chore: adjust prometheus rule to use Accepted and Programmed conditions

Author: OscarLlamas6

config/telemetry/nso-slo/prometheus-rules.yaml

@@ -6,26 +6,34 @@ metadata:
 spec:
   groups:
   - name: nso-slo
-    rules:
-    - alert: GatewayStuckNotReadySLOViolation
+    interval: 30s
+    rules:  
+    - alert: GatewayNotReadySLOViolation
       expr: |
         (time() - datum_cloud_networking_gateway_created) > 60
-        unless ignoring(type) (datum_cloud_networking_gateway_status{type="Ready"} == 1)
+        unless ignoring(type) (
+          (datum_cloud_networking_gateway_status{type="Accepted"} == 1) and 
+          (datum_cloud_networking_gateway_status{type="Programmed"} == 1)
+        )
       for: 0s
       labels:
         severity: critical
         slo_violation: "true"
       annotations:
-        summary: "Gateway {{ $labels.name }} is stuck for over 60 seconds without reaching Ready status"
-        description: "Gateway {{ $labels.name }} in namespace {{ $labels.namespace }} has been in creation/transition state for {{ $value }} seconds without reaching Ready status, which exceeds the 60-second SLO threshold."
+        summary: "Gateway {{ $labels.name }} is taking longer than 60 seconds to reach Ready status"
+        description: "Gateway {{ $labels.name }} in namespace {{ $labels.namespace }} has been in creation state for {{ $value }} seconds without reaching Ready status (Accepted=True AND Programmed=True), which exceeds the 60-second SLO threshold."
 
-    - alert: GatewayNotReadySLOViolation
+    - alert: GatewayDegradedSLOViolation
       expr: |
-        (time() - datum_cloud_networking_gateway_status_condition_last_transition_time{type="Ready", status="False"}) > 60
+        (
+          (time() - datum_cloud_networking_gateway_status_condition_last_transition_time{type="Accepted", status="False"}) > 60
+        ) or (
+          (time() - datum_cloud_networking_gateway_status_condition_last_transition_time{type="Programmed", status="False"}) > 60
+        )
       for: 0s
       labels:
         severity: critical
         slo_violation: "true"
       annotations:
-        summary: "Gateway {{ $labels.name }} has been not Ready for over 60 seconds"
-        description: "Gateway {{ $labels.name }} in namespace {{ $labels.namespace }} has been in not Ready state for {{ $value }} seconds, which exceeds the 60-second SLO threshold."
+        summary: "Gateway {{ $labels.name }} has been degraded for over 60 seconds"
+        description: "Gateway {{ $labels.name }} in namespace {{ $labels.namespace }} has been in a degraded state (Accepted=False or Programmed=False) for {{ $value }} seconds without recovering, which exceeds the 60-second SLO threshold."