...

Text file src/edge-infra.dev/config/pallets/o11y/otel/base/monitoring.yaml

Documentation: edge-infra.dev/config/pallets/o11y/otel/base

     1apiVersion: monitoring.coreos.com/v1
     2kind: ServiceMonitor
     3metadata:
     4  name: kube-apiserver
     5  labels:
     6    app.kubernetes.io/name: apiserver
     7spec:
     8  selector:
     9    matchLabels:
    10      component: apiserver
    11      provider: kubernetes
    12  endpoints:
    13  - port: https
    14    bearerTokenFile: /var/run/secrets/kubernetes.io/serviceaccount/token
    15    interval: 5m
    16    metricRelabelings:
    17    - action: drop
    18      regex: kubelet_(pod_worker_latency_microseconds|pod_start_latency_microseconds|cgroup_manager_latency_microseconds|pod_worker_start_latency_microseconds|pleg_relist_latency_microseconds|pleg_relist_interval_microseconds|runtime_operations|runtime_operations_latency_microseconds|runtime_operations_errors|eviction_stats_age_microseconds|device_plugin_registration_count|device_plugin_alloc_latency_microseconds|network_plugin_operations_latency_microseconds)
    19      sourceLabels:
    20      - __name__
    21    - action: drop
    22      regex: scheduler_(e2e_scheduling_latency_microseconds|scheduling_algorithm_predicate_evaluation|scheduling_algorithm_priority_evaluation|scheduling_algorithm_preemption_evaluation|scheduling_algorithm_latency_microseconds|binding_latency_microseconds|scheduling_latency_seconds)
    23      sourceLabels:
    24      - __name__
    25    - action: drop
    26      regex: apiserver_(request_count|request_latencies|request_latencies_summary|dropped_requests|storage_data_key_generation_latencies_microseconds|storage_transformation_failures_total|storage_transformation_latencies_microseconds|proxy_tunnel_sync_latency_secs|request_sli_.*|request_slo_.*)
    27      sourceLabels:
    28      - __name__
    29    - action: drop
    30      regex: kubelet_docker_(operations|operations_latency_microseconds|operations_errors|operations_timeout)
    31      sourceLabels:
    32      - __name__
    33    - action: drop
    34      regex: reflector_(items_per_list|items_per_watch|list_duration_seconds|lists_total|short_watches_total|watch_duration_seconds|watches_total)
    35      sourceLabels:
    36      - __name__
    37    - action: drop
    38      regex: etcd_(helper_cache_hit_count|helper_cache_miss_count|helper_cache_entry_count|object_counts|request_cache_get_latencies_summary|request_cache_add_latencies_summary|request_latencies_summary|request_duration_seconds.*)
    39      sourceLabels:
    40      - __name__
    41    - action: drop
    42      regex: transformation_(transformation_latencies_microseconds|failures_total)
    43      sourceLabels:
    44      - __name__
    45    - action: drop
    46      regex: (admission_quota_controller_adds|admission_quota_controller_depth|admission_quota_controller_longest_running_processor_microseconds|admission_quota_controller_queue_latency|admission_quota_controller_unfinished_work_seconds|admission_quota_controller_work_duration|APIServiceOpenAPIAggregationControllerQueue1_adds|APIServiceOpenAPIAggregationControllerQueue1_depth|APIServiceOpenAPIAggregationControllerQueue1_longest_running_processor_microseconds|APIServiceOpenAPIAggregationControllerQueue1_queue_latency|APIServiceOpenAPIAggregationControllerQueue1_retries|APIServiceOpenAPIAggregationControllerQueue1_unfinished_work_seconds|APIServiceOpenAPIAggregationControllerQueue1_work_duration|APIServiceRegistrationController_adds|APIServiceRegistrationController_depth|APIServiceRegistrationController_longest_running_processor_microseconds|APIServiceRegistrationController_queue_latency|APIServiceRegistrationController_retries|APIServiceRegistrationController_unfinished_work_seconds|APIServiceRegistrationController_work_duration|autoregister_adds|autoregister_depth|autoregister_longest_running_processor_microseconds|autoregister_queue_latency|autoregister_retries|autoregister_unfinished_work_seconds|