/google-cloud-slo-alert-configuration

Service Metrics Mapping (SLI Types)

When determining the ServiceLevelIndicator (SLI), map the user's Google Cloud resource and desired Google Cloud metric type to the correct Prometheus metric type.

IMPORTANT DOCUMENTATION LINK: If the user's Google Cloud resource is not listed below, you MUST use the read_url_content tool to fetch the live infrastructure list from: https://docs.cloud.google.com/monitoring/docs/application-monitoring-services.md.txt

Cloud Run (run.googleapis.com)

• Availability (RatioSli):
  • total_metric: run_googleapis_com:request_count
  • bad_metric filter: response_code_class="5xx"
• Latency (DistributionSli):
  • distribution_metric: run_googleapis_com:request_latencies

Google Kubernetes Engine (GKE)

Scenario A: Registered in App Hub (Preferred)

For GKE deployments, StatefulSets, and DaemonSets registered in App Hub, use the automatically generated SLIs:

• Availability (RatioSli):
  • total_metric: service:server_request_count (mapping for service/server/request_count)
  • bad_metric filter: response_code_class="5xx"
• Latency (DistributionSli):
  • distribution_metric: service:server_response_latencies (mapping for service/server/response_latencies)

Scenario B: Custom Telemetry (OTLP / Prometheus)

If the user's apps, services, and workloads are not registered in App Hub, you MUST ask the user for their exact metric names and failure filters if they are not provided in the original request.

Examples of potential metrics:

• Prometheus / Istio: istio_requests_total filter: response_code=~"5.."
• OTLP: http.server.request.duration filter: http.response.status_code=~"5.."

Vertex AI (aiplatform.googleapis.com)

• Availability (RatioSli):
  • total_metric: aiplatform_googleapis_com:reasoning_engine_request_count (Example for Reasoning Engine)
  • bad_metric filter: response_code=~"5.."