X-Git-Url: https://git.openstreetmap.org./chef.git/blobdiff_plain/a4777dddfb2c769864094ec928d5bdcfdd411cc2..cc57b9e9a6f06eb2fdce5ed83aa5f9ed6cd90cad:/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 7a5353372..239731607 100644 --- a/cookbooks/prometheus/templates/default/alert_rules.yml.erb +++ b/cookbooks/prometheus/templates/default/alert_rules.yml.erb @@ -105,6 +105,15 @@ groups: alertgroup: database annotations: delay: "{{ $value | humanizeDuration }}" + - name: discourse + rules: + - alert: discourse job failure rate + expr: rate(discourse_job_failures[5m]) > 0 + for: 5m + labels: + alertgroup: discourse + annotations: + failure_rate: "{{ $value }} jobs/s" - name: dublin rules: - alert: pdu current draw @@ -611,7 +620,7 @@ groups: - name: statuscake rules: - alert: statuscake uptime check failing - expr: statuscake_uptime{status="down",paused="false"} > 0 + expr: statuscake_paused == 0 and statuscake_up == 0 for: 10m labels: alertgroup: statuscake