Creating a Node Label
In a Kubernetes cluster, if a node containing the Ascend AI Processor is used as the Kubernetes master node, it also functions as the worker node. In addition to the master node label, you also need to add a worker node label to it based on the Ascend AI Processor type. In a production environment, the master node is generally a general-purpose server that does not contain any Ascend AI Processor.
Label the corresponding node on the master node based on the label information in Table 1, so that the cluster scheduling component can schedule worker nodes in different forms. The following is an example of the command for labeling a node, where host name is a node name in the Kubernetes cluster.
kubectl label nodes host name label
Example:
kubectl label nodes ubuntu masterselector=dls-master-node
Node Type |
Processor Type |
Label |
Description |
|---|---|---|---|
Master node |
- |
masterselector=dls-master-node |
Select host-arch=huawei-arm or host-arch=huawei-x86 based on the node architecture. |
Worker node |
Ascend 910 AI Processor |
|
|
Ascend 910 AI Processor (half configuration of NPUs) |
|
||
Servers (with Atlas 300T training cards) |
|
||
Ascend 310 AI Processor |
|
||
Ascend 310P AI Processor |
|
||
Atlas 200I Soc A1 core board (Ascend 310P AI Processor) |
|