autoregister_work_duration|AvailableConditionController_adds|AvailableConditionController_depth|AvailableConditionController_longest_running_processor_microseconds|AvailableConditionController_queue_latency|AvailableConditionController_retries|AvailableConditionController_unfinished_work_seconds|AvailableConditionController_work_duration|crd_autoregistration_controller_adds|crd_autoregistration_controller_depth|crd_autoregistration_controller_longest_running_processor_microseconds|crd_autoregistration_controller_queue_latency|crd_autoregistration_controller_retries|crd_autoregistration_controller_unfinished_work_seconds|crd_autoregistration_controller_work_duration|crdEstablishing_adds|crdEstablishing_depth|crdEstablishing_longest_running_processor_microseconds|crdEstablishing_queue_latency|crdEstablishing_retries|crdEstablishing_unfinished_work_seconds|crdEstablishing_work_duration|crd_finalizer_adds|crd_finalizer_depth|crd_finalizer_longest_running_processor_microseconds|crd_finalizer_queue_latency|crd_finalizer_retries|crd_finalizer_unfinished_work_seconds|crd_finalizer_work_duration|crd_naming_condition_controller_adds|crd_naming_condition_controller_depth|crd_naming_condition_controller_longest_running_processor_microseconds|crd_naming_condition_controller_queue_latency|crd_naming_condition_controller_retries|crd_naming_condition_controller_unfinished_work_seconds|crd_naming_condition_controller_work_duration|crd_openapi_controller_adds|crd_openapi_controller_depth|crd_openapi_controller_longest_running_processor_microseconds|crd_openapi_controller_queue_latency|crd_openapi_controller_retries|crd_openapi_controller_unfinished_work_seconds|crd_openapi_controller_work_duration|DiscoveryController_adds|DiscoveryController_depth|DiscoveryController_longest_running_processor_microseconds|DiscoveryController_queue_latency|DiscoveryController_retries|DiscoveryController_unfinished_work_seconds|DiscoveryController_work_duration|kubeproxy_sync_proxy_rules_latency_microseconds|non_structural_schema_condition_controller_adds|non_structural_schema_condition_controller_depth|non_structural_schema_condition_controller_longest_running_processor_microseconds|non_structural_schema_condition_controller_queue_latency|non_structural_schema_condition_controller_retries|non_structural_schema_condition_controller_unfinished_work_seconds|non_structural_schema_condition_controller_work_duration|rest_client_request_latency_seconds|storage_operation_errors_total|storage_operation_status_count)
    47      sourceLabels:
    48      - __name__
    49    - action: drop
    50      regex: etcd_(debugging|disk|server).*
    51      sourceLabels:
    52      - __name__
    53    - action: drop
    54      regex: apiserver_admission_controller_admission_latencies_seconds_.*
    55      sourceLabels:
    56      - __name__
    57    - action: drop
    58      regex: apiserver_admission_step_admission_latencies_seconds_.*
    59      sourceLabels:
    60      - __name__
    61    - action: drop
    62      regex: apiserver_request_duration_seconds_bucket;(0.15|0.25|0.3|0.35|0.4|0.45|0.6|0.7|0.8|0.9|1.25|1.5|1.75|2.5|3|3.5|4.5|6|7|8|9|15|25|30|50)
    63      sourceLabels:
    64      - __name__
    65      - le
    66    scheme: https
    67    scrapeTimeout: 50s
    68    tlsConfig:
    69      caFile: /var/run/secrets/kubernetes.io/serviceaccount/ca.crt
    70      serverName: kubernetes
    71  jobLabel: component
    72  namespaceSelector:
    73    matchNames:
    74    - default
    75---
    76apiVersion: monitoring.coreos.com/v1
    77kind: ServiceMonitor
    78metadata:
    79  name: kubelet
    80  labels:
    81    app.kubernetes.io/name: kubelet
    82spec:
    83  selector:
    84    matchLabels:
    85      app.kubernetes.io/name: kubelet
    86  endpoints:
    87  - port: https-metrics
    88    bearerTokenFile: /var/run/secrets/kubernetes.io/serviceaccount/token
    89    honorLabels: true
    90    interval: 60s
    91    metricRelabelings:
    92    - action: drop
