livrare lot 2

backend/observability/alertmanager/alertmanager.yml

@@ -0,0 +1,62 @@
+global:
+  resolve_timeout: 5m
+  # Configure SMTP for email alerts
+  smtp_from: 'alerts@didi365.eu'
+  smtp_smarthost: 'smtp.sendgrid.net:587'
+  smtp_auth_username: 'apikey'
+  # smtp_auth_password set via env $SENDGRID_API_KEY in compose
+
+route:
+  group_by: ['alertname', 'cluster', 'service']
+  group_wait: 30s
+  group_interval: 5m
+  repeat_interval: 12h
+  receiver: 'default'
+  routes:
+    - matchers:
+        - severity = critical
+      receiver: 'critical'
+      group_wait: 10s
+      repeat_interval: 1h
+
+    - matchers:
+        - team = revenue
+      receiver: 'revenue-team'
+
+    - matchers:
+        - team = ai-platform
+      receiver: 'ai-team'
+
+receivers:
+  - name: 'default'
+    email_configs:
+      - to: 'office@clossers.com'
+        send_resolved: true
+        headers:
+          Subject: '[DiDi] {{ .Status | toUpper }}: {{ .GroupLabels.alertname }}'
+
+  - name: 'critical'
+    email_configs:
+      - to: 'office@clossers.com'
+        send_resolved: true
+        headers:
+          Subject: '[DiDi CRITICAL] {{ .GroupLabels.alertname }}'
+    # Add webhook for Slack/Teams when ready:
+    # webhook_configs:
+    #   - url: 'https://hooks.slack.com/services/T.../B.../...'
+
+  - name: 'revenue-team'
+    email_configs:
+      - to: 'office@clossers.com'
+
+  - name: 'ai-team'
+    email_configs:
+      - to: 'office@clossers.com'
+
+inhibit_rules:
+  # Suppress non-critical alerts if a critical alert is already firing for same service
+  - source_matchers:
+      - severity = critical
+    target_matchers:
+      - severity = warning
+    equal: ['alertname', 'cluster', 'service']