apiVersion: v1
kind: ConfigMap
metadata:
  name: machine-controller-manager-monitoring-config
  namespace: {{ .Release.Namespace }}
  labels:
    extensions.gardener.cloud/configuration: monitoring
data:
  scrape_config: |
    - job_name: machine-controller-manager
      honor_labels: false
      kubernetes_sd_configs:
      - role: endpoints
        namespaces:
          names: [{{ .Release.Namespace }}]
      relabel_configs:
      - source_labels:
        - __meta_kubernetes_service_name
        - __meta_kubernetes_endpoint_port_name
        action: keep
        regex: machine-controller-manager;metrics
      # common metrics
      - action: labelmap
        regex: __meta_kubernetes_service_label_(.+)
      - source_labels: [ __meta_kubernetes_pod_name ]
        target_label: pod
      metric_relabel_configs:
      - source_labels: [ __name__ ]
        regex: ^(mcm_cloud_api_requests_failed_total|mcm_cloud_api_requests_total|mcm_machine_controller_frozen|mcm_machine_current_status_phase|mcm_machine_deployment_failed_machines|mcm_machine_items_total|mcm_machine_set_failed_machines|mcm_machine_deployment_items_total|mcm_machine_set_items_total|mcm_scrape_failure_total|process_max_fds|process_open_fds|mcm_workqueue_adds_total|mcm_workqueue_depth|mcm_workqueue_queue_duration_seconds_bucket|mcm_workqueue_queue_duration_seconds_sum|mcm_workqueue_queue_duration_seconds_count|mcm_workqueue_work_duration_seconds_bucket|mcm_workqueue_work_duration_seconds_sum|mcm_workqueue_work_duration_seconds_count|mcm_workqueue_unfinished_work_seconds|mcm_workqueue_longest_running_processor_seconds|mcm_workqueue_retries_total)$
        action: keep

  alerting_rules: |
    machine-controller-manager.rules.yaml: |
      groups:
      - name: machine-controller-manager.rules
        rules:
        - alert: MachineControllerManagerDown
          expr: absent(up{job="machine-controller-manager"} == 1)
          for: 15m
          labels:
            service: machine-controller-manager
            severity: critical
            type: seed
            visibility: operator
          annotations:
            description: There are no running machine controller manager instances. No shoot nodes can be created/maintained.
            summary: Machine controller manager is down.

  dashboard_operators: |
    machine-controller-manager-dashboard.json: |-
{{ .Files.Get "mcm-monitoring-dashboard.json" | indent 6 }}