X-Git-Url: https://git.openstreetmap.org./chef.git/blobdiff_plain/cba7532d66e48bf49632eaf22d4ba9407b8182af..923a789051a29fbbce773c56505be4b78fcc4287:/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 a52901233..61a7370b1 100644
--- a/cookbooks/prometheus/templates/default/alert_rules.yml.erb
+++ b/cookbooks/prometheus/templates/default/alert_rules.yml.erb
@@ -114,13 +114,18 @@ groups:
           delay: "{{ $value | humanizeDuration }}"
   - name: fastly
     rules:
-      - alert: error rate
+      - alert: fastly error rate
         expr: sum(rate(fastly_rt_status_group_total{status_group="5xx"}[5m])) by (service_name, datacenter) / sum(rate(fastly_rt_status_group_total[5m])) by (service_name, datacenter) > 0.005
         for: 15m
         labels:
           alertgroup: fastly
         annotations:
           error_rate: "{{ $value | humanizePercentage }}"
+      - alert: fastly healthcheck failing
+        expr: fastly_healthcheck_status == 0
+        for: 5m
+        labels:
+          alertgroup: fastly
   - name: filesystem
     rules:
       - alert: readonly filesystem
@@ -417,6 +422,18 @@ groups:
           alertgroup: "{{ $labels.instance }}"
         annotations:
           queries: "{{ $value }}"
+  - name: raid
+    rules:
+      - alert: raid array degraded
+        expr: ohai_array_info{status="degraded"} > 0
+        for: 5m
+        labels:
+          alertgroup: "{{ $labels.instance }}"
+      - alert: raid disk failed
+        expr: ohai_disk_info{status="failed"} > 0
+        for: 5m
+        labels:
+          alertgroup: "{{ $labels.instance }}"
   - name: smart
     rules:
       - alert: smart failure