alerts: add check for healthy garage cluster

2024-09-10 14:26:26 +02:00 · 2024-09-10 14:26:26 +02:00 · faa71b7797
parent 21a1ae15cb
commit faa71b7797
1 changed files with 6 additions and 0 deletions
--- a/modules/prometheus/alert-rules.nix
+++ b/modules/prometheus/alert-rules.nix
@ -250,4 +250,10 @@ lib.mapAttrsToList
    #   description =
    #     "alertmanager: number of active silences has changed: {{$value}}";
    # };
+
+    garage_cluster_healthy = {
+      condition = "cluster_healthy = 0";
+      time = "15m";
+      description = "garage cluster on {{$labels.instance}} is not healthy: {{$labels.result}}!";
+    };
  })