diff options
Diffstat (limited to 'roles/monitoring/prometheus/server/defaults/main/rules_coredns.yml')
| -rw-r--r-- | roles/monitoring/prometheus/server/defaults/main/rules_coredns.yml | 29 |
1 files changed, 29 insertions, 0 deletions
diff --git a/roles/monitoring/prometheus/server/defaults/main/rules_coredns.yml b/roles/monitoring/prometheus/server/defaults/main/rules_coredns.yml new file mode 100644 index 00000000..126a7ba4 --- /dev/null +++ b/roles/monitoring/prometheus/server/defaults/main/rules_coredns.yml @@ -0,0 +1,29 @@ +--- +prometheus_server_rules_coredns_extra: [] +prometheus_server_rules_coredns: + - alert: CorednsPanicCount + expr: increase(coredns_panics_total[15m]) > 0 + for: 0m + labels: + severity: critical + annotations: + summary: CoreDNS Panic (instance {{ '{{' }} $labels.instance {{ '}}' }}) + description: "Number of CoreDNS panics encountered has been increasing in the last 15 minutes\n VALUE = {{ '{{' }} $value {{ '}}' }}\n LABELS = {{ '{{' }} $labels {{ '}}' }}" + + - alert: CorednsFailedReloadCount + expr: increase(coredns_reload_failed_total[15m]) > 0 + for: 0m + labels: + severity: critical + annotations: + summary: CoreDNS reload failed (instance {{ '{{' }} $labels.instance {{ '}}' }}) + description: "Number of CoreDNS failed reloads has been increasing in the last 15 minutes\n VALUE = {{ '{{' }} $value {{ '}}' }}\n LABELS = {{ '{{' }} $labels {{ '}}' }}" + + - alert: CorednsBrokenForwardHealthchecks + expr: increase(coredns_forward_healthcheck_broken_total[15m]) > 0 + for: 0m + labels: + severity: warning + annotations: + summary: CoreDNS broken forward healthchecks (instance {{ '{{' }} $labels.instance {{ '}}' }}) + description: "Number of CoreDNS broken forward healthchecks has been increasing in the last 15 minutes\n VALUE = {{ '{{' }} $value {{ '}}' }}\n LABELS = {{ '{{' }} $labels {{ '}}' }}" |