Disable Workload Uploading
If you require a high level of confidentiality for your cluster information and do not permit transmitting details such as Namespace and Workload Name to CloudPilot AI, you can disable workload information uploading by following the steps outlined in this document.
Disable when installing CloudPilot AI
You must disable this capability when deploying CloudPilot AI; otherwise, by default, workload information will be automatically uploaded to CloudPilot AI.
When deploying Phase 1, you need to manually set the DISABLE_WORKLOAD_UPLOADING environment variable to true:
Later, when deploying Phase 2, make sure to set the DISABLE_WORKLOAD_UPLOADING environment variable to true before starting the deployment.
export DISABLE_WORKLOAD_UPLOADING=trueWorkload Configuration
Once workload information uploading is disabled, you will no longer be able to use the features provided on the Workload Configuration page, such as setting Spot Friendly, Min Non-Spot Replicas, or Rebalanceable for workloads.
However, you can still configure these settings using labels and annotations.
For more details on the available configuration options, please refer to the Workload Configuration Guide.
Rebalanceable Configuration
If you want a specific workload to be completely unaffected by CloudPilot AI, you can add the following annotation to the target workload:
karpenter.sh/do-not-disrupt: "true"Spot Friendly Configuration
If you don’t want a specific workload to be scheduled onto Spot instances, you can add the following label to the target workload:
workload.cloudpilot.ai/spot-friendly: "false"Min Non-Spot Replicas Configuration
If you want to ensure that at least X replicas of a workload are scheduled onto non-Spot instances, you can add the following label to the target workload:
workload.cloudpilot.ai/min-nonspot: "2"Dashboard
When workload information uploading is disabled, you will still be able to view aggregated workload data; however, detailed information for individual workloads will no longer be available.
We will continue to upload the UID of the workload and namespace (instead of their actual names) to CloudPilot AI, so that you can still see the overall workload status summary.
Optimize Simulation
When workload information uploading is disabled, you can still customize the scheduling behavior of specific workloads using labels or annotations. This allows you to view simulated scheduling results during Phase 1 based on those custom configurations.