admin: Add wordpress site monitoring with regexp checks

Related to swh/infra/sysadm-environment#4525

admin/websites-alerting.yaml

@@ -23,6 +23,50 @@ spec:
 ---
 apiVersion: monitoring.coreos.com/v1
 kind: PrometheusRule
+metadata:
+  labels:
+    app: website
+  name: websites-body-critical-alertmanager.rules
+  namespace: cattle-monitoring-system
+spec:
+  groups:
+  - name: critical-website-body.rules
+    rules:
+    - alert: WebsiteIncorrectBody
+      annotations:
+        description: The {{ $labels.domain }} website {{ $labels.instance }} has an incorrect body since more than 5m.
+        summary: A blackbox probe regex check is failing since more then 5mn
+      expr: |-
+        max_over_time(probe_failed_due_to_regex{level="critical"}[1m]) == 1
+      for: 5m
+      labels:
+        severity: critical
+        namespace: cattle-monitoring-system
+---
+apiVersion: monitoring.coreos.com/v1
+kind: PrometheusRule
+metadata:
+  labels:
+    app: website
+  name: websites-body-warning-alertmanager.rules
+  namespace: cattle-monitoring-system
+spec:
+  groups:
+  - name: warning-website-body.rules
+    rules:
+    - alert: WebsiteIncorrectBody
+      annotations:
+        description: The {{ $labels.domain }} website {{ $labels.instance }} has an incorrect body since more than 5m.
+        summary: A blackbox probe regex check is failing since more then 5mn
+      expr: |-
+        max_over_time(probe_failed_due_to_regex{level="critical"}[1m]) == 1
+      for: 5m
+      labels:
+        severity: warning
+        namespace: cattle-monitoring-system
+---
+apiVersion: monitoring.coreos.com/v1
+kind: PrometheusRule
 metadata:
   labels:
     app: website

admin/websites-monitoring.yaml

 ---
 apiVersion: monitoring.coreos.com/v1
 kind: Probe
+metadata:
+  name: production-wordpress-monitoring
+  namespace: cattle-monitoring-system
+  labels:
+    release: "rancher-monitoring"
+spec:
+  interval: 60s
+  jobName: production-websites-monitoring
+  module: swh_www
+  prober:
+    url: admin-blackbox-exporter-application-prometheus-blackbox-exporte:9115
+    path: /probe
+    scheme: http
+  targets:
+    staticConfig:
+      static:
+        - "https://www.softwareheritage.org"
+      labels:
+        level: critical
+        domain: production
+---
+apiVersion: monitoring.coreos.com/v1
+kind: Probe
 metadata:
   name: production-websites-monitoring
   namespace: cattle-monitoring-system
@@ -17,7 +40,6 @@ spec:
   targets:
     staticConfig:
       static:
-        - "https://www.softwareheritage.org"
         - "https://archive.softwareheritage.org"
         - "https://hedgedoc.softwareheritage.org"
       labels:

argocd-configuration/applications/admin/blacbox-exporter-application.yaml → argocd-configuration/applications/admin/blackbox-exporter-application.yaml

@@ -21,6 +21,17 @@ spec:
             enabled: true
           prometheusRule:
             enabled: true
+        config:
+          modules:
+            swh_www:
+              prober: http
+              timeout: 5s
+              http:
+                valid_http_versions: ["HTTP/1.1", "HTTP/2.0"]
+                follow_redirects: true
+                preferred_ip_protocol: "ip4"
+                fail_if_body_not_matches_regexp:
+                  - '<title>Software Heritage</title>'
   destination:
     server: https://rancher.euwest.azure.internal.softwareheritage.org/k8s/clusters/c-q2wd4
     namespace: cattle-monitoring-system