ACK Pro preset control plane for ultra-large clusters-Container Service for Kubernetes(ACK)-阿里云帮助中心

The preset control plane pre-allocates dedicated resources for AI training and inference, ultra-large clusters, and mission-critical workloads, keeping API request concurrency and pod scheduling capacity at a high, predictable level.

ACK Pro clusters dynamically adjust the resource configurations of control plane components such as the API Server and etcd based on actual cluster load. This is suitable for most scenarios. However, for ultra-large clusters or scenarios with sudden high-concurrency traffic, the response latency of auto scaling can affect business continuity. For example, when you start large numbers of pods in batches for large-scale AI training, you may encounter scheduling delays, or pods may remain unscheduled. During peak hours, slow API responses can affect business continuity.

The ACK Pro preset control plane eliminates the uncertainty of auto scaling by dedicating control plane resources and establishing a baseline configuration for the API Server. Instead of relying on reactive mechanisms to catch up, this approach ensures that control plane performance is always predictable.

The ACK Pro preset control plane runs in parallel with the standard ACK Pro control plane. Enabling the preset control plane does not affect the operation of existing ACK managed clusters.

Use cases

AI training and inference
Large-scale AI training jobs can involve the concurrent creation and scheduling of tens of thousands of pods, and they often require computation to start immediately after scheduling. The auto scaling latency of a standard control plane can cause pods to remain unscheduled or interrupt the training process. A preset control plane ensures that the required scheduling capacity is ready the moment training starts.
Large-scale cluster deployments
For scenarios that consistently handle high API request concurrency, such as agent-based applications and microservices clusters, a preset control plane simplifies cluster operations for large clusters and reduces complexity through standardized level management.
Mission-critical web applications
For latency-sensitive core containerized workloads, control plane response latency can directly affect service availability. A preset control plane provides deterministic API request concurrency and pod scheduling rates to ensure business continuity.

Levels

Level specifications

The preset control plane is available in three levels. Each level defines the control plane capacity based on three core metrics.

Important

Access to the Pro 4XL level requires an allowlist. Contact customer support to request access.

Performance metrics	Pro XL	Pro 2XL	Pro 4XL
API request concurrency (Seats)①	3,900	7,800	15,600
Pod scheduling rate (pods/second)②	200	400	600
etcd database size (GB)③	16	16	16

Metric descriptions:

① API request concurrency (Seats): The maximum concurrent request processing capacity of the API Server, expressed in Seats. The preset control plane ensures consistent request handling capacity in high-concurrency scenarios by dedicating baseline resources to the API Server.
② Pod scheduling rate (pods/second): The rate at which the cluster scheduler assigns pods to nodes. The preset control plane guarantees a predictable scheduling rate through a fixed ResourceQuota.
③ etcd database size: The capacity for storing the full data of the cluster. The preset control plane uses separate storage for Events data to prevent high-frequency event writes from affecting core state operations.

Level switching rules

Upgrades and downgrades are supported between the standard Pro level and the preset control plane levels (Pro XL, Pro 2XL, and Pro 4XL), as well as between the preset levels themselves. For example, you can upgrade from Pro to Pro XL or from Pro XL to Pro 2XL.
You cannot switch to an ACK Basic cluster.
Automatic level switching is not supported. Once a level is selected, the control plane operates at that fixed specification to ensure predictable performance. To adjust the capacity for your workload, you must monitor control plane metrics and manually upgrade or downgrade the level.

Standard control plane compared with preset control plane

Dimension	Standard ACK Pro control plane	ACK Pro preset control plane
Control plane resource configuration	Auto scaling, dynamically adjusted by load	Pre-allocated fixed capacity, always sufficient
API request concurrency guarantee	Dynamic, capped at the Pro XL level	Each level has a defined Seats specification
Pod scheduling rate guarantee	Dynamic, capped at the Pro XL level	Each level has a defined pods/second value
etcd configuration	8 GB	16 GB
Billing method	Resource plan or pay-as-you-go	Pay-as-you-go only
Use cases	General workloads that can tolerate auto scaling latency	Large-scale, high-concurrency workloads that require predictable performance

Billing

The preset control plane uses the same billing logic as ACK Pro cluster.

