v-m: alerts: Add Longhorn alerts

2024-08-17 10:51:13 -05:00 · 2024-08-17 10:51:13 -05:00 · 809676f691
parent 9977bb3de4
commit 809676f691
1 changed files with 11 additions and 0 deletions
--- a/victoria-metrics/alerts.yml
+++ b/victoria-metrics/alerts.yml
@ -148,3 +148,14 @@ groups:
    expr: >-
      {__name__=~"collectd_.*_temperature", sensors!~"i350bb.*"} > 80
    for: 10m
+
+- name: Longhorn
+  rules:
+  - alert: Degraded Volumes
+    expr: >-
+      count(longhorn_volume_robustness==2) > 0
+    for: 1h
+  - alert: Faulted Volumes
+    expr: >-
+      count(longhorn_volume_robustness==3) > 0
+    for: 5m