Adding new metrics: pod_cpu_usage_total and pod_memory_working_set (#…

…1294)
aws · Aug 16, 2024 · 6f8ea66 · 6f8ea66
1 parent 05bdddc
commit 6f8ea66
Show file tree

Hide file tree

Showing 7 changed files with 11 additions and 3 deletions.
diff --git a/.github/workflows/integration-test.yml b/.github/workflows/integration-test.yml
@@ -1344,4 +1344,4 @@ jobs:
             else
               cd terraform/eks/addon/gpu
             fi
-            terraform destroy --auto-approve
+            terraform destroy --auto-approve
diff --git a/translator/tocwconfig/sampleConfig/emf_and_kubernetes_config.yaml b/translator/tocwconfig/sampleConfig/emf_and_kubernetes_config.yaml
@@ -134,6 +134,8 @@ exporters:
                 - pod_memory_request
                 - pod_memory_limit
                 - pod_cpu_limit
+                - pod_cpu_usage_total
+                - pod_memory_working_set
                 - pod_cpu_request
                 - pod_container_status_running
                 - pod_container_status_terminated

diff --git a/translator/tocwconfig/sampleConfig/emf_and_kubernetes_with_gpu_config.yaml b/translator/tocwconfig/sampleConfig/emf_and_kubernetes_with_gpu_config.yaml
@@ -135,6 +135,8 @@ exporters:
                 - pod_memory_limit
                 - pod_cpu_limit
                 - pod_cpu_request
+                - pod_cpu_usage_total
+                - pod_memory_working_set
                 - pod_container_status_running
                 - pod_container_status_terminated
                 - pod_container_status_waiting

diff --git a/translator/tocwconfig/sampleConfig/kubernetes_on_prem_config.yaml b/translator/tocwconfig/sampleConfig/kubernetes_on_prem_config.yaml
@@ -99,6 +99,8 @@ exporters:
                 - pod_status_unknown
                 - pod_status_succeeded
                 - pod_memory_request
+                - pod_cpu_usage_total
+                - pod_memory_working_set
                 - pod_memory_limit
                 - pod_cpu_limit
                 - pod_cpu_request

diff --git a/translator/tocwconfig/sampleConfig/logs_and_kubernetes_config.yaml b/translator/tocwconfig/sampleConfig/logs_and_kubernetes_config.yaml
@@ -133,6 +133,8 @@ exporters:
                 - pod_memory_limit
                 - pod_cpu_limit
                 - pod_cpu_request
+                - pod_cpu_usage_total
+                - pod_memory_working_set
                 - pod_container_status_running
                 - pod_container_status_terminated
                 - pod_container_status_waiting

diff --git a/translator/translate/otel/exporter/awsemf/kubernetes.go b/translator/translate/otel/exporter/awsemf/kubernetes.go
@@ -99,7 +99,7 @@ func getPodMetricDeclarations(conf *confmap.Conf) []*awsemfexporter.MetricDeclar
 		podMetricDeclarations[0].Dimensions = append(podMetricDeclarations[0].Dimensions, []string{"FullPodName", "PodName", "Namespace", "ClusterName"})
 		selectors = append(selectors, []string{"pod_number_of_container_restarts", "pod_number_of_containers", "pod_number_of_running_containers",
 			"pod_status_ready", "pod_status_scheduled", "pod_status_running", "pod_status_pending", "pod_status_failed", "pod_status_unknown",
-			"pod_status_succeeded", "pod_memory_request", "pod_memory_limit", "pod_cpu_limit", "pod_cpu_request",
+			"pod_status_succeeded", "pod_memory_request", "pod_memory_limit", "pod_cpu_limit", "pod_cpu_request", "pod_cpu_usage_total", "pod_memory_working_set",
 			"pod_container_status_running", "pod_container_status_terminated", "pod_container_status_waiting", "pod_container_status_waiting_reason_crash_loop_back_off",
 			"pod_container_status_waiting_reason_image_pull_error", "pod_container_status_waiting_reason_start_error", "pod_container_status_waiting_reason_create_container_error",
 			"pod_container_status_waiting_reason_create_container_config_error", "pod_container_status_terminated_reason_oom_killed",

diff --git a/translator/translate/otel/exporter/awsemf/translator_test.go b/translator/translate/otel/exporter/awsemf/translator_test.go
@@ -306,7 +306,7 @@ func TestTranslator(t *testing.T) {
 						Dimensions: [][]string{{"PodName", "Namespace", "ClusterName"}, {"ClusterName"}, {"FullPodName", "PodName", "Namespace", "ClusterName"}, {"Service", "Namespace", "ClusterName"}},
 						MetricNameSelectors: []string{"pod_cpu_reserved_capacity", "pod_memory_reserved_capacity", "pod_number_of_container_restarts", "pod_number_of_containers", "pod_number_of_running_containers",
 							"pod_status_ready", "pod_status_scheduled", "pod_status_running", "pod_status_pending", "pod_status_failed", "pod_status_unknown",
-							"pod_status_succeeded", "pod_memory_request", "pod_memory_limit", "pod_cpu_limit", "pod_cpu_request",
+							"pod_status_succeeded", "pod_memory_request", "pod_memory_limit", "pod_cpu_limit", "pod_cpu_request", "pod_cpu_usage_total", "pod_memory_working_set",
 							"pod_container_status_running", "pod_container_status_terminated", "pod_container_status_waiting", "pod_container_status_waiting_reason_crash_loop_back_off",
 							"pod_container_status_waiting_reason_image_pull_error", "pod_container_status_waiting_reason_start_error", "pod_container_status_waiting_reason_create_container_error",
 							"pod_container_status_waiting_reason_create_container_config_error", "pod_container_status_terminated_reason_oom_killed",