Billing occurs on the hour. At the start of each hour, the system bills you based on the current control plane level of your cluster.
After you switch levels, the new billing rate takes effect from the beginning of the hour in which the switch occurred. For example, if you switch from Pro to Pro XL at 10:20, you are billed at the Pro XL rate for the entire 10:00 to 11:00 billing cycle.
Only the pay-as-you-go billing method is supported. Existing upfront resource plans for ACK Pro cluster apply only to standard ACK Pro clusters and not to the preset control plane levels.
For more information, see Cluster management fees.

Capacity planning and metric monitoring

To decide whether to switch levels and select the appropriate capacity, monitor the following dashboards. The procedure for opening each dashboard and interpreting its panels is described in ACK Pro Preset Control Plane Observability.

Core capacity metrics

On the Cluster Information page of your cluster, click the Cluster Monitoring tab. Focus on the following capacity metrics and compare them with the preset control plane Level specifications to select the right level:

API request concurrency: Check this metric against the Seats specification of each level to assess the concurrent processing pressure on the API Server.
Pod scheduling rate: Observe the number of pod scheduling requests processed per second to determine whether the scheduler is a bottleneck.
etcd database size: Monitor etcd storage usage to ensure that the data volume stays within the safe limits of the current level.

Resource usage

In the left-side navigation pane of the cluster details page, choose Operations > Prometheus Monitoring.
Click the Key Component Monitoring tab and switch between the add-on tabs to view details. Monitor the memory and CPU resource usage of the API Server, etcd, and scheduler components. If the usage level remains in the high state (utilization ≥ 80%) for a sustained period, the current control plane resources are under pressure. Upgrade to a higher level promptly.
For detailed metric descriptions, see the References.

Usage notes

Optimal level selection: To find the optimal level for your cluster, switch to a higher level, run a stress test that simulates peak loads, and observe the control plane's resource usage. Then, select the level that best fits the performance data.
Rollback limitation: The etcd database of a standard ACK Pro control plane has a capacity limit of 8 GB. If the etcd database usage exceeds 8 GB while your cluster runs in preset control plane mode, you must reduce the database size to below 8 GB before you revert to the standard ACK Pro control plane.
Data plane add-on scaling: After you enable the preset control plane, also monitor the resource usage of data plane add-ons such as CoreDNS and metrics-server. Scale these add-ons appropriately based on the actual load to prevent the data plane from becoming a performance bottleneck.
Cluster and workload tuning: Configure your clusters and workloads according to Cluster best practices to further improve cluster performance. When you plan and tune a large-scale cluster, follow Recommendations for using large-scale clusters.

Get started

You can configure a preset control plane level for a new cluster or an existing cluster.

To upgrade a cluster, see Upgrade a cluster.

New cluster

By default, the creation process is the same as for ACK Pro cluster. You must explicitly select a preset control plane level to enable this feature.

Log on to the ACK console. In the left-side navigation pane, choose Clusters and then click Create Kubernetes Cluster.
For more information, see Create an ACK managed Pro cluster.
In the Cluster Configurations step, set Cluster Specification to a preset control plane level (Pro XL, Pro 2XL, or Pro 4XL) as prompted.
After the cluster is created, the control plane runs with the specifications of the selected level.

Existing cluster

For an existing ACK Pro cluster, check its current resource usage on the control plane component monitoring dashboard. You can then upgrade the level based on resource usage alerts or in preparation for peak traffic.

On the ACK Clusters page, click the name of your cluster. In the left navigation pane, click Cluster Information.
Click the Basic Information tab. In the Basic Information section, find the Cluster Specification area and follow the prompts to enable or configure a preset control plane level.

References

Service-Level Agreement (SLA): The preset control plane feature adds new performance levels to ACK Pro clusters. These levels offer the same guarantees as a standard ACK Pro cluster. For more information, see Alibaba Cloud ACK Service-Level Agreement.
Monitoring of control plane components: