update enhanced CI metrics and aggregations and update tests

aws · Sep 28, 2023 · 78f4c67 · 78f4c67
1 parent bb61f9a
commit 78f4c67
Show file tree

Hide file tree

Showing 16 changed files with 520 additions and 137 deletions.
diff --git a/translator/tocwconfig/sampleConfig/base_container_insights_config.yaml b/translator/tocwconfig/sampleConfig/base_container_insights_config.yaml
@@ -4,7 +4,7 @@ exporters:
     certificate_file_path: ""
     emf_only: true
     endpoint: "https://fake_endpoint"
-    "imds_retries": 1
+    imds_retries: 1
     local_mode: false
     log_group_name: emf/logs/default
     log_retention: 0
@@ -42,7 +42,7 @@ exporters:
     certificate_file_path: ""
     endpoint: ""
     enhanced_container_insights: false
-    "imds_retries": 1
+    imds_retries: 1
     local_mode: false
     retain_initial_value_of_delta_metric: false
     parse_json_encoded_attr_values: [ Sources, kubernetes ]
@@ -144,7 +144,7 @@ receivers:
     enable_control_plane_metrics: false
     certificate_file_path: ""
     endpoint: ""
-    "imds_retries": 1
+    imds_retries: 1
     prefer_full_pod_name: true
     leader_lock_name: cwagent-clusterleader
     leader_lock_using_config_map_only: true

diff --git a/translator/tocwconfig/sampleConfig/complete_darwin_config.yaml b/translator/tocwconfig/sampleConfig/complete_darwin_config.yaml
@@ -20,7 +20,7 @@ exporters:
         emf_only: true
         certificate_file_path: ""
         endpoint: "https://logs-fips.us-west-2.amazonaws.com"
-        "imds_retries": 1
+        imds_retries: 1
         local_mode: false
         log_group_name: emf/logs/default
         log_retention: 0
@@ -49,7 +49,7 @@ exporters:
         aws_log_groups: []
         certificate_file_path: ""
         endpoint: https://x-ray-endpoint.us-west-2.amazonaws.com
-        "imds_retries": 1
+        imds_retries: 1
         index_all_attributes: false
         indexed_attributes: []
         local_mode: true

diff --git a/translator/tocwconfig/sampleConfig/complete_linux_config.yaml b/translator/tocwconfig/sampleConfig/complete_linux_config.yaml
@@ -23,7 +23,7 @@ exporters:
         emf_only: true
         certificate_file_path: ""
         endpoint: "https://logs-fips.us-west-2.amazonaws.com"
-        "imds_retries": 1
+        imds_retries: 1
         local_mode: false
         log_group_name: emf/logs/default
         log_retention: 0
@@ -52,7 +52,7 @@ exporters:
         aws_log_groups: []
         certificate_file_path: ""
         endpoint: https://x-ray-endpoint.us-west-2.amazonaws.com
-        "imds_retries": 1
+        imds_retries: 1
         index_all_attributes: false
         indexed_attributes: []
         local_mode: true

diff --git a/translator/tocwconfig/sampleConfig/complete_windows_config.yaml b/translator/tocwconfig/sampleConfig/complete_windows_config.yaml
@@ -20,7 +20,7 @@ exporters:
         certificate_file_path: ""
         emf_only: true
         endpoint: "https://logs-fips.us-west-2.amazonaws.com"
-        "imds_retries": 1
+        imds_retries: 1
         local_mode: false
         log_group_name: emf/logs/default
         log_retention: 0
@@ -49,7 +49,7 @@ exporters:
         aws_log_groups: []
         certificate_file_path: ""
         endpoint: https://x-ray-endpoint.us-west-2.amazonaws.com
-        "imds_retries": 1
+        imds_retries: 1
         index_all_attributes: false
         indexed_attributes: []
         local_mode: true

diff --git a/translator/tocwconfig/sampleConfig/config_with_env.yaml b/translator/tocwconfig/sampleConfig/config_with_env.yaml
@@ -4,7 +4,7 @@ exporters:
     certificate_file_path: ""
     emf_only: true
     endpoint: ""
-    "imds_retries": 1
+    imds_retries: 1
     local_mode: false
     log_group_name: emf/logs/default
     log_retention: 0

