X-Git-Url: https://git.openstreetmap.org./chef.git/blobdiff_plain/8a772a64629eb2e7fe271191daea7579a35cdf80..88f820a038a839a687db7374235eca3003790296:/cookbooks/prometheus/templates/default/alert_rules.yml.erb?ds=inline

diff --git a/cookbooks/prometheus/templates/default/alert_rules.yml.erb b/cookbooks/prometheus/templates/default/alert_rules.yml.erb
index 2ab437c30..396de8de4 100644
--- a/cookbooks/prometheus/templates/default/alert_rules.yml.erb
+++ b/cookbooks/prometheus/templates/default/alert_rules.yml.erb
@@ -3,6 +3,13 @@
 groups:
   - name: amsterdam
     rules:
+      - alert: uplink
+        expr: ifOperStatus{site="amsterdam",ifName=~"ge-[01]/2/2"} != 1
+        for: 6m
+        labels:
+          alertgroup: "amsterdam"
+        annotations:
+          status: "{{ $value }}"
       - alert: pdu current draw
         expr: rPDU2PhaseStatusCurrent{site="amsterdam",rPDU2PhaseStatusIndex="1"} / 10 > 28
         for: 6m
@@ -109,6 +116,13 @@ groups:
           failure_rate: "{{ $value }} jobs/s"
   - name: dublin
     rules:
+      - alert: uplink
+        expr: ifOperStatus{site="dublin",ifName=~"ge-[01]/2/2"} != 1
+        for: 6m
+        labels:
+          alertgroup: "dublin"
+        annotations:
+          status: "{{ $value }}"
       - alert: pdu current draw
         expr: rPDU2PhaseStatusCurrent{site="dublin",rPDU2PhaseStatusIndex="1"} / 10 > 28
         for: 6m
@@ -249,10 +263,12 @@ groups:
   - name: juniper
     rules:
       - alert: juniper cpu alarm
-        expr: jnxOperatingCPU{jnxOperatingContentsIndex="7"} > 30
+        expr: jnxOperating5MinLoadAvg{jnxOperatingContentsIndex="9"} / 200 > 0.5
         for: 5m
         labels:
           alertgroup: "{{ $labels.site }}"
+        annotations:
+          load_average: "{{ $value | humanizePercentage }}"
       - alert: juniper fan alarm
         expr: jnxOperatingState{jnxOperatingContentsIndex="4",jnxOperatingState!~"running.*"} > 0
         for: 5m
@@ -271,7 +287,7 @@ groups:
         labels:
           alertgroup: "{{ $labels.instance }}"
       - alert: exim queue length
-        expr: exim_queue > exim_queue_limit
+        expr: exim_queue > ignoring(job) exim_queue_limit
         for: 60m
         labels:
           alertgroup: mail