huawei-cloud-cce-metric-analyzer

Run the dispatcher with huawei_get_cce_pod_metrics_topN, passing region, cluster_id, and optionally namespace, label_selector, top_n, and hours. Results come from AOM Prometheus and include CPU, memory, and disk usage with critical/warning/normal status classification.

Use the dedicated tools huawei_get_ecs_metrics, huawei_get_elb_metrics, huawei_get_eip_metrics, and huawei_get_nat_gateway_metrics with the resource ID and region. These queries go through hcloud/CES and return structured metric keys such as cpu_util, m1_cps, and snat_connection.

Empty series usually mean the required AOM collection is not enabled. Controller-manager, scheduler, and etcd need their ServiceMonitors enabled, while autoscaler, ingress-controller, and GPU metrics need the corresponding PodMonitors; nginx request metrics also require nginx_ingress_controller_requests to be allowed.

No. All tools are read-only R3 queries that only read metrics and resource metadata. They never create, update, delete, restart, or scale Huawei Cloud or Kubernetes resources, and remediation is delegated to separate skills only when explicitly requested.

You need Huawei Cloud credentials via hcloud profile, environment variables, or explicit AK/SK parameters, plus IAM permissions such as cce:cluster:get, aom:metricsData:get, and ces:metricsData:get. AOM Prometheus queries use signed HTTPS requests and support temporary security tokens.

kubectl is used only for Kubernetes reads that AOM and hcloud cannot derive: Pod label_selector filtering, Ingress TLS Secret certificate checks, and LoadBalancer Service discovery for ELB/EIP association. Clusters without an external EIP require the kubectl cce plugin.