diff --git a/translator/tocwconfig/sampleConfig/emf_and_kubernetes_config.yaml b/translator/tocwconfig/sampleConfig/emf_and_kubernetes_config.yaml
@@ -4,7 +4,7 @@ exporters:
     emf_only: true
     certificate_file_path: ""
     endpoint: "https://fake_endpoint"
-    "imds_retries": 2
+    imds_retries: 2
     local_mode: false
     log_group_name: emf/logs/default
     log_retention: 0
@@ -36,8 +36,8 @@ exporters:
     eks_fargate_container_insights_enabled: false
     certificate_file_path: ""
     endpoint: ""
-    "imds_retries": 2
     enhanced_container_insights: true
+    imds_retries: 2
     local_mode: false
     log_group_name: /aws/containerinsights/{ClusterName}/performance
     log_retention: 0
@@ -53,11 +53,20 @@ exporters:
           - container_memory_utilization
           - container_memory_utilization_over_container_limit
           - container_memory_failures_total
+          - container_memory_limit
+          - container_memory_request
           - container_filesystem_usage
+          - container_filesystem_available
+          - container_filesystem_utilization
           - container_status_running
           - container_status_terminated
           - container_status_waiting
-          - container_status_waiting_reason_crashed
+          - container_status_waiting_reason_crash_loop_back_off
+          - container_status_waiting_reason_image_pull_error
+          - container_status_waiting_reason_start_error
+          - container_status_waiting_reason_create_container_error
+          - container_status_waiting_reason_create_container_config_error
+          - container_status_terminated_reason_oom_killed
       # pod metrics
       - dimensions: [ [ ClusterName, Namespace, PodName ], [ ClusterName ], [ ClusterName, Namespace, Service ], [ ClusterName, Namespace ], [ ClusterName, FullPodName, Namespace, PodName ] ]
         label_matchers: [ ]
@@ -72,9 +81,7 @@ exporters:
         label_matchers: [ ]
         metric_name_selectors:
           - pod_interface_network_rx_dropped
-          - pod_interface_network_rx_errors
           - pod_interface_network_tx_dropped
-          - pod_interface_network_tx_errors
       - dimensions: [ [ ClusterName, Namespace, PodName ], [ ClusterName ], [ ClusterName, FullPodName, Namespace, PodName ], [ ClusterName, Namespace, Service ] ]
         label_matchers: []
         metric_name_selectors:
@@ -90,6 +97,8 @@ exporters:
           - pod_status_failed
           - pod_status_unknown
           - pod_status_succeeded
+          - pod_memory_request
+          - pod_memory_limit
       # node metrics
       - dimensions: [ [ ClusterName, InstanceId, NodeName ], [ ClusterName ] ]
         label_matchers: [ ]
@@ -117,9 +126,7 @@ exporters:
         label_matchers: [ ]
         metric_name_selectors:
           - node_interface_network_rx_dropped
-          - node_interface_network_rx_errors
           - node_interface_network_tx_dropped
-          - node_interface_network_tx_errors
           - node_diskio_io_service_bytes_total
           - node_diskio_io_serviced_total
       # node fs metrics
@@ -164,56 +171,118 @@ exporters:
       - dimensions: [ [ClusterName, endpoint], [ ClusterName ] ]
         label_matchers: [ ]
         metric_name_selectors:
+          - apiserver_storage_size_bytes
+          - apiserver_storage_size_bytes
           - etcd_db_total_size_in_bytes
+          - etcd_request_duration_seconds
       - dimensions: [ [ClusterName, resource], [ ClusterName ] ]
         label_matchers: [ ]
         metric_name_selectors:
           - apiserver_storage_list_duration_seconds
-      - dimensions: [ [ClusterName, priority_level], [ ClusterName ] ]
+          - apiserver_longrunning_requests
+          - apiserver_storage_objects
+      - dimensions: [ [ClusterName, verb], [ ClusterName ] ]
         label_matchers: [ ]
         metric_name_selectors:
-          - apiserver_flowcontrol_request_concurrency_limit
-      - dimensions: [ [ ClusterName ] ]
+          - apiserver_request_duration_seconds
+          - rest_client_request_duration_seconds
+      - dimensions: [ [ClusterName, code, verb], [ ClusterName ] ]
         label_matchers: [ ]
         metric_name_selectors:
-          - apiserver_storage_objects
           - apiserver_request_total
           - apiserver_request_total_5xx
-          - apiserver_request_duration_seconds
+      - dimensions: [ [ClusterName, operation], [ ClusterName ] ]
+        label_matchers: [ ]
+        metric_name_selectors:
           - apiserver_admission_controller_admission_duration_seconds
