Scheduling CCE Workloads to CCI 2.0 Using AI CLI and Skills

Environment and Function Constraints

• Only CCE standard and CCE Turbo clusters that use the VPC network model are supported.
• The subnet where a cluster is located cannot overlap with 10.247.0.0/16, or the subnet conflicts with the Service CIDR block in the CCI namespace.
• DaemonSets are not supported.
• After the CCE Cloud Bursting Engine for CCI add-on is installed, a namespace named bursting-{cluster-ID} will be created in CCI and managed by the add-on. Do not use this namespace directly in CCI.
• For details about the Kubernetes versions and constraints, see Functions of CCE Cloud Bursting Engine for CCI.

IAM Permission Requirements

cce:cluster:get

cce:addon:list

cce:addon:create

cce:addon:update

vpcep:endpoint:create

vpcep:endpoint:list

vpcep:service:list

vpc:subnet:list

vpc:routetable:list

Step 1: Create a CCE Turbo Cluster

Create a CCE Turbo cluster and record the cluster ID and region. If a cluster that meets the requirements already exists, skip this step.

Step 2: Enable Auto Scaling Using AI CLI

cce-cci-bursting-deployer uses a preview-first design. Read operations (precheck, verify, discover, and diagnose) can be executed immediately, while a preview solution of write operations (VPC endpoint creation, add-on installation, and workload deployment) is returned first. The write operations are executed only after user confirmation.

Enter the following prompt in AI CLI to start the entire process:

I need to enable auto scaling of CCI 2.0 for my CCE cluster (cluster ID: xxx, region: cn-north-4).
Please follow the complete process: precheck → VPC endpoint creation → add-on installation → smoke test deployment → verification.
Preview each write operation before executing it.

The Skill will automatically proceed in the following sequence: (Each step involving a write operation will be paused for confirmation.)

1. Cluster precheck

   The Skill invokes huawei_precheck_cce_cci_bursting to automatically parse the cluster network topology, distinguishes the subnet roles of cci_subnet_id (Neutron UUID) and vpcep_subnet_id (VPC UUID), checks the status of the virtual-kubelet add-on, and performs NodeCheck to check the physical node's addon headroom.

2. Node capacity check (NodeCheck)

   If the precheck report indicates that physical node resources are insufficient, the Skill invokes huawei_check_cce_cci_node_capacity to view capacity details and preview the node pool scale-out solution. After user confirmation, the scale-out is performed.

3. VPC endpoint creation

   The Skill invokes huawei_ensure_cce_cci_vpcep to automatically discover and create VPC endpoints compatible with SWR and OBS. Existing VPC endpoints are automatically reused and will not be created again.

   

   The VPC endpoint for OBS requires accurate obs_endpoint_service_name. The value needs to be obtained from the Huawei Cloud service ticket and cannot be inferred from the public service name in a similar region. If this value is not obtained in advance, the Skill will be interrupted in this step and prompt you to provide the value. After obs_endpoint_service_name is added, the Skill continues to create the VPC endpoint for OBS and resumes the subsequent process.

4. Add-on installation

   The Skill invokes huawei_setup_cce_cci_bursting to install or update the virtual-kubelet add-on after confirming that the VPC endpoint dependencies are ready. It automatically parses and writes the region project ID. This operation is idempotent. The existing add-on is only updated, and will not be uninstalled and reinstalled.

5. Auto scaling readiness verification

   The Skill invokes huawei_verify_cce_cci_bursting to check whether the virtual-kubelet add-on and the bursting-node are ready. If the verification fails, the Skill invokes huawei_diagnose_cce_cci_bursting_addon to return a structured diagnosis report.

6. Smoke test deployment

   The Skill invokes huawei_discover_cce_cci_smoke_images to discover the tenant's image in SWR Basic Edition and then invokes huawei_deploy_cce_cci_smoke_workload to create a Deployment. The Deployment automatically adds the bursting.cci.io/burst-to-cci: enforce label to forcibly schedule the workload to CCI. If the image parameter is not specified, the discovered tenant image is automatically used.

7. Final verification

   The Skill invokes huawei_verify_cce_cci_bursting again to ensure that all pods are in the Running state on the CCI virtual node and are visible on the CCE console.

Step 3: Configure a ScheduleProfile Policy

After the basic auto scaling configuration is complete, you can create a ScheduleProfile to control workload scheduling.

Enter the following prompt in AI CLI to create a ScheduleProfile:

Help me create a ScheduleProfile in the default namespace.
Match the app=nginx label and set the policy to localPrefer.
The maximum number of local instances is 20, and the CCI scale-in priority is 10.

The label policy has a higher priority than the ScheduleProfile. If a pod has both the bursting.cci.io/burst-to-cci: off label and the enforce profile, the pod will not be scheduled to CCI.