gitlab-org--gitlab-foss/config/prometheus/pod_metrics.yml

dashboard: 'K8s pod health'
priority: 2

templating:
  variables:
    pod:
      label: 'Pod name'
      type: metric_label_values
      options:
        series_selector: 'container_memory_working_set_bytes'
        label: 'pod'

panel_groups:
- group: CPU metrics
  panels:
  - title: "CPU usage"
    type: "line-chart"
    y_label: "Cores per container"
    metrics:
    - id: pod_cpu_usage_seconds_total
      query_range: >-
        sum(
          rate(container_cpu_usage_seconds_total{pod="{{pod}}",container!="POD"}[5m])
        )
        by (container)
      unit: "cores"
      label: container

  - title: "CPU throttling"
    type: "line-chart"
    y_label: "Cores per container"
    metrics:
    - id: pod_cpu_cfs_throttle
      query_range: >-
        sum(
          rate(container_cpu_cfs_throttled_seconds_total{pod="{{pod}}"}[5m])
        )
        by (container)
      unit: "cores"
      label: container

- group: Memory metrics
  panels:
  - title: "Memory usage working set"
    type: "line-chart"
    y_label: "Working set memory"
    metrics:
    - id: pod_memory_working_set
      query_range: >-
        sum(
          container_memory_working_set_bytes{pod="{{pod}}",container!="POD"}
        ) by (container)
      unit: "bytes"
      label: container

- group: Network metrics
  panels:
  - title: "Network Receive (In)"
    type: "line-chart"
    y_label: "Received (bytes/sec)"
    metrics:
    - id: pod_network_receive
      query_range: >-
        sum(
          rate(
            container_network_receive_bytes_total{pod="{{pod}}"}[5m]
          )
        ) by (pod)
      unit: "bytes"
      label: pod

  - title: "Network Transmit (Out)"
    type: "line-chart"
    y_label: "Transmitted (bytes/sec)"
    metrics:
    - id: pod_network_transmit
      query_range: >-
        sum(
          rate(
            container_network_transmit_bytes_total{pod="{{pod}}"}[5m]
          )
        ) by (pod)
      unit: bytes
      label: pod

- group: Disk metrics
  panels:
  - title: "Disk Reads"
    type: "line-chart"
    y_label: "Disk reads (bytes/sec)"
    metrics:
    - id: pod_disk_reads
      query_range: >-
        sum(
          rate(
            container_fs_reads_bytes_total{pod="{{pod}}", container!="POD"}[5m]
          )
        ) by (container,device)

      unit: "bytes / sec"
      label: "{{container}} {{device}}"

  - title: "Disk Writes"
    type: "line-chart"
    y_label: "Disk writes (bytes/sec)"
    metrics:
    - id: pod_disk_writes
      query_range: >-
        sum(
          rate(
            container_fs_writes_bytes_total{pod="{{pod}}", container!="POD"}[5m]
          )
        ) by (container,device)
      unit: "bytes / sec"
      label: "{{container}} {{device}}"