安装seldon

上一篇, 我们已使用 xgboost 完成训练并且产生model档, 这个model的档名为bst_save_model.pkl

现在我们来到了第二个范例的第三个步骤:serving, 如下图所示.

这次我们要使用seldon部署我们的模型. Seldon是最近很热门的model部署工具, 它建立在kubernetes平台上, 也因为我们在第二天就已经安装好一个K8s cluster, 所以要安装seldon core是一件很方便的事.
Seldon core的官网在这里: https://www.seldon.io/tech/products/core/

首先, 我们来安装istio, 因为我们在安装seldon时会启用istio.

安装istio

• 下载istio

  $curl -L https://istio.io/downloadIstio | sh -
  

• 设定PATH

  $cd istio-1.11.1  #版本号要根据下载的版本做修改
  $export PATH=$PWD/bin:$PATH
  

• 开始安装

  $kubectl create namespace istio-system
  $helm install istio-base manifests/charts/base -n istio-system
  $helm install istiod manifests/charts/istio-control/istio-discovery -n istio-system
  $helm install istio-ingress manifests/charts/gateways/istio-ingress -n istio-system
  $helm install istio-egress manifests/charts/gateways/istio-egress -n istio-system
  

• 建立seldon GATEWAY

  apiVersion: networking.istio.io/v1alpha3
  kind: Gateway
  metadata:
    name: seldon-gateway
    namespace: istio-system
  spec:
    selector:
      istio: ingressgateway # use istio default controller
    servers:
    - port:
        number: 80
        name: http
        protocol: HTTP
      hosts:
      - "*"
  

安装seldon

• 建立seldon所使用的namespace

  kubectl create namespace seldon-system
  

• 执行helm指令安装seldon
  ps: 我们在Day4 安装JupyterHub 已安装过hlem

  helm install seldon-core seldon-core-operator \
    --repo https://storage.googleapis.com/seldon-charts \
    --set usageMetrics.enabled=true \
    --namespace seldon-system \
    --set istio.enabled=true \
    --set istio.gateway=istio-system/seldon-gateway
  

  执行完成的讯息如下:

