Add support for Elastic Fabric Adapter (EFA) metrics. (aws#1117)

straussb · Apr 3, 2024 · 5f676d1 · 5f676d1
1 parent 072eac4
commit 5f676d1
Show file tree

Hide file tree

Showing 3 changed files with 147 additions and 0 deletions.
diff --git a/translator/tocwconfig/sampleConfig/emf_and_kubernetes_with_gpu_config.yaml b/translator/tocwconfig/sampleConfig/emf_and_kubernetes_with_gpu_config.yaml
@@ -538,6 +538,76 @@ exporters:
                 - node_neurondevice_hw_ecc_events_total_mem_ecc_uncorrected
                 - node_neurondevice_hw_ecc_events_total_sram_ecc_corrected
                 - node_neurondevice_hw_ecc_events_total_sram_ecc_uncorrected
+            - dimensions:
+                - - ClusterName
+                - - ClusterName
+                  - ContainerName
+                  - Namespace
+                  - PodName
+                - - ClusterName
+                  - ContainerName
+                  - FullPodName
+                  - Namespace
+                  - PodName
+                - - ClusterName
+                  - ContainerName
+                  - EfaDevice
+                  - FullPodName
+                  - Namespace
+                  - PodName
+              label_matchers: []
+              metric_name_selectors:
+                - container_efa_rx_bytes
+                - container_efa_tx_bytes
+                - container_efa_rx_dropped
+                - container_efa_rdma_read_bytes
+                - container_efa_rdma_write_bytes
+                - container_efa_rdma_write_recv_bytes
+            - dimensions:
+                - - ClusterName
+                - - ClusterName
+                  - Namespace
+                - - ClusterName
+                  - Namespace
+                  - Service
+                - - ClusterName
+                  - Namespace
+                  - PodName
+                - - ClusterName
+                  - FullPodName
+                  - Namespace
+                  - PodName
+                - - ClusterName
+                  - EfaDevice
+                  - FullPodName
+                  - Namespace
+                  - PodName
+              label_matchers: []
+              metric_name_selectors:
+                - pod_efa_rx_bytes
+                - pod_efa_tx_bytes
+                - pod_efa_rx_dropped
+                - pod_efa_rdma_read_bytes
+                - pod_efa_rdma_write_bytes
+                - pod_efa_rdma_write_recv_bytes
+            - dimensions:
+                - - ClusterName
+                - - ClusterName
+                  - InstanceId
+                  - NodeName
+                - - ClusterName
+                  - EfaDevice
+                  - InstanceId
+                  - InstanceType
+                  - NodeName
+              label_matchers: []
+              metric_name_selectors:
+                - node_efa_rx_bytes
+                - node_efa_tx_bytes
+                - node_efa_rx_dropped
+                - node_efa_rdma_read_bytes
+                - node_efa_rdma_write_bytes
+                - node_efa_rdma_write_recv_bytes
         metric_descriptors:
             - metric_name: apiserver_admission_controller_admission_duration_seconds
               overwrite: true

diff --git a/translator/translate/otel/exporter/awsemf/kubernetes.go b/translator/translate/otel/exporter/awsemf/kubernetes.go
@@ -52,6 +52,8 @@ func setKubernetesMetricDeclaration(conf *confmap.Conf, cfg *awsemfexporter.Conf
 	// Setup Aws Neuron metrics
 	kubernetesMetricDeclarations = append(kubernetesMetricDeclarations, getAwsNeuronMetricDeclarations(conf)...)
 
+	kubernetesMetricDeclarations = append(kubernetesMetricDeclarations, getEFAMetricDeclarations(conf)...)
+
 	cfg.MetricDeclarations = kubernetesMetricDeclarations
 	cfg.MetricDescriptors = getControlPlaneMetricDescriptors(conf)
 