-          - rest_client_request_duration_seconds
-          - rest_client_requests_total
+          - apiserver_admission_step_admission_duration_seconds
           - etcd_request_duration_seconds
+      - dimensions: [ [ClusterName, code, method], [ ClusterName ] ]
+        label_matchers: [ ]
+        metric_name_selectors:
+          - rest_client_requests_total
+      - dimensions: [ [ClusterName, request_kind], [ ClusterName ] ]
+        label_matchers: [ ]
+        metric_name_selectors:
+          - apiserver_current_inflight_requests
+      - dimensions: [ [ClusterName, name], [ ClusterName ] ]
+        label_matchers: [ ]
+        metric_name_selectors:
+          - apiserver_admission_webhook_admission_duration_seconds
+      - dimensions: [ [ClusterName, group], [ ClusterName ] ]
+        label_matchers: [ ]
+        metric_name_selectors:
+          - apiserver_requested_deprecated_apis
+      - dimensions: [ [ClusterName, reason], [ ClusterName ] ]
+        label_matchers: [ ]
+        metric_name_selectors:
           - apiserver_flowcontrol_rejected_requests_total
+      - dimensions: [ [ ClusterName, priority_level ], [ ClusterName ] ]
+        label_matchers: [ ]
+        metric_name_selectors:
+          - apiserver_flowcontrol_request_concurrency_limit
     metric_descriptors:
-          - metric_name: apiserver_storage_objects
+          - metric_name: apiserver_admission_controller_admission_duration_seconds
+            unit: Seconds
+            overwrite: true
+          - metric_name: apiserver_admission_step_admission_duration_seconds
+            unit: Seconds
+            overwrite: true
+          - metric_name: apiserver_admission_webhook_admission_duration_seconds
+            unit: Seconds
+            overwrite: true
+          - metric_name: apiserver_current_inflight_requests
             unit: Count
             overwrite: true
-          - metric_name: apiserver_request_total
+          - metric_name: apiserver_current_inqueue_requests
             unit: Count
             overwrite: true
-          - metric_name: apiserver_request_total_5xx
+          - metric_name: apiserver_flowcontrol_rejected_requests_total
+            unit: Count
+            overwrite: true
+          - metric_name: apiserver_flowcontrol_request_concurrency_limit
+            unit: Count
+            overwrite: true
+          - metric_name: apiserver_longrunning_requests
             unit: Count
             overwrite: true
           - metric_name: apiserver_request_duration_seconds
             unit: Seconds
             overwrite: true
-          - metric_name: apiserver_admission_controller_admission_duration_seconds
+          - metric_name: apiserver_request_total
+            unit: Count
+            overwrite: true
+          - metric_name: apiserver_request_total_5xx
+            unit: Count
+            overwrite: true
+          - metric_name: apiserver_storage_objects
+            unit: Count
+            overwrite: true
+          - metric_name: etcd_request_duration_seconds
             unit: Seconds
             overwrite: true
-          - metric_name: rest_client_request_duration_seconds
+          - metric_name: apiserver_storage_list_duration_seconds
             unit: Seconds
             overwrite: true
-          - metric_name: rest_client_requests_total
+          - metric_name: apiserver_storage_objects
             unit: Count
             overwrite: true
+          - metric_name: apiserver_storage_db_total_size_in_bytes
+            unit: Bytes
+            overwrite: true
+          - metric_name: apiserver_storage_size_bytes
+            unit: Bytes
+            overwrite: true
+          - metric_name: etcd_db_total_size_in_bytes
+            unit: Bytes
+            overwrite: true
           - metric_name: etcd_request_duration_seconds
             unit: Seconds
             overwrite: true
-          - metric_name: apiserver_flowcontrol_rejected_requests_total
-            unit: Count
+          - metric_name: rest_client_request_duration_seconds
+            unit: Seconds
             overwrite: true
-          - metric_name: apiserver_flowcontrol_request_concurrency_limit
+          - metric_name: rest_client_requests_total
             unit: Count
             overwrite: true
     namespace: ContainerInsights
@@ -270,7 +339,7 @@ receivers:
     enable_control_plane_metrics: true
     certificate_file_path: ""
     endpoint: ""
-    "imds_retries": 2
+    imds_retries: 2
     leader_lock_name: cwagent-clusterleader
     leader_lock_using_config_map_only: true
     local_mode: false