Fix for prometheus alerts for luminous/nautilus different metrics names.
Related-Prod: PROD-35115
Change-Id: If29bfbe6beb79a41171d3ad03d5423157b29fcb9
diff --git a/ceph/meta/prometheus.yml b/ceph/meta/prometheus.yml
index 2969b6f..04a4362 100644
--- a/ceph/meta/prometheus.yml
+++ b/ceph/meta/prometheus.yml
@@ -156,6 +156,7 @@
{%- else -%}
{%- if mon is defined and mon.get('enabled') %}
+ {%- set pool_used_space = 'ceph_pool_bytes_used' if common.version in ['luminous'] else 'ceph_pool_stored' -%}
{%- raw %}
server:
recording:
@@ -366,7 +367,7 @@
CephPool{{pool_name|replace(".", "")|replace("-", "")}}SpaceUsageWarning:
{%- set threshold = monitoring_pool.pool_space_used_utilization_warning_threshold|default('0.75')|float %}
if: >-
- ceph_pool_stored / (ceph_pool_stored + ceph_pool_max_avail) * on(pool_id) group_left(name) ceph_pool_metadata{name="{{pool_name}}"} > {{threshold}}
+ {{pool_used_space}} / ({{pool_used_space}} + ceph_pool_max_avail) * on(pool_id) group_left(name) ceph_pool_metadata{name="{{pool_name}}"} > {{threshold}}
for: 3m
labels:
severity: warning
@@ -377,7 +378,7 @@
CephPool{{pool_name|replace(".", "")|replace("-", "")}}SpaceUsageCritical:
{%- set threshold = monitoring_pool.pool_space_used_critical_threshold|default('0.85')|float %}
if: >-
- ceph_pool_stored / (ceph_pool_stored + ceph_pool_max_avail) * on(pool_id) group_left(name) ceph_pool_metadata{name="{{pool_name}}"} > {{threshold}}
+ {{pool_used_space}} / ({{pool_used_space}} + ceph_pool_max_avail) * on(pool_id) group_left(name) ceph_pool_metadata{name="{{pool_name}}"} > {{threshold}}
for: 3m
labels:
severity: minor
@@ -390,7 +391,7 @@
{%- set threshold = monitoring.prediction_threshold %}
{%- set space_threshold = monitoring_pool.space_threshold %}
if: >-
- predict_linear(ceph_pool_stored[{{threshold}}d], {{threshold}} * 86400) * on(pool_id) group_left(name) ceph_pool_metadata{name="{{pool_name}}"} > (ceph_pool_stored + ceph_pool_max_avail) * {{space_threshold}} * on(pool_id) group_left(name) ceph_pool_metadata{name="{{pool_name}}"}
+ predict_linear({{pool_used_space}}[{{threshold}}d], {{threshold}} * 86400) * on(pool_id) group_left(name) ceph_pool_metadata{name="{{pool_name}}"} > ({{pool_used_space}} + ceph_pool_max_avail) * {{space_threshold}} * on(pool_id) group_left(name) ceph_pool_metadata{name="{{pool_name}}"}
for: 30m
labels:
severity: minor