    93      regex: kubelet_(pod_worker_latency_microseconds|pod_start_latency_microseconds|cgroup_manager_latency_microseconds|pod_worker_start_latency_microseconds|pleg_relist_latency_microseconds|pleg_relist_interval_microseconds|runtime_operations|runtime_operations_latency_microseconds|runtime_operations_errors|eviction_stats_age_microseconds|device_plugin_registration_count|device_plugin_alloc_latency_microseconds|network_plugin_operations_latency_microseconds)
    94      sourceLabels:
    95      - __name__
    96    - action: drop
    97      regex: scheduler_(e2e_scheduling_latency_microseconds|scheduling_algorithm_predicate_evaluation|scheduling_algorithm_priority_evaluation|scheduling_algorithm_preemption_evaluation|scheduling_algorithm_latency_microseconds|binding_latency_microseconds|scheduling_latency_seconds)
    98      sourceLabels:
    99      - __name__
   100    - action: drop
   101      regex: apiserver_(request_count|request_latencies|request_latencies_summary|dropped_requests|storage_data_key_generation_latencies_microseconds|storage_transformation_failures_total|storage_transformation_latencies_microseconds|proxy_tunnel_sync_latency_secs)
   102      sourceLabels:
   103      - __name__
   104    - action: drop
   105      regex: kubelet_docker_(operations|operations_latency_microseconds|operations_errors|operations_timeout)
   106      sourceLabels:
   107      - __name__
   108    - action: drop
   109      regex: reflector_(items_per_list|items_per_watch|list_duration_seconds|lists_total|short_watches_total|watch_duration_seconds|watches_total)
   110      sourceLabels:
   111      - __name__
   112    - action: drop
   113      regex: etcd_(helper_cache_hit_count|helper_cache_miss_count|helper_cache_entry_count|object_counts|request_cache_get_latencies_summary|request_cache_add_latencies_summary|request_latencies_summary)
   114      sourceLabels:
   115      - __name__
   116    - action: drop
   117      regex: transformation_(transformation_latencies_microseconds|failures_total)
   118      sourceLabels:
   119      - __name__
   120    - action: drop
   121      regex: (admission_quota_controller_adds|admission_quota_controller_depth|admission_quota_controller_longest_running_processor_microseconds|admission_quota_controller_queue_latency|admission_quota_controller_unfinished_work_seconds|admission_quota_controller_work_duration|APIServiceOpenAPIAggregationControllerQueue1_adds|APIServiceOpenAPIAggregationControllerQueue1_depth|APIServiceOpenAPIAggregationControllerQueue1_longest_running_processor_microseconds|APIServiceOpenAPIAggregationControllerQueue1_queue_latency|APIServiceOpenAPIAggregationControllerQueue1_retries|APIServiceOpenAPIAggregationControllerQueue1_unfinished_work_seconds|APIServiceOpenAPIAggregationControllerQueue1_work_duration|APIServiceRegistrationController_adds|APIServiceRegistrationController_depth|APIServiceRegistrationController_longest_running_processor_microseconds|APIServiceRegistrationController_queue_latency|APIServiceRegistrationController_retries|APIServiceRegistrationController_unfinished_work_seconds|APIServiceRegistrationController_work_duration|autoregister_adds|autoregister_depth|autoregister_longest_running_processor_microseconds|autoregister_queue_latency|autoregister_retries|autoregister_unfinished_work_seconds|autoregister_work_duration|AvailableConditionController_adds|AvailableConditionController_depth|AvailableConditionController_longest_running_processor_microseconds|AvailableConditionController_queue_latency|AvailableConditionController_retries|AvailableConditionController_unfinished_work_seconds|AvailableConditionController_work_duration|crd_autoregistration_controller_adds|crd_autoregistration_controller_depth|crd_autoregistration_controller_longest_running_processor_microseconds|crd_autoregistration_controller_queue_latency|crd_autoregistration_controller_retries|crd_autoregistration_controller_unfinished_work_seconds|crd_autoregistration_controller_work_duration|crdEstablishing_adds|crdEstablishing_depth|crdEstablishing_longest_running_processor_microseconds|crdEstablishing_queue_latency|crdEstablishing_retries|crdEstablishing_unfinished_work_seconds|crdEstablishing_work_duration|crd_finalizer_adds|crd_finalizer_depth|crd_finalizer_longest_running_processor_microseconds|crd_finalizer_queue_latency|crd_finalizer_retries|crd_finalizer_unfinished_work_seconds|crd_finalizer_work_duration|crd_naming_condition_controller_adds|crd_naming_condition_controller_depth|crd_naming_condition_controller_longest_running_processor_microseconds|crd_naming_condition_controller_queue_latency|crd_naming_condition_controller_retries|crd_naming_condition_controller_unfinished_work_seconds|crd_naming_condition_controller_work_duration|crd_openapi_controller_adds|crd_openapi_controller_depth|crd_openapi_controller_longest_running_processor_microseconds|crd_openapi_controller_queue_latency|crd_openapi_controller_retries|crd_openapi_controller_unfinished_work_seconds|crd_openapi_controller_work_duration|DiscoveryController_adds|DiscoveryController_depth|DiscoveryController_longest_running_processor_microseconds|DiscoveryController_queue_latency|DiscoveryController_retries|DiscoveryController_unfinished_work_seconds|DiscoveryController_work_duration|kubeproxy_sync_proxy_rules_latency_microseconds|non_structural_schema_condition_controller_adds|non_structural_schema_condition_controller_depth|non_structural_schema_condition_controller_longest_running_processor_microseconds|non_structural_schema_condition_controller_queue_latency|non_structural_schema_condition_controller_retries|non_structural_schema_condition_controller_unfinished_work_seconds|non_structural_schema_condition_controller_work_duration|rest_client_request_latency_seconds|storage_operation_errors_total|storage_operation_status_count)
   122      sourceLabels:
   123      - __name__
   124    relabelings:
   125    - sourceLabels:
   126      - __metrics_path__
   127      targetLabel: metrics_path
   128    scheme: https
   129    scrapeTimeout: 50s
   130    tlsConfig:
   131      insecureSkipVerify: true
   132  - port: https-metrics
   133    bearerTokenFile: /var/run/secrets/kubernetes.io/serviceaccount/token
   134    honorLabels: false
   135    honorTimestamps: false
   136    interval: 60s
   137    metricRelabelings:
   138    - action: drop
   139      regex: container_(network_tcp_usage_total|network_udp_usage_total|tasks_state|cpu_load_average_10s)
   140      sourceLabels:
   141      - __name__
   142    - action: drop
   143      regex: (container_spec_.*|container_file_descriptors|container_sockets|container_threads_max|container_threads|container_start_time_seconds|container_last_seen);;
   144      sourceLabels:
   145      - __name__
   146      - pod
   147      - namespace
   148    - action: drop
   149      regex: (container_blkio_device_usage_total);.+
   150      sourceLabels:
   151      - __name__
   152      - container
   153    - action: replace
   154      regex: (.+)
   155      replacement: $1
   156      sourceLabels:
   157      - namespace
   158      targetLabel: exported_namespace
   159    path: /metrics/cadvisor
   160    relabelings:
   161    - sourceLabels:
   162      - __metrics_path__
   163      targetLabel: metrics_path
   164    - action: labeldrop
   165      regex: (__meta_kubernetes_namespace|namespace)
   166    scheme: https
   167    scrapeTimeout: 50s
   168    tlsConfig:
   169      insecureSkipVerify: true
   170  - port: https-metrics
   171    bearerTokenFile: /var/run/secrets/kubernetes.io/serviceaccount/token
   172    honorLabels: true
   173    interval: 60s
   174    path: /metrics/probes
   175    relabelings:
   176    - sourceLabels:
   177      - __metrics_path__
   178      targetLabel: metrics_path
   179    scheme: https
   180    scrapeTimeout: 50s
   181    tlsConfig:
   182      insecureSkipVerify: true
   183  jobLabel: app.kubernetes.io/name
   184  namespaceSelector:
   185    matchNames:
   186    - kube-system

View as plain text