refactor: unique names for odh-controller alerts

opendatahub-io · Nov 27, 2024 · fab03b6 · fab03b6
1 parent 928b1c8
commit fab03b6
Show file tree

Hide file tree

Showing 2 changed files with 22 additions and 16 deletions.
diff --git a/config/monitoring/prometheus/apps/prometheus-configs.yaml b/config/monitoring/prometheus/apps/prometheus-configs.yaml
@@ -1180,11 +1180,11 @@ data:
     groups:
       - name: SLOs-probe_success_model_controller
         rules:
-        - alert: ODH Model Controller Probe Success Burn Rate
+        - alert: ODH Model Controller Probe Success 5m and 1h Burn Rate high
           annotations:
             message: 'High error budget burn for {{ $labels.instance }} (current value: {{ $value }}).'
             triage: "https://gitlab.cee.redhat.com/service/managed-tenants-sops/-/blob/main/RHODS/Model-Serving/rhods-odh-controller-probe-success-burn-rate.md"
-            summary: ODH Model Controller Probe Success Burn Rate
+            summary: ODH Model Controller Probe Success 5m and 1h Burn Rate high
           expr: |
             sum(probe_success:burnrate5m{instance=~"odh-model-controller"}) by (instance) > (14.40 * (1-0.98000))
             and
@@ -1193,11 +1193,11 @@ data:
           labels:
             severity: critical
             namespace: redhat-ods-applications
-        - alert: ODH Model Controller Probe Success Burn Rate
+        - alert: ODH Model Controller Probe Success 30m and 6h Burn Rate high
           annotations:
             message: 'High error budget burn for {{ $labels.instance }} (current value: {{ $value }}).'
             triage: "https://gitlab.cee.redhat.com/service/managed-tenants-sops/-/blob/main/RHODS/Model-Serving/rhods-odh-controller-probe-success-burn-rate.md"
-            summary: ODH Model Controller Probe Success Burn Rate
+            summary: ODH Model Controller Probe Success 30m and 6h Burn Rate high
           expr: |
             sum(probe_success:burnrate30m{instance=~"odh-model-controller"}) by (instance) > (6.00 * (1-0.98000))
             and
@@ -1206,11 +1206,11 @@ data:
           labels:
             severity: critical
             namespace: redhat-ods-applications
-        - alert: ODH Model Controller Probe Success Burn Rate
+        - alert: ODH Model Controller Probe Success 2h and 1d Burn Rate high
           annotations:
             message: 'High error budget burn for {{ $labels.instance }} (current value: {{ $value }}).'
             triage: "https://gitlab.cee.redhat.com/service/managed-tenants-sops/-/blob/main/RHODS/Model-Serving/rhods-odh-controller-probe-success-burn-rate.md"
-            summary: ODH Model Controller Probe Success Burn Rate
+            summary: ODH Model Controller Probe Success 2h and 1d Burn Rate high
           expr: |
             sum(probe_success:burnrate2h{instance=~"odh-model-controller"}) by (instance) > (3.00 * (1-0.98000))
             and

diff --git a/tests/prometheus_unit_tests/odh-model-controller-alerting.unit-tests.yaml b/tests/prometheus_unit_tests/odh-model-controller-alerting.unit-tests.yaml
@@ -20,7 +20,13 @@ tests:
         values: "0x60"
     alert_rule_test:
       - eval_time: 1h
-        alertname: ODH Model Controller Probe Success Burn Rate
+        alertname: ODH Model Controller Probe Success 5m and 1h Burn Rate high
+        exp_alerts: []
+      - eval_time: 1h
+        alertname: ODH Model Controller Probe Success 30m and 6h Burn Rate high
+        exp_alerts: []
+      - eval_time: 1h
+        alertname: ODH Model Controller Probe Success 2h and 1d Burn Rate high
         exp_alerts: []
 
   - interval: 1m
@@ -31,15 +37,15 @@ tests:
         values: "1+1x60"
     alert_rule_test:
       - eval_time: 2m
-        alertname: ODH Model Controller Probe Success Burn Rate
+        alertname: ODH Model Controller Probe Success 5m and 1h Burn Rate high
         exp_alerts:
           - exp_labels:
-              alertname: ODH Model Controller Probe Success Burn Rate
+              alertname: ODH Model Controller Probe Success 5m and 1h Burn Rate high
               namespace: "redhat-ods-applications"
               instance: "odh-model-controller"
               severity: critical
             exp_annotations:
-              summary: "ODH Model Controller Probe Success Burn Rate"
+              summary: "ODH Model Controller Probe Success 5m and 1h Burn Rate high"
               message: "High error budget burn for odh-model-controller (current value: 3)."
               triage: 'https://gitlab.cee.redhat.com/service/managed-tenants-sops/-/blob/main/RHODS/Model-Serving/rhods-odh-controller-probe-success-burn-rate.md'
 
@@ -51,15 +57,15 @@ tests:
         values: "1+1x60"
     alert_rule_test:
       - eval_time: 15m
-        alertname: ODH Model Controller Probe Success Burn Rate
+        alertname: ODH Model Controller Probe Success 30m and 6h Burn Rate high
         exp_alerts:
           - exp_labels:
-              alertname: ODH Model Controller Probe Success Burn Rate
+              alertname: ODH Model Controller Probe Success 30m and 6h Burn Rate high
               namespace: "redhat-ods-applications"
               instance: "odh-model-controller"
               severity: critical
             exp_annotations:
-              summary: "ODH Model Controller Probe Success Burn Rate"
+              summary: "ODH Model Controller Probe Success 30m and 6h Burn Rate high"
               message: "High error budget burn for odh-model-controller (current value: 16)."
               triage: 'https://gitlab.cee.redhat.com/service/managed-tenants-sops/-/blob/main/RHODS/Model-Serving/rhods-odh-controller-probe-success-burn-rate.md'
 
@@ -71,14 +77,14 @@ tests:
         values: "1+1x60"
     alert_rule_test:
       - eval_time: 1h
-        alertname: ODH Model Controller Probe Success Burn Rate
+        alertname: ODH Model Controller Probe Success 2h and 1d Burn Rate high
         exp_alerts:
           - exp_labels:
-              alertname: ODH Model Controller Probe Success Burn Rate
+              alertname: ODH Model Controller Probe Success 2h and 1d Burn Rate high
               namespace: "redhat-ods-applications"
               instance: "odh-model-controller"
               severity: warning
             exp_annotations:
-              summary: "ODH Model Controller Probe Success Burn Rate"
+              summary: "ODH Model Controller Probe Success 2h and 1d Burn Rate high"
               message: "High error budget burn for odh-model-controller (current value: 61)."
               triage: 'https://gitlab.cee.redhat.com/service/managed-tenants-sops/-/blob/main/RHODS/Model-Serving/rhods-odh-controller-probe-success-burn-rate.md'