From: Tom Hughes Date: Sun, 14 Nov 2021 16:25:05 +0000 (+0000) Subject: Add some SMART alert rules X-Git-Url: https://git.openstreetmap.org./chef.git/commitdiff_plain/d1633aaa414c1d6126ed0d1a60ee33537ccd7c21 Add some SMART alert rules --- diff --git a/cookbooks/prometheus/templates/default/alert_rules.yml.erb b/cookbooks/prometheus/templates/default/alert_rules.yml.erb index fd25f2523..4511e1eac 100644 --- a/cookbooks/prometheus/templates/default/alert_rules.yml.erb +++ b/cookbooks/prometheus/templates/default/alert_rules.yml.erb @@ -204,6 +204,20 @@ groups: alertgroup: "{{ $labels.instance }}" annotations: queries: "{{ $value }}" + - name: smart + rules: + - alert: smart failure + expr: smart_health_status == 0 + for: 60m + labels: + alertgroup: "{{ $labels.instance }}" + - alert: smart ssd wearout approaching + expr: smart_percentage_used >= 90 + for: 60m + labels: + alertgroup: "{{ $labels.instance }}" + annotations: + percentage_used: "{{ $value | humanizePercentage }}" - name: ssl rules: - alert: ssl certificate probe failed