PodConditionUnknownMetric, 1)
+			}
 		}
+
 	}
 }
 

diff --git a/receiver/awscontainerinsightreceiver/internal/stores/podstore_test.go b/receiver/awscontainerinsightreceiver/internal/stores/podstore_test.go
@@ -339,14 +339,13 @@ func TestPodStore_addContainerCount(t *testing.T) {
 }
 
 const (
-	PodFailedMetricName      = "pod_status_failed"
-	PodPendingMetricName     = "pod_status_pending"
-	PodRunningMetricName     = "pod_status_running"
-	PodSucceededMetricName   = "pod_status_succeeded"
-	PodUnknownMetricName     = "pod_status_unknown"
-	PodReadyMetricName       = "pod_status_ready"
-	PodScheduledMetricName   = "pod_status_scheduled"
-	PodInitializedMetricName = "pod_status_initialized"
+	PodFailedMetricName    = "pod_status_failed"
+	PodPendingMetricName   = "pod_status_pending"
+	PodRunningMetricName   = "pod_status_running"
+	PodSucceededMetricName = "pod_status_succeeded"
+	PodUnknownMetricName   = "pod_status_unknown"
+	PodReadyMetricName     = "pod_status_ready"
+	PodScheduledMetricName = "pod_status_scheduled"
 )
 
 func TestPodStore_enhanced_metrics_disabled(t *testing.T) {
@@ -356,7 +355,6 @@ func TestPodStore_enhanced_metrics_disabled(t *testing.T) {
 	assert.False(t, decoratedResultMetric.HasField(PodPendingMetricName))
 	assert.False(t, decoratedResultMetric.HasField(PodRunningMetricName))
 	assert.False(t, decoratedResultMetric.HasField(PodSucceededMetricName))
-	assert.False(t, decoratedResultMetric.HasField(PodUnknownMetricName))
 }
 
 func TestPodStore_addStatus_adds_pod_failed_metric(t *testing.T) {
@@ -366,7 +364,6 @@ func TestPodStore_addStatus_adds_pod_failed_metric(t *testing.T) {
 	assert.Equal(t, 0, decoratedResultMetric.GetField(PodPendingMetricName))
 	assert.Equal(t, 0, decoratedResultMetric.GetField(PodRunningMetricName))
 	assert.Equal(t, 0, decoratedResultMetric.GetField(PodSucceededMetricName))
-	assert.Equal(t, 0, decoratedResultMetric.GetField(PodUnknownMetricName))
 }
 
 func TestPodStore_addStatus_adds_pod_pending_metric(t *testing.T) {
@@ -376,7 +373,6 @@ func TestPodStore_addStatus_adds_pod_pending_metric(t *testing.T) {
 	assert.Equal(t, 1, decoratedResultMetric.GetField(PodPendingMetricName))
 	assert.Equal(t, 0, decoratedResultMetric.GetField(PodRunningMetricName))
 	assert.Equal(t, 0, decoratedResultMetric.GetField(PodSucceededMetricName))
-	assert.Equal(t, 0, decoratedResultMetric.GetField(PodUnknownMetricName))
 }
 
 func TestPodStore_addStatus_adds_pod_running_metric(t *testing.T) {
@@ -386,7 +382,6 @@ func TestPodStore_addStatus_adds_pod_running_metric(t *testing.T) {
 	assert.Equal(t, 0, decoratedResultMetric.GetField(PodPendingMetricName))
 	assert.Equal(t, 1, decoratedResultMetric.GetField(PodRunningMetricName))
 	assert.Equal(t, 0, decoratedResultMetric.GetField(PodSucceededMetricName))
-	assert.Equal(t, 0, decoratedResultMetric.GetField(PodUnknownMetricName))
 }
 
