From 7da7c10f1d812129647b92849c2ab4d8c20b0a5f Mon Sep 17 00:00:00 2001
From: Christian Pointner <equinox@spreadspace.org>
Date: Mon, 29 Jul 2024 20:32:54 +0200
Subject: promethues: add coredns

---
 .../prometheus/server/defaults/main/main.yml       |  1 +
 .../server/defaults/main/rules_coredns.yml         | 29 ++++++++++++++++++++++
 2 files changed, 30 insertions(+)
 create mode 100644 roles/monitoring/prometheus/server/defaults/main/rules_coredns.yml

(limited to 'roles/monitoring/prometheus/server')

diff --git a/roles/monitoring/prometheus/server/defaults/main/main.yml b/roles/monitoring/prometheus/server/defaults/main/main.yml
index 9f867568..e94aaaf5 100644
--- a/roles/monitoring/prometheus/server/defaults/main/main.yml
+++ b/roles/monitoring/prometheus/server/defaults/main/main.yml
@@ -41,6 +41,7 @@ prometheus_server_rules:
   nftables: "{{ prometheus_server_rules_nftables + prometheus_server_rules_nftables_extra }}"
   whawty-nginx-sso: "{{ prometheus_server_rules_whawty_nginx_sso + prometheus_server_rules_whawty_nginx_sso_extra }}"
   mosquitto: "{{ prometheus_server_rules_mosquitto + prometheus_server_rules_mosquitto_extra }}"
+  coredns: "{{ prometheus_server_rules_coredns + prometheus_server_rules_coredns_extra }}"
 
 # prometheus_server_alertmanager:
 #   url: "127.0.0.1:9093"
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 {{ '}}' }}"
-- 
cgit v1.2.3