Use labels to schedule pods to FPGA-accelerated nodes - Container Service for Kubernetes

FPGA-accelerated nodes in an ACK cluster carry dedicated node labels that expose FPGA hardware attributes. Use these labels in your workload's nodeSelector to ensure pods land only on FPGA nodes—preventing ordinary workloads from consuming FPGA capacity.

Prerequisites

Before you begin, ensure that you have:

An ACK cluster with FPGA-accelerated nodes. See Create an ACK cluster with FPGA-accelerated nodes.
kubectl configured to connect to the cluster. See Obtain the kubeconfig file of a cluster and use kubectl to connect to the cluster.

FPGA node labels

When you add FPGA-accelerated nodes to an ACK cluster, ACK automatically attaches node labels that describe the FPGA hardware. The following label is used for scheduling:

Label	Value	Description
`fpga.k8s.aliyun.com`	`f3`	Alibaba Cloud ECS f3 FPGA instance

Set this label as a nodeSelector in your workload spec to target FPGA nodes. Set resources.limits with the corresponding FPGA resource key to request an FPGA unit.

Step 1: View FPGA node labels

Method 1: ACK console

Log on to the ACK console. In the left-side navigation pane, click Clusters.
On the Clusters page, find the cluster and click its name. In the left-side navigation pane, choose Nodes > Nodes.
On the Nodes page, find the FPGA-accelerated node, then choose More > Details in the Actions column. View the labels attached to the node.

Method 2: kubectl

List all nodes in the cluster:

kubectl get nodes

Expected output:

NAME                       STATUS   ROLES    AGE     VERSION
cn-beijing.192.168.XX.X1   Ready    <none>   3h51m   v1.18.8-aliyun.1
cn-beijing.192.168.XX.X2   Ready    <none>   3h41m   v1.18.8-aliyun.1

Describe the FPGA-accelerated node to view its labels:

kubectl describe node cn-beijing.192.168.XX.X2

Expected output:

Name:               cn-beijing.192.168.XX.X2
Roles:              <none>
Labels:             ack.aliyun.com=c05888610e***
                    alibabacloud.com/nodepool-id=npfda879b6***
                    beta.kubernetes.io/arch=amd64
                    beta.kubernetes.io/instance-type=ecs.f3-c4f1.xlarge
                    beta.kubernetes.io/os=linux
                    failure-domain.beta.kubernetes.io/region=cn-beijing
                    failure-domain.beta.kubernetes.io/zone=cn-beijing-h
                    fpga.k8s.aliyun.com=f3
                    kubernetes.io/arch=amd64
                    kubernetes.io/hostname=cn-beijing.192.168.XX.X2
                    kubernetes.io/os=linux
                    node.kubernetes.io/instance-type=ecs.f3-c4f1.xlarge
                    topology.diskplugin.csi.alibabacloud.com/zone=cn-beijing-h
                    topology.kubernetes.io/region=cn-beijing
                    topology.kubernetes.io/zone=cn-beijing-h

The fpga.k8s.aliyun.com=f3 label identifies this as an ECS f3 FPGA instance. Use this label in the next step to target the node.

Step 2: Schedule pods to FPGA-accelerated nodes

The following Job spec uses nodeSelector: fpga.k8s.aliyun.com: f3 to target f3 FPGA nodes and xilinx.com/fpga-aliyun-f3: 1 to request one FPGA unit.

Log on to the ACK console. In the left-side navigation pane, click Clusters.
On the Clusters page, find the cluster and click its name. In the left-side navigation pane, choose Workloads > Deployments.
On the Deployments page, click Create from YAML.

Select a custom template from the sample templates drop-down list, then copy the following content to the Template field.

apiVersion: batch/v1
kind: Job
metadata:
  name: fpga-run-task1
spec:
  backoffLimit: 0
  completions: 1
  parallelism: 1
  template:
    spec:
      nodeSelector:
        fpga.k8s.aliyun.com: f3
      containers:
      - image: <your image>
        imagePullPolicy: Always
        name: fpga-run-task1
        resources:
          limits:
            xilinx.com/fpga-aliyun-f3: 1
        securityContext:
          privileged: true

Replace the value of <your image> with the address of your custom image.

Click Create. In the left-side navigation pane of the cluster details page, choose Workloads > Pods. In the list of pods, you can view that the specified pod is scheduled to the required FPGA-accelerated node. You can use labels to schedule pods to specific FPGA-accelerated nodes with ease.