 func TestPodStore_addStatus_adds_pod_succeeded_metric(t *testing.T) {
@@ -396,41 +391,38 @@ func TestPodStore_addStatus_adds_pod_succeeded_metric(t *testing.T) {
 	assert.Equal(t, 0, decoratedResultMetric.GetField(PodPendingMetricName))
 	assert.Equal(t, 0, decoratedResultMetric.GetField(PodRunningMetricName))
 	assert.Equal(t, 1, decoratedResultMetric.GetField(PodSucceededMetricName))
-	assert.Equal(t, 0, decoratedResultMetric.GetField(PodUnknownMetricName))
-}
-
-func TestPodStore_addStatus_adds_pod_unknown_metric(t *testing.T) {
-	decoratedResultMetric := runAddStatusToGetDecoratedCIMetric("./test_resources/pod_in_phase_unknown.json", true)
-
-	assert.Equal(t, 0, decoratedResultMetric.GetField(PodFailedMetricName))
-	assert.Equal(t, 0, decoratedResultMetric.GetField(PodPendingMetricName))
-	assert.Equal(t, 0, decoratedResultMetric.GetField(PodRunningMetricName))
-	assert.Equal(t, 0, decoratedResultMetric.GetField(PodSucceededMetricName))
-	assert.Equal(t, 1, decoratedResultMetric.GetField(PodUnknownMetricName))
 }
 
 func TestPodStore_addStatus_enhanced_metrics_disabled(t *testing.T) {
 	decoratedResultMetric := runAddStatusToGetDecoratedCIMetric("./test_resources/all_pod_conditions_valid.json", false)
 
-	assert.False(t, decoratedResultMetric.HasField(PodInitializedMetricName))
 	assert.False(t, decoratedResultMetric.HasField(PodReadyMetricName))
 	assert.False(t, decoratedResultMetric.HasField(PodScheduledMetricName))
+	assert.False(t, decoratedResultMetric.HasField(PodUnknownMetricName))
 }
 
 func TestPodStore_addStatus_adds_all_pod_conditions_as_metrics_when_true_false_unknown(t *testing.T) {
 	decoratedResultMetric := runAddStatusToGetDecoratedCIMetric("./test_resources/all_pod_conditions_valid.json", true)
 
-	assert.Equal(t, 1, decoratedResultMetric.GetField(PodInitializedMetricName))
 	assert.Equal(t, 0, decoratedResultMetric.GetField(PodReadyMetricName))
 	assert.Equal(t, 0, decoratedResultMetric.GetField(PodScheduledMetricName))
+	assert.Equal(t, 1, decoratedResultMetric.GetField(PodUnknownMetricName))
+}
+
+func TestPodStore_addStatus_adds_all_pod_conditions_as_metrics_when_Ready_Scheduled_Condition_Unknown(t *testing.T) {
+	decoratedResultMetric := runAddStatusToGetDecoratedCIMetric("./test_resources/pod_Ready_Scheduled_Condition_Unknown.json", true)
+
+	assert.Equal(t, 0, decoratedResultMetric.GetField(PodReadyMetricName))
+	assert.Equal(t, 0, decoratedResultMetric.GetField(PodScheduledMetricName))
+	assert.Equal(t, 1, decoratedResultMetric.GetField(PodUnknownMetricName))
 }
 
 func TestPodStore_addStatus_adds_all_pod_conditions_as_metrics_when_unexpected(t *testing.T) {
 	decoratedResultMetric := runAddStatusToGetDecoratedCIMetric("./test_resources/one_pod_condition_invalid.json", true)
 
-	assert.Equal(t, 0, decoratedResultMetric.GetField(PodInitializedMetricName))
 	assert.Equal(t, 1, decoratedResultMetric.GetField(PodReadyMetricName))
 	assert.Equal(t, 1, decoratedResultMetric.GetField(PodScheduledMetricName))
+	assert.Equal(t, 0, decoratedResultMetric.GetField(PodUnknownMetricName))
 }
 
 func TestPodStore_addStatus_enhanced_metrics(t *testing.T) {