@@ -615,3 +617,60 @@ func getAwsNeuronMetricDeclarations(conf *confmap.Conf) []*awsemfexporter.Metric
 	}
 	return metricDeclarations
 }
+
+func getEFAMetricDeclarations(conf *confmap.Conf) []*awsemfexporter.MetricDeclaration {
+	var metricDeclarations []*awsemfexporter.MetricDeclaration
+	if awscontainerinsight.EnhancedContainerInsightsEnabled(conf) && awscontainerinsight.AcceleratedComputeMetricsEnabled(conf) {
+		metricDeclarations = []*awsemfexporter.MetricDeclaration{
+			{
+				Dimensions: [][]string{
+					{"ClusterName"},
+					{"ClusterName", "Namespace", "PodName", "ContainerName"},
+					{"ClusterName", "Namespace", "PodName", "FullPodName", "ContainerName"},
+					{"ClusterName", "Namespace", "PodName", "FullPodName", "ContainerName", "EfaDevice"}},
+				MetricNameSelectors: []string{
+					"container_efa_rx_bytes",
+					"container_efa_tx_bytes",
+					"container_efa_rx_dropped",
+					"container_efa_rdma_read_bytes",
+					"container_efa_rdma_write_bytes",
+					"container_efa_rdma_write_recv_bytes",
+				},
+			},
+			{
+				Dimensions: [][]string{
+					{"ClusterName"},
+					{"ClusterName", "Namespace"},
+					{"ClusterName", "Namespace", "Service"},
+					{"ClusterName", "Namespace", "PodName"},
+					{"ClusterName", "Namespace", "PodName", "FullPodName"},
+					{"ClusterName", "Namespace", "PodName", "FullPodName", "EfaDevice"},
+				},
+				MetricNameSelectors: []string{
+					"pod_efa_rx_bytes",
+					"pod_efa_tx_bytes",
+					"pod_efa_rx_dropped",
+					"pod_efa_rdma_read_bytes",
+					"pod_efa_rdma_write_bytes",
+					"pod_efa_rdma_write_recv_bytes",
+				},
+			},
+			{
+				Dimensions: [][]string{
+					{"ClusterName"},
+					{"ClusterName", "NodeName", "InstanceId"},
+					{"ClusterName", "NodeName", "InstanceId", "InstanceType", "EfaDevice"},
+				},
+				MetricNameSelectors: []string{
+					"node_efa_rx_bytes",
+					"node_efa_tx_bytes",
+					"node_efa_rx_dropped",
+					"node_efa_rdma_read_bytes",
+					"node_efa_rdma_write_bytes",
+					"node_efa_rdma_write_recv_bytes",
+				},
+			},
+		}
+	}
+	return metricDeclarations
+}
diff --git a/translator/translate/otel/exporter/awsemf/translator_test.go b/translator/translate/otel/exporter/awsemf/translator_test.go
@@ -513,6 +513,24 @@ func TestTranslator(t *testing.T) {
 							"node_neurondevice_hw_ecc_events_total_sram_ecc_uncorrected",
 						},
 					},
+					{
+						Dimensions: [][]string{{"ClusterName"}, {"ClusterName", "Namespace", "PodName", "ContainerName"}, {"ClusterName", "Namespace", "PodName", "FullPodName", "ContainerName"}, {"ClusterName", "Namespace", "PodName", "FullPodName", "ContainerName", "EfaDevice"}},
+						MetricNameSelectors: []string{
+							"container_efa_rx_bytes", "container_efa_tx_bytes", "container_efa_rx_dropped", "container_efa_rdma_read_bytes", "container_efa_rdma_write_bytes", "container_efa_rdma_write_recv_bytes",
+						},
+					},
+					{
+						Dimensions: [][]string{{"ClusterName"}, {"ClusterName", "Namespace"}, {"ClusterName", "Namespace", "Service"}, {"ClusterName", "Namespace", "PodName"}, {"ClusterName", "Namespace", "PodName", "FullPodName"}, {"ClusterName", "Namespace", "PodName", "FullPodName", "EfaDevice"}},
+						MetricNameSelectors: []string{
+							"pod_efa_rx_bytes", "pod_efa_tx_bytes", "pod_efa_rx_dropped", "pod_efa_rdma_read_bytes", "pod_efa_rdma_write_bytes", "pod_efa_rdma_write_recv_bytes",
+						},
+					},
+					{
+						Dimensions: [][]string{{"ClusterName"}, {"ClusterName", "NodeName", "InstanceId"}, {"ClusterName", "NodeName", "InstanceId", "InstanceType", "EfaDevice"}},
+						MetricNameSelectors: []string{
+							"node_efa_rx_bytes", "node_efa_tx_bytes", "node_efa_rx_dropped", "node_efa_rdma_read_bytes", "node_efa_rdma_write_bytes", "node_efa_rdma_write_recv_bytes",
+						},
+					},
 				},
 				"metric_descriptors": []awsemfexporter.MetricDescriptor{
 					{