ceph/meta/prometheus.yml

{%- from "ceph/map.jinja" import thresholds, mon, monitoring, setup with context %}

{%- if (mon is defined and mon.get('enabled')) or (monitoring.cluster_stats.get('enabled') and monitoring.cluster_stats.ceph_user is defined) %}
{% raw %}
server:
  alert:
    CephClusterHealthMinor:
      if: >-
        ceph_overall_health == 2
      for: 3m
      labels:
        severity: minor
        service: ceph
      annotations:
        summary: "Ceph cluster health is WARNING"
        description: "The Ceph cluster is in the WARNING state. For details, run 'ceph -s'."
    CephClusterHealthCritical:
      if: >-
        ceph_overall_health == 3
      for: 3m
      labels:
        severity: critical
        service: ceph
      annotations:
        summary: "Ceph cluster health is CRITICAL"
        description: "The Ceph cluster is in the CRITICAL state. For details, run 'ceph -s'."
    CephMonitorDownMinor:
      if: >-
        100 * (1 - ceph_num_mon_quorum / ceph_num_mon) > 0
      for: 3m
      labels:
        severity: minor
        service: ceph
      annotations:
        summary: "Ceph Monitors are down"
        description: "{{ $value }}% of Ceph Monitors are down. For details, run 'ceph -s'."
    CephOsdDownMinor:
      if: >-
        100 * (1 - ceph_osdmap_num_up_osds / ceph_osdmap_num_osds) > 0
      for: 3m
      labels:
        severity: minor
        service: ceph
      annotations:
        summary: "Ceph OSDs are down"
        description: "{{ $value }}% of Ceph OSDs are down. For details, run 'ceph osd tree'."
    CephOsdSpaceUsageWarning:
      {%- endraw %}
      {%- set threshold = monitoring.space_used_warning_threshold|default('0.75')|float %}
      if: >-
        ceph_osd_bytes_used > ceph_osd_bytes * {{threshold}}
      {%- raw %}
      for: 3m
      labels:
        severity: warning
        service: ceph
      annotations:
        summary: "{%-endraw %}{{100*threshold}}{%- raw %}% of Ceph space is used"
        description: "{{ $value }} bytes of Ceph OSD space (>= {%-endraw %}{{100*threshold}}{%- raw %}%) is used for 3 minutes. For details, run 'ceph df'."
    CephOsdSpaceUsageMajor:
      {%- endraw %}
      {%- set threshold = monitoring.space_used_critical_threshold|default('0.85')|float %}
      if: >-
        ceph_osd_bytes_used > ceph_osd_bytes * {{threshold}}
      {%- raw %}
      for: 3m
      labels:
        severity: major
        service: ceph
      annotations:
        summary: "{%-endraw %}{{100*threshold}}{%- raw %}% of Ceph space is used"
        description: "{{ $ value }} bytes of Ceph OSD space (>= {%-endraw %}{{100*threshold}}{%- raw %}%) is used for 3 minutes. For details, run 'ceph df'."
{% endraw %}
{%- if setup.pool is defined %}
{%- for pool_name, pool in setup.pool.iteritems() %}
    {%- if monitoring.pool is defined and monitoring.pool[pool_name] is defined %}
       {%- set monitoring_pool = monitoring.pool[pool_name] %}
    {%- else %}
       {%- set monitoring_pool = monitoring %}
    {%- endif %}
    CephPool{{pool_name|replace(".", "")|replace("-", "")}}SpaceUsageWarning:
      {%- set threshold = monitoring_pool.pool_space_used_utilization_warning_threshold|default('0.75')|float %}
      if: >-
        ceph_pool_usage_bytes_used{name="{{pool_name}}"} / (ceph_pool_usage_max_avail{name="{{pool_name}}"} + ceph_pool_usage_bytes_used{name="{{pool_name}}"}) > {{threshold}}
      for: 3m
      labels:
        severity: warning
        service: ceph
      annotations:
        summary: "{{100*threshold}}% of Ceph pool space is used"
        description: "The Ceph {{pool_name}} pool uses {{100*threshold}}% of available space for 3 minutes. For details, run 'ceph df'."
    CephPool{{pool_name|replace(".", "")|replace("-", "")}}SpaceUsageCritical:
      {%- set threshold = monitoring_pool.pool_space_used_critical_threshold|default('0.85')|float %}
      if: >-
        ceph_pool_usage_bytes_used{name="{{pool_name}}"} / (ceph_pool_usage_max_avail{name="{{pool_name}}"} + ceph_pool_usage_bytes_used{name="{{pool_name}}"}) > {{threshold}}
      for: 3m
      labels:
        severity: minor
        service: ceph
      annotations:
        summary: "{{100*threshold}}% of Ceph pool space is used"
        description: "The Ceph {{pool_name}} pool uses {{100*threshold}}% of available space for 3 minutes. For details, run 'ceph df'."
    {%- if monitoring.cluster_stats.extra_alerts.get("enabled", False) %}
    CephPool{{pool_name|replace(".", "")|replace("-", "")}}WriteOpsTooHigh:
      {%- set threshold = monitoring_pool.pool_write_ops_threshold|default('200')|float %}
      if: >-
        ceph_pool_stats_write_op_per_sec{name="{{pool_name}}"} > {{threshold}}
      for: 3m
      labels:
        severity: warning
        service: ceph
      annotations:
        summary: "{{threshold}} Ceph pool write operations per second"
        description: "The number of Ceph {{pool_name}} pool write operations per second is {{threshold}} for 3 minutes."
    CephPool{{pool_name|replace(".", "")|replace("-", "")}}WriteBytesTooHigh:
      {%- set threshold = monitoring_pool.pool_write_bytes_threshold|default('70000000')|float %}
      if: >-
        ceph_pool_stats_write_bytes_sec{name="{{pool_name}}"} > {{threshold}}
      for: 3m
      labels:
        severity: warning
        service: ceph
      annotations:
        summary: "{{threshold}} Ceph pool write bytes per second"
        description: "The number of Ceph {{pool_name}} pool write bytes per second is {{threshold}} for 3 minutes."
    CephPool{{pool_name|replace(".", "")|replace("-", "")}}ReadOpsTooHigh:
      {%- set threshold = monitoring_pool.pool_read_ops_threshold|default('1000')|float %}
      if: >-
        ceph_pool_stats_read_op_per_sec{name="{{pool_name}}"} > {{threshold}}
      for: 3m
      labels:
        severity: warning
        service: ceph
      annotations:
        summary: "{{threshold}} Ceph pool read operations per second"
        description: "The number of Ceph {{pool_name}} pool read operations per second is {{threshold}} for 3 minutes."
    CephPool{{pool_name|replace(".", "")|replace("-", "")}}ReadBytesTooHigh:
      {%- set threshold = monitoring_pool.pool_read_bytes_threshold|default('70000000')|float %}
      if: >-
        ceph_pool_stats_read_bytes_sec{name="{{pool_name}}"} > {{threshold}}
      for: 3m
      labels:
        severity: warning
        service: ceph
      annotations:
        summary: "{{threshold}} Ceph pool read bytes per second"
        description: "The number of Ceph {{pool_name}} pool read bytes per second is {{threshold}} for 3 minutes."
    {%- endif %}
{%- endfor %}
{%- endif %}
{%- endif %}