An ACK Lingjun managed cluster from Container Service for Kubernetes (ACK) is a specialized cluster for PAI-Lingjun AI Computing Service. It features a fully managed, highly available Kubernetes control plane and uses Lingjun compute nodes as its worker nodes. This topic describes how to create an ACK Lingjun managed cluster.
Prerequisites
-
You have purchased the required PAI-Lingjun AI Computing Service products, such as compute nodes and Lingjun connections. For more information, see Purchase products.
-
You have purchased and configured other required cloud products, such as Cloud Enterprise Network (CEN), Application Real-Time Monitoring Service (ARMS), Virtual Private Cloud (VPC), and ACK Lingjun managed clusters. For more information, see Purchase and configure other cloud products.
-
Your account is real-name verified and has a cash or credit balance of at least CNY 100. For more information, see Set credit limits for RAM users.
Background information
An ACK Lingjun managed cluster is a standard Kubernetes cluster service that provides a fully managed and highly available control plane. It supports the efficient management of heterogeneous resources and scheduling of heterogeneous tasks, serves as the cloud-native foundation for PAI, and provides cloud-native enhanced capabilities for high-performance computing (HPC) scenarios such as AI. For more information about ACK Lingjun managed clusters, see What is ACK Lingjun?.
Procedure
-
Log on to the Intelligent Computing Lingjun console.
-
In the left-side navigation pane, choose Resources & Nodes > Cluster Management.
-
Click One-click Cluster Creation.
-
Click the PAI Lingjun Cluster (including PAI, ACK, and CPFS) card.
Follow the on-screen instructions to complete the configurations for Cluster and Group Configuration, Network Configuration, Basic Software Instance Parameter Configuration, and Software Instance and Group Mapping Configuration.
ACK Lingjun managed clusters are billed separately. For more information, see Billing details.
Configure clusters and node groups
You can plan multiple clusters based on your business requirements and divide the compute nodes in a cluster into node groups. You can improve the resource utilization of compute nodes by planning your clusters and compute nodes. After you plan your clusters and compute nodes, perform the steps described in this section to configure clusters and node groups.
Configure the cluster information.
Specify information such as the cluster name, root password of cluster nodes, and resource group. For more information about how to create a resource group, see Create a resource group.
Click Create Group to create a node group.
In the Create Group dialog box, specify the group name and information about the nodes that belong to the group such as the node model and image based on your plan.
Click Select Node Instance next to Node Instance to select the nodes to be added to the group.
Click Save and go to the next step. Network Configurations.
Configure networks
A Lingjun cluster initially resides in an isolated network. You must connect the cluster to the Alibaba Cloud public cloud by using a Lingjun connection instance and a CEN instance, and specify a VPC to monitor network connectivity.
As shown in the preceding figure, the network topology involves the following core networks:
Cluster network: the CIDR block that is used by the cluster to assign IP addresses to compute nodes. The CIDR block is a private CIDR block.
Monitoring network: the VPC that is used to monitor network connectivity.
When you plan and configure networks, make sure that the CIDR blocks of the preceding networks do not conflict with each other. After you plan your networks, you can perform the following steps to configure networks for your cluster.
After you configure the networks of the cluster, you must check whether the network configurations of the CEN instance are correct. For more information about how to configure a CEN instance, see the "CEN configurations" section of the Activate and configure other Alibaba Cloud services topic.
Configure a Lingjun VPD.
The Lingjun Virtual Private Datacenter (VPD) is used to assign IP addresses to compute nodes in the Lingjun cluster. Enter a valid private CIDR block.
A Lingjun subnet is a subnet of the Lingjun VPD. For more information about Lingjun VPDs and their subnets, see Manage Lingjun VPDs.
NoteYou must plan a Lingjun VPD in advance. The Lingjun VPD cannot conflict with the CIDR blocks of other networks to which the Lingjun cluster is to be connected, such as CIDR blocks of VPCs or data centers.
The number of available IP addresses in a Lingjun VPD determines the maximum number of nodes that can be deployed in the Lingjun cluster. You must prepare a CIDR block whose subnet mask length is greater than 22 bits to prevent the cluster from being unable to be scaled up.
-
Optional. Configure the bond allocation policy of a Lingjun subnet. If you select specific node models, you must configure the bond allocation policy for the physical NICs of Lingjun nodes. Bonds are associated with Lingjun nodes. You can configure bonds by configuring the bond allocation policy, node model allocation policy, or node allocation policy.
Configure a bond allocation policy
Different node models have different numbers of bond interfaces. The number of bond interfaces for the cluster is equal to the maximum number of bond interfaces on any single node model in the cluster. Bond interfaces in a cluster are named in the bondx format, where x is a number starting from 0.
