Refactor alert routing logic for slack and opsgenie.

Streamlined the handling of alert routes by consolidating redundant match conditions and modifying specific drop logic. This improves clarity and reduces complexity, particularly for "Unhealthy" and failure-related event filtering. Applied consistent logic across Slack and Opsgenie configurations.
pagopa · Jan 14, 2025 · a72be9e · a72be9e
1 parent 8e1de5c
commit a72be9e
Showing 1 changed file with 21 additions and 51 deletions.
diff --git a/kubernetes_event_exporter/templates/default.tftpl b/kubernetes_event_exporter/templates/default.tftpl
@@ -38,64 +38,34 @@ config:
       - match:
         - receiver: dump
       %{ if enable_slack }
-      - match:
+      - drop:
         - reason: "Unhealthy"
-          message: ".*OOMKilled.*"
-          type: "Warning|Error"
-        route:
+        - reason: "FailedToUpdateEndpoint"
+        - reason: "FailedScheduling"
+        - reason: "EgressBlocked"
+        - kind: "HorizontalPodAutoscaler"
+        - kind: "ScaledObjectCheckFailed"
+        match:
           - receiver: "slack"
-
-      - match:
-        - reason: "Unhealthy"
-          message: ".*CrashLoopBackOff.*"
-          type: "Warning|Error"
-        route:
+            type: "Warning"
           - receiver: "slack"
-
-      - match:
-        - drop:
-            - reason: "Unhealthy"
-              message: ".*Readiness probe failed.*"
-
-      - match:
-        - drop:
-          - reason: "Unhealthy"
-            message: ".*Liveness probe failed.*"
-
-      - match:
-        - reason: "Failed"
-          type: "Warning|Error"
-        route:
-          - receiver:
+            reason: "*Failed*"
+          - receiver: "slack"
+            reason: "NotTriggerScaleUp"
       %{ endif }
       %{ if enable_opsgenie }
-      - match:
+      - drop:
         - reason: "Unhealthy"
-          message: ".*CrashLoopBackOff.*"
-          type: "Warning|Error"
-        route:
+        - reason: "FailedToUpdateEndpoint"
+        - reason: "FailedScheduling"
+        - reason: "EgressBlocked"
+        - kind: "HorizontalPodAutoscaler"
+        - kind: "ScaledObjectCheckFailed"
+        match:
           - receiver: "${opsgenie_receiver_name}"
-
-      - match:
-        - reason: "Unhealthy"
-          message: ".*OOMKilled.*"
-          type: "Warning|Error"
-        route:
+            type: "Warning"
           - receiver: "${opsgenie_receiver_name}"
-
-      - match:
-        - drop:
-            - reason: "Unhealthy"
-              message: ".*Readiness probe failed.*"
-
-      - match:
-        - drop:
-          - reason: "Unhealthy"
-            message: ".*Liveness probe failed.*"
-
-      - match:
-        - reason: "Failed"
-          type: "Warning|Error"
-        route:
+            reason: "*Failed*"
           - receiver: "${opsgenie_receiver_name}"
+            reason: "NotTriggerScaleUp"
       %{ endif }