diff --git a/modules/nixos/monitoring/alert-rules.nix b/modules/nixos/monitoring/alert-rules.nix index 9930c70..bd1c391 100644 --- a/modules/nixos/monitoring/alert-rules.nix +++ b/modules/nixos/monitoring/alert-rules.nix @@ -26,6 +26,12 @@ annotations.description = "{{$labels.host}} should have a running {{$labels.name}}"; }; + OfBorgEvalQueue = { + expr = ''ofborg_queue_evaluator_waiting > (2 * ofborg_queue_evaluator_consumers)''; + for = "1h"; + annotations.description = "ofborg evaluator queue is more than 2x the number of evaluators"; + }; + SmartErrors.expr = lib.mkForce ''smart_device_health_ok{enabled!="Disabled", host!="build01"} != 1''; }; }; diff --git a/modules/nixos/monitoring/telegraf.nix b/modules/nixos/monitoring/telegraf.nix index 4aa1582..a071e47 100644 --- a/modules/nixos/monitoring/telegraf.nix +++ b/modules/nixos/monitoring/telegraf.nix @@ -41,5 +41,8 @@ timeout = "10s"; }) hosts; + prometheus.urls = [ + "https://events.ofborg.org/prometheus.php" + ]; }; }