For example, consider a cluster with two node models, A and B, which have 3 and 4 bond interfaces, respectively. The cluster has 4 bond interfaces, named from bond0 to bond3. Node model A uses the policies for interfaces bond0 to bond2.
NoteYou can configure only one bond allocation policy for a cluster.
Procedure
Configure a bond allocation policy for the cluster.
Optional. Configure the default bond allocation policy. The bonds that are not assigned a policy use the default bond allocation policy. Select Apply to all to assign the default bond policy to all bonds.
Configure a node model allocation policy
You can specify a node model allocation policy for each node model in a cluster. The maximum number of node model allocation policies in a cluster is equal to the number of node groups in the cluster.
Procedure
Click Model Type. In the AddModel Type dialog box, select a node model from the Model drop-down list.
Configure a node model allocation policy. The policy is applied to all nodes of the selected model.
Configure a node allocation policy
You can configure a node allocation policy for each node in a cluster. Different bonds of a node can be connected to different CIDR blocks or subnets of a Lingjun cluster.
Procedure
Click Node Policy. In the AddNode Policy dialog box, select a node form the Node drop-down list.
Configure a node allocation policy. The policy is applied to the selected node.
-
Configure a Lingjun connection instance.
-
Click Authorize to grant permissions to the Lingjun connection instance.
You can use the Lingjun connection instance to connect the Lingjun cluster to a CEN instance and access other Alibaba Cloud services. Therefore, you must authorize Lingjun to access other Alibaba Cloud services. For more information, see Appendix: Service-linked role for Lingjun connection instances.
Select the ID of the Lingjun connection instance from the InstanceID drop-down list. The Lingjun connection instance is used by the cluster to connect to the Alibaba Cloud public cloud.
Select a CEN instance from the CEN drop-down list. The cluster is connected to the CEN instance by using the Lingjun connection instance.
ImportantYou must create a transit router in the CEN instance. The region of the transit router must be the same as that of the Lingjun nodes. For more information, see Transit routers.
-
Configure the monitoring network.
Configure the CEN instance. Connect a VPC to the transit router of the CEN instance that is created in the previous step. You can create a VPC or use an existing VPC. For more information, see the "CEN configurations" section of the Activate and configure other Alibaba Cloud services topic. Make sure that the vSwitch in the VPC has at least one idle IP address. The Lingjun cluster uses this vSwitch to monitor the network connectivity of the Lingjun connection instance.
ImportantYou can select a VPC from the drop-down list only if you connect the VPC to the selected transit router.
The Lingjun VPD and the CIDR block of the VPC that is used as the monitoring network cannot conflict with each other, and the CIDR block of the VPC that is used as the monitoring network cannot conflict with the CIDR blocks of other networks to which the Lingjun cluster is to be connected, such as CIDR blocks of other VPCs or data centers.
Click the
icons next to the VPC drop-down list and the Switch(VSwitch) drop-down list. Then, select the VPC and vSwitch that you create.
Click Save and go to the next step. Basic Parameters of Software Instance.
Basic software instance parameters
-
Configure basic parameters.
ACK tab
For more information about the cluster parameters, see Create an ACK managed cluster.
ImportantEnsure that the Service CIDR, the Lingjun cluster CIDR block, the public CIDR block, and the VPC CIDR block do not overlap.
CPFS tab
For more information about how to create a CPFS file system, see Create a file system.
NoteAfter the CPFS instance is created, you can view the instance details in the CPFS console.
PAI tab
NoteFor more information about how to configure ApsaraDB RDS, cloud storage, Container Registry (ACR), and OAuth authentication, see Purchase and configure other cloud products.
-
Click Save and Go to Next Step: Software Instance and Group Mapping to continue.
Software instance and group mappings
ACK Lingjun managed clusters use Lingjun node pools to manage Lingjun compute nodes in groups. This simplifies tasks such as node configuration, batch management, targeted scheduling, and GPU configuration. For more information about node pools, see Lingjun node pool overview.
-
Click Create Node Pool to create an ACK node pool.
-
Configure the node pool name and maximum number of nodes.
-
Click Select Associated Group. In the dialog box that appears, select the cluster groups to associate and click OK.
-
Click Save and Go to Next Step: Confirm Configuration to continue.
Confirm configuration
-
On the Confirm Configuration page, review all settings. If they are correct, click Submit Configuration to create the cluster.
-
In the Dependency Check section, click Complete Authorization to grant the required permissions to Container Service for Kubernetes.