开启智能预测功能

更新时间：2025-08-22

Sugar BI 的智能预测功能包含内置预测服务、BML/EasyDL 预测服务、训练预测服务三部分：

内置预测服无需进行额外部署，在 SaaS 高级版、私有部署 2 账号以上版本可用；

训练预测服务部分只在私有部署版本中提供，须购买包含智能预测功能的 License。然后按照下面步骤部署模型训练和推理服务后即可使用；

BML/EasyDL 预测服务在 SaaS 高级版中支持。在私有部署版本中则和训练预测服务一致，须购买包含智能预测功能的 License。

前期准备

智能预测模型的储存依赖对象存储，该部分与自助取数功能复用，如果您已经完成了自助取数功能的部署和开启，则不需要额外配置；如果未完成，请参照开启自助取数功能进行配置，并完成该文档最后的部署验证部分。

下面进行预测模型的训练和推理服务的部署。

安装环境

模型训练和推理服务建议部署在与 Sugar BI 本体独立的服务器上，安装环境要求如下：

软件环境

Docker

需要您的机器安装Docker，推荐版本为v17及以上，Sugar BI可以直接运行在单机 Docker 环境上，如果需要集群化、高可用，可以使用 Docker 自带的 swarm 或者 kubernetes；Sugar BI的安装部署过程中需要您对 Docker 的基础知识有所了解，详见Docker 官网。

Docker 的安装建议参考官方文档，可以安装在Centos、Ubuntu、Docker Desktop on Windows、Mac等系统上。Windows 上的 Docker 启动需要使用「以管理员身份运行」。

硬件环境

如果训练的数据量不大（同时进行训练的任务总数据量在 10G 以下）且不严格要求高可用，单机运行即可，否则视具体情况推荐使用 kubernetes 集群方式部署（建议最少 3 台机器）；每台机器的配置建议：内存（最低 32G）、CPU（16 核或以上）、硬盘（200GB SSD 或以上）。

下载

需要下载训练服务和推理服务两个镜像，我们提供了两种获取安装包镜像的方式：

1、百度智能云的 docker 镜像服务

镜像已经上传到百度智能云 docker 镜像服务，如果机器可以访问公网，可以使用以下命令获取：

Plain Text

1docker pull registry.baidubce.com/sugarbi/sugar-ml-train:1.1.1
2docker tag registry.baidubce.com/sugarbi/sugar-ml-train:1.1.1 sugarbi/sugar-ml-train:1.1.1
3docker pull registry.baidubce.com/sugarbi/sugar-ml-predict:1.0.1
4docker tag registry.baidubce.com/sugarbi/sugar-ml-predict:1.0.1 sugarbi/sugar-ml-predict:1.0.1
5// 查看刚刚拉取的镜像
6docker images

另外，如果您的服务器 CPU、操作系统是国产化的，那么你可能需要 ARM 架构的镜像，命令如下：

Plain Text

1docker pull registry.baidubce.com/sugarbi/sugar-ml-train-arm64:1.1.1
2docker tag registry.baidubce.com/sugarbi/sugar-ml-train-arm64:1.1.1 sugarbi/sugar-ml-train-arm64:1.1.1
3docker pull registry.baidubce.com/sugarbi/sugar-ml-predict-arm64:1.0.1
4docker tag registry.baidubce.com/sugarbi/sugar-ml-predict-arm64:1.0.1 sugarbi/sugar-ml-predict-arm64:1.0.1
5// 查看刚刚拉取的镜像
6docker images

2、直接在线下载

如果您的部署机器不能连接互联网，您可以在联网的机器上访问下列链接下载镜像（如果鼠标点击下载不了，可以复制该地址在浏览器中打开），然后将下载的安装包拷贝到部署机器上，注意：如果您需要同时部署到多台机器上，需要将下载的安装包文件拷贝到所有部署机器上，并且以下的操作也需要在所有部署机器上执行。

点击这里下载: https://sugar-docker-image.cdn.bcebos.com/sugarbi-ml-train-1.1.1.tar.gz

点击这里下载: https://sugar-docker-image.cdn.bcebos.com/sugarbi-ml-predict-1.0.1.tar.gz

在部署的机器上，进入到您拷贝的安装包的同目录下，

Plain Text

1docker load -i ./sugarbi-ml-train-1.1.1.tar.gz
2docker load -i ./sugarbi-ml-predict-1.0.1.tar.gz
3// 上面的命令会执行数十秒，然后执行以下命令查看刚刚load的镜像
4docker images

另外，如果您的服务器 CPU、操作系统是国产化的，那么你可能需要 ARM 架构的镜像：

点击这里下载: https://sugar-docker-image.cdn.bcebos.com/sugarbi-ml-train-arm64-1.1.1.tar.gz

点击这里下载: https://sugar-docker-image.cdn.bcebos.com/sugarbi-ml-predict-arm64-1.0.1.tar.gz

在部署的机器上，进入到您拷贝的安装包的同目录下，

Plain Text

1docker load -i ./sugarbi-ml-train-arm64-1.1.1.tar.gz
2docker load -i ./sugarbi-ml-predict-arm64-1.0.1.tar.gz
3// 上面的命令会执行数十秒，然后执行以下命令查看刚刚load的镜像
4docker images

运行训练和推理服务

首先检查两个镜像是否成功加载，执行docker images都能看到如下展示即证明下载成功：

注意如果使用 ARM 版本镜像，镜像名称会有 -arm64 后缀，下面的运行命令中也需要做相应调整。

安装有两种方式分别对应 Docker 单机模式和多机集群模式：

一、Docker 单机运行

运行训练服务镜像

运行下面的命令即可在 8090 端口上启用训练服务，您也可以调整 8090 为其他端口。

Plain Text

1docker run --shm-size=4g --restart unless-stopped -d -p 8090:54321 --name sugar-ml-train  sugarbi/sugar-ml-train:1.1.1

