Skip to Content
GuideRebalance ConfigurationNode Pool Configuration Guide

Node Pool Configuration Guide

Node Pool Configuration lets you split cluster capacity into multiple pools with different cost, placement, and migration policies. This guide covers the common node pool controls for AWS, AlibabaCloud, GKE, and AKS so you can shape capacity around workload requirements without losing operational safety.

Use NodePool Disruption Budgets to limit native voluntary node disruption by count, percentage, reason, or UTC restriction window. The same budget model applies across all supported providers.


AWS Node Provision Settings

The following parameters are available for AWS node provisioning:

ParameterDescriptionDefault
EnableControls whether this node pool can provision new nodes. CloudPilot also turns this off automatically when a pool enters decommissioning.true
Node ClassSelect the node class to use for this node pool.cloudpilot
Provision PriorityThe priority of this node pool.2
Instance FamilySpecifies instance types (e.g., t3, m5). Multiple comma-separated values are allowed.All instance families
Instance ArchCluster architecture (supports amd64, arm64). Automatically detected or manually configured.amd64
Capacity TypeNode purchase option: on-demand, spot, or both.spot, on-demand
Instance TagsKey-value tags for resource management (e.g., cloudpilot.ai/managed=true).cloudpilot.ai/managed=true
Zone IDDeployment zone (e.g., us-west-1a). Multiple zones enhance disaster recovery.All available zones
System Disk SizeOS disk size (adjustable).20GiB
Extra CPU/Memory AllocationReserved CPU (mCore) or memory (MiB) for burst tasks.0 (disabled)
Min/Max CPUMinimum/maximum CPU cores per node (0 = unlimited).0 (Unlimited)
Min/Max MemoryMinimum/maximum memory (MiB) per node (0 = unlimited).0 (Unlimited)
Enable GPU Node ProvisionActivates GPU node provisioning.Disabled
GPU Instance FamilyGPU instance types (e.g., p4d, p4). Multiple choices allowed.All GPU families
Min Graphic Cards/MemoryMinimum GPU cards or memory (MiB) per node.0 (Unlimited)

AlibabaCloud Node Provision Settings

The following parameters are available for AlibabaCloud node provisioning:

ParameterDescriptionDefault
EnableControls whether this node pool can provision new nodes. CloudPilot also turns this off automatically when a pool enters decommissioning.true
Node ClassSelect the node class to use for this node pool.cloudpilot
Provision PriorityThe priority of this node pool.2
Instance FamilySpecifies instance types (e.g., ecs.g5, ecs.r6). Multiple comma-separated values allowed.All instance families
Instance ArchCluster architecture (supports amd64, arm64). Automatically detected or manually configured.amd64
Capacity TypeNode purchase option: on-demand, spot, or both.spot, on-demand
Instance TagsKey-value tags for resource management (e.g., cloudpilot.ai/managed=true).cloudpilot.ai/managed=true
Image AliasOS image (supports ContainerOS for x86, AlibabaCloudLinux3 for x86/arm64).AlibabaCloudLinux3
Image IDCustom image ID (overrides Image Alias if specified).null
Zone IDDeployment zone (e.g., cn-beijing-a). Multiple zones supported.All available zones
System Disk SizeOS disk size (adjustable).20GiB
System Disk CategoriesDisk types (e.g., cloud_ssd, cloud_essd). Cheapest available is selected.cloud, cloud_ssd, cloud_efficiency, cloud_essd, cloud_auto, cloud_essd_entry
Min/Max CPUMinimum/maximum CPU cores per node (0 = unlimited).0 (Unlimited)
Min/Max MemoryMinimum/maximum memory (MiB) per node (0 = unlimited).0 (Unlimited)
Enable GPU Node ProvisionActivates GPU node provisioning.Disabled
GPU Instance FamilyGPU instance types (e.g., ecs.gn5i-c2g1, ecs.gn7i-c8g1). Multiple choices allowed.All GPU families
Min Graphic Cards/MemoryMinimum GPU cards or memory (MiB) per node.0 (Unlimited)

GKE Node Provision Settings

The following parameters are available for GKE node provisioning:

ParameterDescriptionDefault
EnableControls whether this node pool can provision new nodes. CloudPilot also turns this off automatically when a pool enters decommissioning.true
Node ClassSelect the GCENodeClass to use for this node pool.cloudpilot
Provision PriorityThe priority of this node pool.2
Instance FamilyLimits the allowed GCE families (for example n2, c3, or c3d). Multiple comma-separated values are allowed.All families that satisfy the remaining requirements
Instance ArchThe node CPU architecture.amd64
Capacity TypeNode purchase option: on-demand, spot, or both.on-demand
Zone IDDeployment zone (e.g., us-central1-a). Multiple zones are supported.All available zones
Min/Max CPUMinimum/maximum CPU cores per node (0 = unlimited).0 / 16
Min/Max MemoryMinimum/maximum memory in MiB per node (0 = unlimited).0 / 32768 MiB

AKS Node Provision Settings

The following parameters are available for AKS node provisioning. CloudPilot AI keeps optimized AKS capacity on user pools and uses AKSNodeClass as the provider-specific node class.

ParameterDescriptionDefault
EnableControls whether this node pool can provision new nodes. CloudPilot also turns this off automatically when a pool enters decommissioning.true
Node ClassSelect the AKSNodeClass to use for this node pool.cloudpilot
Provision PriorityThe priority of this node pool.2
Instance FamilyLimits the allowed Azure VM families (for example D, E, or F). Multiple comma-separated values are allowed.All families that satisfy the remaining requirements
Instance ArchThe node CPU architecture.amd64
Capacity TypeNode purchase option: on-demand, spot, or both.on-demand
Zone IDDeployment zone (for example 1 or 2). Multiple zones are supported.All available zones
Min/Max CPUMinimum/maximum CPU cores per node (0 = unlimited).0 / 16
Min/Max MemoryMinimum/maximum memory in MiB per node (0 = unlimited).0 / 32768 MiB

Delete a Node Pool Safely

On supported Phase 2 clusters, you can enable NodePool Decommission in the cluster Settings page before removing a node pool. This changes deletion from an immediate remove action into a drain-first workflow:

  1. Delete the node pool from the console.
  2. CloudPilot marks the pool as Decommissioning and stops it from provisioning any new nodes.
  3. Existing nodes in that pool keep running, so already-scheduled workloads are not interrupted immediately.
  4. After the remaining workloads and backing node capacity drain away, CloudPilot finalizes deletion automatically.
  • Make sure another enabled node pool can accept the workloads that still need to run.
  • Move or scale down workloads that are pinned to the target node pool.
  • Expect the decommissioning pool to stop handling any new scale-up requests.
  • Keep the pool visible until the final delete completes.

This workflow is useful when you want to retire a node pool gradually instead of dropping its backing capacity right away.

Adjust configurations based on workload requirements and validate changes in a staging environment. For further assistance, consult the CloudPilot Documentation  or contact support.

Last updated on