X-Git-Url: https://git.openstreetmap.org./chef.git/blobdiff_plain/c4f0a0e16117988b7f9488712b101bb08f8bc151..28326173066466bb20f54bad702b2885b64f925c:/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 5b2ec56ac..8c1aeee48 100644
--- a/cookbooks/prometheus/templates/default/alert_rules.yml.erb
+++ b/cookbooks/prometheus/templates/default/alert_rules.yml.erb
@@ -145,12 +145,12 @@ groups:
         annotations:
           error_rate: "{{ $value | humanizePercentage }}"
       - alert: fastly healthcheck failing
-        expr: count(fastly_healthcheck_status == 0) > 0
+        expr: count(fastly_healthcheck_status == 0) by (service) > 0
         for: 15m
         labels:
           alertgroup: fastly
       - alert: multiple fastly healthchecks failing
-        expr: count(fastly_healthcheck_status == 0) > 4
+        expr: count(fastly_healthcheck_status == 0) by (service) > 4
         for: 5m
         labels:
           alertgroup: fastly
@@ -247,12 +247,12 @@ groups:
   - name: juniper
     rules:
       - alert: juniper fan alarm
-        expr: sum_over_time(jnxOperatingState{jnxOperatingContentsIndex="4",jnxOperatingState!="running"}[6m]) > 0
+        expr: sum_over_time(jnxOperatingState{jnxOperatingContentsIndex="4",jnxOperatingState!~"running.*"}[6m]) > 0
         for: 5m
         labels:
           alertgroup: "{{ $labels.site }}"
       - alert: juniper power alarm
-        expr: sum_over_time(jnxOperatingState{jnxOperatingContentsIndex="2",jnxOperatingState!="running"}[6m]) > 0
+        expr: sum_over_time(jnxOperatingState{jnxOperatingContentsIndex="2",jnxOperatingState!~"running.*"}[6m]) > 0
         for: 5m
         labels:
           alertgroup: "{{ $labels.site }}"