Add hyperpod metrics

aws · Aug 19, 2024 · 74398de · 74398de
1 parent 6f8ea66
commit 74398de
Show file tree

Hide file tree

Showing 6 changed files with 22 additions and 1 deletion.
diff --git a/translator/tocwconfig/sampleConfig/emf_and_kubernetes_config.yaml b/translator/tocwconfig/sampleConfig/emf_and_kubernetes_config.yaml
@@ -181,6 +181,10 @@ exporters:
                 - node_interface_network_tx_dropped
                 - node_diskio_io_service_bytes_total
                 - node_diskio_io_serviced_total
+                - hyper_pod_node_health_status_schedulable
+                - hyper_pod_node_health_status_unschedulable_pending_replacement
+                - hyper_pod_node_health_status_unschedulable_pending_reboot
+                - hyper_pod_node_health_status_unschedulable
             - dimensions:
                 - - ClusterName
                   - InstanceId

diff --git a/translator/tocwconfig/sampleConfig/emf_and_kubernetes_with_gpu_config.yaml b/translator/tocwconfig/sampleConfig/emf_and_kubernetes_with_gpu_config.yaml
@@ -188,6 +188,10 @@ exporters:
                 - node_interface_network_tx_dropped
                 - node_diskio_io_service_bytes_total
                 - node_diskio_io_serviced_total
+                - hyper_pod_node_health_status_schedulable
+                - hyper_pod_node_health_status_unschedulable_pending_replacement
+                - hyper_pod_node_health_status_unschedulable_pending_reboot
+                - hyper_pod_node_health_status_unschedulable
             - dimensions:
                 - - ClusterName
                   - InstanceId

diff --git a/translator/tocwconfig/sampleConfig/kubernetes_on_prem_config.yaml b/translator/tocwconfig/sampleConfig/kubernetes_on_prem_config.yaml
@@ -148,6 +148,10 @@ exporters:
                 - node_interface_network_tx_dropped
                 - node_diskio_io_service_bytes_total
                 - node_diskio_io_serviced_total
+                - hyper_pod_node_health_status_schedulable
+                - hyper_pod_node_health_status_unschedulable_pending_replacement
+                - hyper_pod_node_health_status_unschedulable_pending_reboot
+                - hyper_pod_node_health_status_unschedulable
             - dimensions:
                 - - ClusterName
                   - InstanceId

diff --git a/translator/tocwconfig/sampleConfig/logs_and_kubernetes_config.yaml b/translator/tocwconfig/sampleConfig/logs_and_kubernetes_config.yaml
@@ -179,6 +179,10 @@ exporters:
                 - node_interface_network_tx_dropped
                 - node_diskio_io_service_bytes_total
                 - node_diskio_io_serviced_total
+                - hyper_pod_node_health_status_schedulable
+                - hyper_pod_node_health_status_unschedulable_pending_replacement
+                - hyper_pod_node_health_status_unschedulable_pending_reboot
+                - hyper_pod_node_health_status_unschedulable
             - dimensions:
                 - - ClusterName
                   - InstanceId

diff --git a/translator/translate/otel/exporter/awsemf/kubernetes.go b/translator/translate/otel/exporter/awsemf/kubernetes.go
@@ -168,6 +168,8 @@ func getNodeMetricDeclarations(conf *confmap.Conf) []*awsemfexporter.MetricDecla
 				MetricNameSelectors: []string{
 					"node_interface_network_rx_dropped", "node_interface_network_tx_dropped",
 					"node_diskio_io_service_bytes_total", "node_diskio_io_serviced_total",
+					"hyper_pod_node_health_status_schedulable", "hyper_pod_node_health_status_unschedulable_pending_replacement",
+					"hyper_pod_node_health_status_unschedulable_pending_reboot", "hyper_pod_node_health_status_unschedulable",
 				},
 			},
 		}

diff --git a/translator/translate/otel/exporter/awsemf/translator_test.go b/translator/translate/otel/exporter/awsemf/translator_test.go
@@ -329,7 +329,10 @@ func TestTranslator(t *testing.T) {
 						},
 						MetricNameSelectors: []string{
 							"node_interface_network_rx_dropped", "node_interface_network_tx_dropped",
-							"node_diskio_io_service_bytes_total", "node_diskio_io_serviced_total"},
+							"node_diskio_io_service_bytes_total", "node_diskio_io_serviced_total",
+							"hyper_pod_node_health_status_schedulable", "hyper_pod_node_health_status_unschedulable_pending_replacement",
+							"hyper_pod_node_health_status_unschedulable_pending_reboot",
+							"hyper_pod_node_health_status_unschedulable"},
 					},
 					{
 						Dimensions:          [][]string{{"NodeName", "InstanceId", "ClusterName"}, {"ClusterName"}},