Add alerts for juniper alarms

[chef.git] / cookbooks / prometheus / templates / default / alert_rules.yml.erb

diff --git a/cookbooks/prometheus/templates/default/alert_rules.yml.erb b/cookbooks/prometheus/templates/default/alert_rules.yml.erb
index 899c765e45efa858401f84355115bc3ed54b9166..d717e4f0aeedacf3f5c8fc628485c9b945eedc05 100644 (file)
--- a/cookbooks/prometheus/templates/default/alert_rules.yml.erb
+++ b/cookbooks/prometheus/templates/default/alert_rules.yml.erb
@@ -4,7 +4,7 @@ groups:
   - name: amsterdam
     rules:
       - alert: uplink
-        expr: ifOperStatus{site="amsterdam",ifName=~"ge-[01]/2/2"} != 1
+        expr: junos_interface_up{site="amsterdam",name=~"ge-[01]/2/2"} != 1
         for: 6m
         labels:
           alertgroup: "amsterdam"
@@ -117,7 +117,7 @@ groups:
   - name: dublin
     rules:
       - alert: uplink
-        expr: ifOperStatus{site="dublin",ifName=~"ge-[01]/2/2"} != 1
+        expr: junos_interface_up{site="dublin",name=~"ge-[01]/2/2"} != 1
         for: 6m
         labels:
           alertgroup: "dublin"
@@ -272,20 +272,34 @@ groups:
           alertgroup: "{{ $labels.instance }}"
   - name: juniper
     rules:
+      - alert: juniper red alarms
+        expr: juniper_alarms_red_count > 0
+        for: 5m
+        labels:
+          alertgroup: "{{ $labels.site }}"
+        annotations:
+          alarm_count: "{{ $value }} alarms"
+      - alert: juniper yellow alarms
+        expr: juniper_alarms_yellow_count > 0
+        for: 5m
+        labels:
+          alertgroup: "{{ $labels.site }}"
+        annotations:
+          alarm_count: "{{ $value }} alarms"
       - alert: juniper cpu alarm
-        expr: jnxOperating5MinLoadAvg{jnxOperatingContentsIndex="9"} / 200 > 0.5
+        expr: junos_route_engine_load_average_five / 2 > 0.5
         for: 5m
         labels:
           alertgroup: "{{ $labels.site }}"
         annotations:
           load_average: "{{ $value | humanizePercentage }}"
       - alert: juniper fan alarm
-        expr: jnxOperatingState{jnxOperatingContentsIndex="4",jnxOperatingState!~"running.*"} > 0
+        expr: junos_environment_fan_up != 1
         for: 5m
         labels:
           alertgroup: "{{ $labels.site }}"
       - alert: juniper power alarm
-        expr: jnxOperatingState{jnxOperatingContentsIndex="2",jnxOperatingState!~"running.*"} > 0
+        expr: junos_environment_power_up != 1
         for: 5m
         labels:
           alertgroup: "{{ $labels.site }}"