  index.go:339: skipping loading invalid entry for chart "seldon-core-crd" "0.1.5a" from https://storage.googleapis.com/seldon-charts: validation: chart.metadata.version "0.1.5a" is invalid
  index.go:339: skipping loading invalid entry for chart "seldon-core-crd" "0.1.6_SNAPSHOT" from https://storage.googleapis.com/seldon-charts: validation: chart.metadata.version "0.1.6_SNAPSHOT" is invalid
  index.go:339: skipping loading invalid entry for chart "seldon-core-crd" "0.1.6_SNAPSHOT" from https://storage.googleapis.com/seldon-charts: validation: chart.metadata.version "0.1.6_SNAPSHOT" is invalid
  index.go:339: skipping loading invalid entry for chart "seldon-core" "0.1.6_SNAPSHOT" from https://storage.googleapis.com/seldon-charts: validation: chart.metadata.version "0.1.6_SNAPSHOT" is invalid
  index.go:339: skipping loading invalid entry for chart "seldon-core" "0.1.6_SNAPSHOT" from /home/k8s/.cache/helm/repository/E3xhv-3+r1pSeDV8I8nqHo7tIjk=-index.yaml: validation: chart.metadata.version "0.1.6_SNAPSHOT" is invalid
  index.go:339: skipping loading invalid entry for chart "seldon-core-crd" "0.1.5a" from /home/k8s/.cache/helm/repository/E3xhv-3+r1pSeDV8I8nqHo7tIjk=-index.yaml: validation: chart.metadata.version "0.1.5a" is invalid
  index.go:339: skipping loading invalid entry for chart "seldon-core-crd" "0.1.6_SNAPSHOT" from /home/k8s/.cache/helm/repository/E3xhv-3+r1pSeDV8I8nqHo7tIjk=-index.yaml: validation: chart.metadata.version "0.1.6_SNAPSHOT" is invalid
  index.go:339: skipping loading invalid entry for chart "seldon-core-crd" "0.1.6_SNAPSHOT" from /home/k8s/.cache/helm/repository/E3xhv-3+r1pSeDV8I8nqHo7tIjk=-index.yaml: validation: chart.metadata.version "0.1.6_SNAPSHOT" is invalid
  W0913 11:51:43.488868   16657 warnings.go:70] rbac.authorization.k8s.io/v1beta1 ClusterRole is deprecated in v1.17+, unavailable in v1.22+; use rbac.authorization.k8s.io/v1 ClusterRole
  W0913 11:51:43.506167   16657 warnings.go:70]   rbac.authorization.k8s.io/v1beta1 ClusterRoleBinding is deprecated in v1.17+, unavailable in v1.22+; use rbac.authorization.k8s.io/v1 ClusterRoleBinding
  W0913 11:51:43.676157   16657 warnings.go:70] admissionregistration.k8s.io/v1beta1 ValidatingWebhookConfiguration is deprecated in v1.16+, unavailable in v1.22+; use admissionregistration.k8s.io/v1 ValidatingWebhookConfiguration
  W0913 11:51:48.918777   16657 warnings.go:70] rbac.authorization.k8s.io/v1beta1 ClusterRole is deprecated in v1.17+, unavailable in v1.22+; use rbac.authorization.k8s.io/v1 ClusterRole
  W0913 11:51:48.988508   16657 warnings.go:70] rbac.authorization.k8s.io/v1beta1 ClusterRoleBinding is deprecated in v1.17+, unavailable in v1.22+; use rbac.authorization.k8s.io/v1 ClusterRoleBinding
  W0913 11:51:50.531620   16657 warnings.go:70] admissionregistration.k8s.io/v1beta1 ValidatingWebhookConfiguration is deprecated in v1.16+, unavailable in v1.22+; use admissionregistration.k8s.io/v1 ValidatingWebhookConfiguration
  NAME: seldon-core
  LAST DEPLOYED: Mon Sep 13 11:51:37 2021
  NAMESPACE: seldon-system
  STATUS: deployed
  REVISION: 1
  TEST SUITE: None
  k8s@k8s-master:~/mlops$
  

准备pvc

因为使用seldon部署model时需要将model放在下列四种储存空间的其中一种:

• Google Cloud Storage (using gs://)
• S3-compatible (using s3://)
• Minio-compatible (using s3://)
• Azure Blob storage (using https://(.+?).blob.core.windows.net/(.+))

除了上述4个环境, 还可以使用pvc, 在这个网页有一行这麽写
A Kubernetes PersistentVolume can be used instead of a bucket using pvc://.

我们先使用pvc空间做一次, 之後再使用MinIO做一次.
而且在我们的K8s环境中原本就有一个NFS空间, 之前是让JupyterHub使用, 现在我们就再使用NFS空间建立pvc放置要部署在seldon上的模型档案.

请撰写下列yaml内容. 请注意, 这个的storageClassName必需是已经存在的名称, 因为我们在安装JupyterHub就建立过storage class, 名称即是nfs-client. 如果你没有这个storage class, 那你要先建立一个.

apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: seldon-pv-claim
spec:
  storageClassName: nfs-client
  accessModes:
    - ReadWriteOnce
  resources:
    requests:
      storage: 3Gi

将这个yaml档存为seldon-pvc.yaml
所以执行下列指令, 这样我们就准备好所需要的空间.

$kubectl apply -f seldon-pvc.yaml

你也可以在github下载这个档案

到这里我们就准备好所需要的环境, 下一篇我们就来把训练好的模型使用seldon部署起来

参考资料

https://istio.io/latest/docs/setup/getting-started/#download
https://istio.io/latest/docs/setup/install/helm/
https://docs.seldon.io/projects/seldon-core/en/v1.9.1/workflow/github-readme.html