计算任务¶

计算任务（kind: compute）适用于纯 CPU 服务部署，如 Web 服务、数据库、缓存、代理等。底层对应 Kubernetes Deployment + NodePort Service，支持多副本和健康检查。

YAML 完整字段¶

kind: compute
version: v0.1

job:
  name: <服务名称>
  priority: medium
  description: "描述"

environment:
  image: <镜像地址>
  command: [...]   # 可选，不填使用镜像默认 entrypoint
  args: [...]      # 可选
  env:
    - name: KEY
      value: VALUE

service:
  replicas: 1       # 副本数（默认 1）
  port: 80          # 服务端口
  healthCheck: /    # 健康检查路径（可选）

resources:
  pool: default
  gpu: 0            # 计算任务设置为 0（纯 CPU）
  cpu: 2
  memory: 4Gi

storage:
  workdirs:
    - path: /data

场景一：部署 Nginx Web 服务¶

nginx-service.yaml

kind: compute
version: v0.1

job:
  name: nginx-web
  priority: medium
  description: "Nginx 静态 Web 服务"

environment:
  image: nginx:latest
  command: []
  args: []
  env:
    - name: NGINX_PORT
      value: "80"

service:
  replicas: 2
  port: 80
  healthCheck: /

resources:
  pool: compute-pool
  gpu: 0
  cpu: 2
  memory: 4Gi

storage:
  workdirs:
    - path: /etc/nginx/conf.d
    - path: /var/www/html

gpuctl create -f nginx-service.yaml
gpuctl describe job nginx-web    # 查看访问地址

场景二：部署 Redis 缓存¶

redis-cache.yaml

kind: compute
version: v0.1

job:
  name: redis-cache
  priority: medium

environment:
  image: redis:7.0-alpine
  command: ["redis-server"]
  args: ["--maxmemory", "1gb", "--maxmemory-policy", "allkeys-lru"]

service:
  replicas: 1
  port: 6379

resources:
  pool: default
  gpu: 0
  cpu: 1
  memory: 2Gi

storage:
  workdirs:
    - path: /data   # Redis 持久化目录

场景三：部署自定义 Python API 服务¶

fastapi-service.yaml

kind: compute
version: v0.1

job:
  name: my-fastapi-service
  priority: medium

environment:
  image: my-registry/fastapi-app:v1.0
  command: ["uvicorn", "main:app"]
  args: ["--host", "0.0.0.0", "--port", "8080"]
  env:
    - name: DATABASE_URL
      value: "postgresql://user:pass@db:5432/mydb"
    - name: LOG_LEVEL
      value: "info"

service:
  replicas: 3
  port: 8080
  healthCheck: /health

resources:
  pool: default
  gpu: 0
  cpu: 4
  memory: 8Gi

更新计算任务¶

# 修改 YAML（如调整副本数、镜像版本等）后执行：
gpuctl apply -f nginx-service.yaml

查看服务日志¶

gpuctl logs nginx-web -f

删除计算任务¶

gpuctl delete job nginx-web

与推理任务的区别

compute 和 inference 底层都使用 K8s Deployment，主要区别在于语义标注（runwhere.ai/job-type label）和资源池选择。计算任务通常 gpu: 0，推理任务通常有 GPU 需求。