ModelDeployment CRD

以下CR 是 ModelDeployment 资源的完整释义，通过 KubeGems 模型商店部署的算法推理服务可在应用中心对已部署的算法进行持续修改。

kind: ModelDeployment
apiVersion: models.kubegems.io/v1beta1
metadata:
  name: gpt2
  namespace: default
spec:
  replicas: 1
  model:
    source: huggingface     # 模型来源
    name: library/gpt2      # 模型名称
    version: v3.1.0         # 模型版本
    url: "https://modelx.kubegems.io"  # 算法服务注册到网关的默认域名
    token: "<token>"        # 模型 Token，从 modelxd 服务中签发
    task: text-generation
  server:
    # seldon server protocol,ref: https://docs.seldon.io/projects/seldon-core/en/latest/graph/protocols.html#graph-protocols--page-root
    protocol: v2 
    # replace default seldon server image
    image: "some/custom-image:latest" 
    # seldon implementation, ref: https://docs.seldon.io/projects/seldon-core/en/latest/servers/overview.html#servers-overview--page-root
    kind: HUGGINGFACE_SERVER 
    # init container, ref: https://docs.seldon.io/projects/seldon-core/en/latest/servers/overview.html#init-containers
    storageInitializerImage: "docker.io/kubegems/modelx-dl:latest"
    mounts:
      # 模型挂载路径
      - kind: Model
        mountPath: /app/model-files 
      # 挂载 HostPath （可选）
      - kind: HostPath
        mountPath: /host-path
        source: /host-path
      # 挂载 ConfigMap（可选）
      - kind: ConfigMap
        mountPath: /config
        source: configmap-name
      # 挂载 PersistentVolume（可选）
      - kind: PVC
        mountPath: /data
        source: some-pvc-name
      # 挂载 EmptyDir（可选）
      - kind: EmptyDir
        mountPath: /tmp
      # 挂载 Secret（可选）
      - kind: Secret
        mountPath: /certs
        source: some-secret-name
    parameters:
      - name: task
        value: text-generation
      - name: pretrained_model
        value: gpt2
    # 运行特权模式
    privileged: true 
    # 自定义启动命令
    command:
      - run.py
    # 自定义启动参数
    args:
      - --arg-foo=value
    # 自定义环境变量
    env:
      - name: ENV_NAME
        value: env_value
      - name: ENV_FOO
        valueFrom: # container's valueFrom
          configMapKeyRef:
            key: some-config
            name: "key_foo"
    # 自定义服务端口
    ports:
      - name: http # seldon 默认仅暴露名称为 http 和 grpc 的端口到 service
        containerPort: 8080
      - name: grpc
        containerPort: 50051
      - name: other-port # 其他名称的端口无法在seldon service上暴露出来
        containerPort: 1234
    # 自定义资源限制
    resources:
      limits:
        cpu: "2"
        memory: 4Gi
      requests:
        cpu: 100m
        memory: 100Mi
    upgradeStrategy: RollingUpdate # deployment strategy
    # pod相关属性,containers名称必须设置为 model
    podSpec:
      contaienrs:
        - name: model
      # 其余 pod 的特性可参考 podspec api 描述
      # https://kubernetes.io/docs/reference/generated/kubernetes-api/v1.19/#podspec-v1-core
      nodeSelector:
        kubernetes.io/hostname: kubegems-gpu-1
  # ingress 配置
  ingress:
    host: api.models.kubegems.io # ingress host
    className: "ingress-nginx" # nginx ingress class name
    gatewayName: default-gateway # kubegems gateway name