This topic describes the limits that apply when you use Container Service for Kubernetes (ACK) clusters.
Before you use ACK, take note of the following limits:
- You must pass real-name verification before you can create an ACK cluster.
- You cannot perform the following operations on ACK clusters after the clusters are
- Change the virtual private cloud (VPC).
- Change an ACK managed cluster to an ACK dedicated cluster, or change an ACK Pro cluster to an ACK standard cluster.
- Change the network plug-in.
- Change the volume plug-in.
- Migrate applications across namespaces.
- The following limits apply to Elastic Compute Service (ECS) instances in ACK clusters:
- ECS instances support the pay-as-you-go and subscription billing methods. Other resources, such as Server Load Balancer (SLB) instances, support only the pay-as-you-go billing method. You can change the billing method from pay-as-you-go to subscription in the ECS console.
- When you create clusters, expand clusters, or perform auto scaling for clusters, you may fail to create nodes due to resource quota limits or insufficient inventory.
- If you want to create subscription ECS instances when you create a cluster, you may
fail to create the ECS instances due to resource quota limits or insufficient inventory.
After a subscription ECS instance is created, you cannot release the ECS instance
before the subscription expires.
Note To avoid these issues, we recommend that you create pay-as-you-go ECS instances when you create a cluster. You can change the billing method from pay-as-you-go to subscription in the ECS console based on your requirements.
- You must select ECS instances with at least 4 vCPUs and 8 GiB of memory.
- Limits on access to control plane components:
Bandwidth limits apply when you access the control pane components, including kube-apiserver and etcd, of a cluster by calling API operations or by using the CLI. If you read a large number of cluster events at a time, bandwidth throttling may be triggered and you may fail to read the events. We recommend that you query cluster events by using the Kubernetes event center. Alternatively, you can add paging parameters, such as
--chunk-size=500, to the API request or command lines to reduce the number of events to be read. If bandwidth throttling is frequently triggered in an ACK standard cluster, we recommend that you migrate workloads from the cluster to an ACK Pro cluster.
For more information about the Kubernetes event center, see Scenario 1: Use node-problem-detector with the Kubernetes event center of Log Service to sink cluster events.
For more information about how to migrate workloads from an ACK standard cluster to an ACK Pro cluster, see Hot migration from ACK standard clusters to ACK Pro clusters.
|Cluster Type||Maximum number of clusters within an Alibaba Cloud account||Maximum number of node pools in a cluster①||Maximum number of nodes in a cluster||Maximum number of pods on a node②||Adjustable|
①To make a quota increase for node pools take effect, you must submit a ticket to request a quota increase for scaling groups.
②The quota of pods on a node takes effect only in Flannel network mode and cannot be increased. In Terway network mode, the quota of pods on a node is determined by the number of IP addresses that can be allocated by the node.
③ASK clusters do not contain nodes. The quota indicates the maximum number of pods in an ASK cluster and cannot be increased.
Quota limits on underlying cloud resources
|Category||Item||Quota for general users||Adjustable|
|Computing||Number of Resource Orchestration Service (ROS) templates||Default: 100|
|Total vCPUs of all pay-as-you-go instances||500 vCPUs|
|High-specification pay-as-you-go instances (with more than 16 vCPUs)||Unavailable for purchase. Only pay-as-you-go instances with less than 16 vCPUs can be purchased.|
|Total vCPUs of all preemptible instances||800 vCPUs|
|Change the billing method of an instance from pay-as-you-go to subscription||Not supported by the following instance families: t1, s1, s2, s3, c1, c2, m1, m2, n1, n2, and e3.|
|Auto Scaling||The maximum number of ECS instances in a scaling group is 2,000.|
Nodes that run the following operating systems can be added to an ACK cluster:
|Network||Number of custom route entries in a route table||48|
|Number of vSwitches in a VPC||24|
|Number of VPCs within an Alibaba Cloud account||10|
|Number of private IP addresses in a VPC||65535||N/A|
|Number of IP addresses that can be added to a basic security group||2000||N/A|
|Number of elastic network interfaces (ENIs)||50000||N/A|
|Number of elastic IP addresses (EIPs) within an Alibaba Cloud account||20|
|SLB||Number of SLB instances within an Alibaba Cloud account||60|
|Number of backend servers that can be attached to an SLB instance||200||N/A|
|Number of listeners that can be added to an SLB instance||50|
|Number of times that an ECS instance can be repeatedly added to SLB instances as a backend server||50||N/A|
|Elastic Block Storage (EBS)||Number of pay-as-you-go disks in all regions within an Alibaba Cloud account||The quota is five times the number of ECS instances across all regions within an Alibaba Cloud account. You can create at least 10 pay-as-you-go disks within each Alibaba Cloud account.|
|Total capacity of all pay-as-you-go disks that are used as data disks within an Alibaba Cloud account||Subject to ECS resource usage, regions, and disk types. You can go to the Privileges & Quotas page in the ECS console to view details. For more information, see View and increase instance quotas.|