Fix error in rasdaemon alerts

[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 2920d2b16b2dc3f3bfb0c6b99d3df1ab1bc8b5ac..23b94727cfc6e47b929467f01baab325559653a0 100644 (file)
--- a/cookbooks/prometheus/templates/default/alert_rules.yml.erb
+++ b/cookbooks/prometheus/templates/default/alert_rules.yml.erb
@@ -447,19 +447,19 @@ groups:
   - name: rasdaemon
     rules:
       - alert: memory controller errors
   - name: rasdaemon
     rules:
       - alert: memory controller errors

-        expr: increase(rasdaemon_mc_events_total) > 0
+        expr: increase(rasdaemon_mc_events_total[1m]) > 0

         for: 0m
         labels:
           alertgroup: "{{ $labels.instance }}"
         annotations:
           new_errors: "{{ $value }}"
       - alert: pcie aer errors
         for: 0m
         labels:
           alertgroup: "{{ $labels.instance }}"
         annotations:
           new_errors: "{{ $value }}"
       - alert: pcie aer errors

-        expr: increase(rasdaemon_aer_events_total) > 0
+        expr: increase(rasdaemon_aer_events_total[1m]) > 0

         for: 0m
         labels:
           alertgroup: "{{ $labels.instance }}"
         annotations:
         for: 0m
         labels:
           alertgroup: "{{ $labels.instance }}"
         annotations:

-          new_errors: "{{ $value }}"
+          new_ercrors: "{{ $value }}"

   - name: smart
     rules:
       - alert: smart failure
   - name: smart
     rules:
       - alert: smart failure