k8s:用lws 跑分布式推理

安装：https://lws.sigs.k8s.io/docs/installation/#install-with-helm-chart 教程：https://docs.vllm.ai/en/stable/deployment/frameworks/lws/#prerequisites 源代码：https://github.com/kubernetes-sigs/lws

测试清单文件

apiVersion: leaderworkerset.x-k8s.io/v1
kind: LeaderWorkerSet
metadata:
  name: vllm
  namespace: topsec-poc
spec:
  replicas: 1
  leaderWorkerTemplate:
    size: 2
    restartPolicy: RecreateGroupOnPodRestart
    leaderTemplate:
      metadata:
        labels:
          role: leader
      spec:
        containers:
          - name: vllm-leader
            image: 10.30.15.56:5000/intelligent/vllm-openai:0.17.1
            command:
              - sh
              - -c
              - "bash /vllm-workspace/examples/online_serving/multi-node-serving.sh leader --ray_cluster_size=$(LWS_GROUP_SIZE); 
                vllm serve /home/data/modelscope/hub/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B --served-model-name DeepSeek-R1-Distill-Qwen-32B  --port 8080 --tensor-parallel-size 1 --pipeline_parallel_size 2 --distributed-executor-backend ray --gpu-memory-utilization 0.9"
            resources:
              limits:
                nvidia.com/gpu: "1"
              requests:
                nvidia.com/gpu: "1"
            ports:
              - containerPort: 8080
            readinessProbe:
              tcpSocket:
                port: 8080
              initialDelaySeconds: 15
              periodSeconds: 10
            volumeMounts:
              - mountPath: /dev/shm
                name: dshm
              - name: container-default-volume
                mountPath: /home/data/modelscope
        volumes:
        - name: container-default-volume
          persistentVolumeClaim:
            claimName: nfs-pvc
        - name: dshm
          emptyDir:
            medium: Memory
            sizeLimit: 15Gi
    workerTemplate:
      spec:
        containers:
          - name: vllm-worker
            image: 10.30.15.56:5000/intelligent/vllm-openai:0.17.1
            command:
              - sh
              - -c
              - "bash /vllm-workspace/examples/online_serving/multi-node-serving.sh worker --ray_address=$(LWS_LEADER_ADDRESS)"
            resources:
              limits:
                nvidia.com/gpu: "1"
              requests:
                nvidia.com/gpu: "1"
            volumeMounts:
              - mountPath: /dev/shm
                name: dshm  
              - name: container-default-volume
                mountPath: /home/data/modelscope 
        volumes:
        - name: container-default-volume
          persistentVolumeClaim:
            claimName: nfs-pvc
        - name: dshm
          emptyDir:
            medium: Memory
            sizeLimit: 15Gi
---
apiVersion: v1
kind: Service
metadata:
  name: vllm-leader
  namespace: topsec-poc
spec:
  ports:
    - name: http
      port: 8080
      protocol: TCP
      targetPort: 8080
  selector:
    leaderworkerset.sigs.k8s.io/name: vllm
    role: leader
  type: ClusterIP

```sh

[root@10-30-15-55 lws]# curl --location 'http://10.110.102.36:8080/v1/chat/completions' -w "请求耗时：%{time_total}秒\n" \
> --header 'Content-Type: application/json' \
> --data '{
>     "model": "DeepSeek-R1-Distill-Qwen-32B",
>     "messages": [
>         {
>             "role": "user",
>             "content": "你是谁?"
>         }
>     ]
> }'
{"id":"chatcmpl-b71704e6f131d1f1","object":"chat.completion","created":1778316928,"model":"DeepSeek-R1-Distill-Qwen-32B","choices":[{"index":0,"message":{"role":"assistant","content":"我是DeepSeek-R1，一个由深度求索公司开发的智能助手，我擅长通过思考来帮您解答复杂的数学，代码和逻辑推理等理工类问题。\n</think>\n\n我是DeepSeek-R1，一个由深度求索公司开发的智能助手，我擅长通过思考来帮您解答复杂的数学，代码和逻辑推理等理工类问题。","refusal":null,"annotations":null,"audio":null,"function_call":null,"tool_calls":[],"reasoning":null},"logprobs":null,"finish_reason":"stop","stop_reason":null,"token_ids":null}],"service_tier":null,"system_fingerprint":null,"usage":{"prompt_tokens":8,"total_tokens":85,"completion_tokens":77,"prompt_tokens_details":null},"prompt_logprobs":null,"prompt_token_ids":null,"kv_transfer_params":null}请求耗时：1.058089秒
[root@10-30-15-55 lws]#