Commit 09c45737 authored by Pedro Eduardo Trujillo's avatar Pedro Eduardo Trujillo
Browse files

Actualiza y corrige reglas de alerta

parent b497f3fa

deploy/rules/swarm_node.rules.yml

+7 −7
Original line number Diff line number Diff line
@@ -2,8 +2,8 @@ groups: 
- name: swarm_node

  rules:

  - alert: node_cpu_usage

    expr: 100 - (avg(irate(node_cpu{mode="idle"}[1m]) * ON(instance) GROUP_LEFT(node_name)

      node_meta * 100) BY (node_name)) > 70

    expr: 100 - (avg(irate(node_cpu_seconds_total{mode="idle"}[1m]) * ON(instance) GROUP_LEFT(node_name)

      node_meta * 100) BY (node_name)) > 50

    for: 1m

    labels:

      severity: warning
@@ -12,7 +12,7 @@ groups: 
        $value}}%.

      summary: CPU alert for Swarm node '{{ $labels.node_name }}'

  - alert: node_memory_usage

    expr: sum(((node_memory_MemTotal - node_memory_MemAvailable) / node_memory_MemTotal)

    expr: sum(((node_memory_MemTotal_bytes - node_memory_MemAvailable_bytes) / node_memory_MemTotal_bytes)

      * ON(instance) GROUP_LEFT(node_name) node_meta * 100) BY (node_name) > 80

    for: 1m

    labels:
@@ -22,8 +22,8 @@ groups: 
        $value}}%.

      summary: Memory alert for Swarm node '{{ $labels.node_name }}'

  - alert: node_disk_usage

    expr: ((node_filesystem_size{mountpoint="/"} - node_filesystem_free{mountpoint="/"})

      * 100 / node_filesystem_size{mountpoint="/"}) * ON(instance) GROUP_LEFT(node_name)

    expr: ((node_filesystem_size_bytes{mountpoint="/"} - node_filesystem_free_bytes{mountpoint="/"})

      * 100 / node_filesystem_size_bytes{mountpoint="/"}) * ON(instance) GROUP_LEFT(node_name)

      node_meta > 85

    for: 1m

    labels:
@@ -33,7 +33,7 @@ groups: 
        $value}}%.

      summary: Disk alert for Swarm node '{{ $labels.node_name }}'

  - alert: node_disk_fill_rate_6h

    expr: predict_linear(node_filesystem_free{mountpoint="/"}[1h], 6 * 3600) * ON(instance)

    expr: predict_linear(node_filesystem_free_bytes{mountpoint="/"}[1h], 6 * 3600) * ON(instance)

      GROUP_LEFT(node_name) node_meta < 0

    for: 1h

    labels:

deploy/rules/swarm_task.rules.yml

+6 −3
Original line number Diff line number Diff line
@@ -13,11 +13,14 @@ groups: 
      summary: CPU alert for Swarm task '{{ $labels.container_label_com_docker_swarm_task_name }}'

        on '{{ $labels.container_label_com_docker_swarm_node_id }}'



  - alert: task_high_memory_usage_90

    expr: (sum(container_memory_rss{container_label_com_docker_swarm_task_name=~".+"})

  - record: task_memory_usage_percent

    expr: 100 * (sum(container_memory_rss{container_label_com_docker_swarm_task_name=~".+"})

      BY (container_label_com_docker_swarm_task_name, container_label_com_docker_swarm_node_id) /

      sum(container_spec_memory_limit_bytes{container_label_com_docker_swarm_task_name=~".+"})

      BY (container_label_com_docker_swarm_task_name, container_label_com_docker_swarm_node_id)) > 0.9

      BY (container_label_com_docker_swarm_task_name, container_label_com_docker_swarm_node_id))



  - alert: task_high_memory_usage_90

    expr: task_memory_usage_percent > 90

    for: 1m

    annotations:

      description: '{{ $labels.container_label_com_docker_swarm_task_name }} on ''{{