运行推理服务镜像

1、新建一个名为env的文件（文件名没有后缀，就是 env），并用文本编辑器打开，复制以下的内容并填写相应部分：

Plain Text

1# 自助取数中使用的对象储存的类型：minio 或 bos
2sugar_ml_model_storage_type=bos
3# 对象存储中为预测模型分配的 bucket
4sugar_ml_model_bucket=sugar-ml
5
6# 如果对象存储的类型是 bos，配置下面 3 个环境变量
7sugar_ml_bos_ak=
8sugar_ml_bos_sk=
9sugar_ml_bos_endpoint=http://bj.bcebos.com
10
11# 如果对象存储的类型是 minio，配置下面 3 个环境变量
12sugar_ml_minio_ak=
13sugar_ml_minio_sk=
14sugar_ml_minio_endpoint=

2、在env文件的目录下执行以下命令来启动推理服务

为了更快速的推理，需要提供一个文件夹来储存预测模型的缓存，在宿主机新建文件夹~/sugar-ml-predict（当然目录名称您可以随意修改，同步替换下面命令中的即可）：

Plain Text

1docker run --ulimit nofile=65100:65100 --restart unless-stopped -d -p 8091:8111 --name sugar-ml-predict -v ~/sugar-ml-predict:/sugar-predict --env-file env sugarbi/sugar-ml-predict:1.0.1

其中 8091 是本地端口，可以修改成自己想要的端口。

二、kubernetes 集群部署

集群方式部署时，建议使用 3 台机器，训练和预测服务都部署 3 个以上 pod ，均匀分布在各机器上。

部署推理服务

部署推理服务，新建一个predict.yaml文件，并用文本编辑器打开，复制以下的内容并填写相应部分：

YAML

1apiVersion: v1
2kind: Namespace
3metadata:
4  name: sugar-ml
5---
6kind: Deployment
7apiVersion: apps/v1
8metadata:
9  labels:
10    app: sugar-ml-predict
11  name: sugar-ml-predict
12  namespace: sugar-ml
13spec:
14  selector:
15    matchLabels:
16      app: sugar-ml-predict
17  replicas: 3 # pod 数目，可根据实际情况调整
18  template:
19    metadata:
20      labels:
21        app: sugar-ml-predict
22    spec:
23      volumes:
24        - name: sugar-predict
25          hostPath:
26            path: /sugar-ml-predict # 模型缓存挂载的volume，可以根据实际情况修改
27      containers:
28        - name: sugar-ml-predict
29          image: registry.baidubce.com/sugarbi/sugar-ml-predict:1.0.1
30          ports:
31            - containerPort: 8111
32          volumeMounts:
33            - name: sugar-predict
34              mountPath: /sugar-predict
35          livenessProbe:
36            httpGet:
37              path: /predict
38              port: 8111
39            initialDelaySeconds: 30
40            timeoutSeconds: 5
41          env:
42            - name: sugar_ml_model_storage_type # 自助取数中使用的对象储存的类型：minio 或 bos
43              value: bos
44            - name: sugar_ml_model_bucket # 对象存储中为预测模型分配的 bucket
45              value: sugar-ml
46
47            - name: sugar_ml_bos_ak # 如果对象存储的类型是 bos，配置下面 3 个环境变量
48              value:
49            - name: sugar_ml_bos_sk
50              value:
51            - name: sugar_ml_bos_endpoint
52              value: http://bj.bcebos.com
53
54            - name: sugar_ml_minio_ak # 如果对象存储的类型是 minio，配置下面 3 个环境变量
55              value:
56            - name: sugar_ml_minio_sk
57              value:
58            - name: sugar_ml_minio_endpoint
59              value:
60
61---
62kind: Service
63apiVersion: v1
64metadata:
65  labels:
66    app: sugar-ml-predict
67  name: sugar-ml-predict-service
68  namespace: sugar-ml
69spec:
70  ports:
71    - port: 8111
72      targetPort: 8111
73      nodePort: 32590 # 推理服务端口号，可根据实际情况调整
74  type: NodePort
75  selector:
76    app: sugar-ml-predict

然后在该文件所在目录执行：

Plain Text

1kubectl apply -f predict.yaml

这样就将推理服务部署在了 32590 端口，可根据实际情况调整端口号。

通过下面命令查看节点状态：

Plain Text

1kubectl get pods -A

等待全部节点运行状态都变为 Ready 即可。例如这里部署了 3 个 POD：

部署 Ingress

1、训练服务需要通过 Ingress 暴露出去，如果您的 k8s 中有自己的 Ingress 服务可以跳过本步骤，在 Ingress 中将下一节部署的训练服务 Service sugar-ml-service 暴露出去即可。

如果没有已有的 Ingress ，这里给出 Ingress 的一种安装方法，Ingress 有很多种实现，也可使用您熟悉的版本。

新建一个ingress-nginx.yaml文件，并用文本编辑器打开，复制以下的内容并填写相应部分：

YAML

1apiVersion: v1
2kind: Namespace
3metadata:
4  name: ingress-nginx
5  labels:
6    app.kubernetes.io/name: ingress-nginx
7    app.kubernetes.io/instance: ingress-nginx
8
9---
10# Source: ingress-nginx/templates/controller-serviceaccount.yaml
11apiVersion: v1
12kind: ServiceAccount
13metadata:
14  labels:
15    helm.sh/chart: ingress-nginx-4.0.15
16    app.kubernetes.io/name: ingress-nginx
17    app.kubernetes.io/instance: ingress-nginx
18    app.kubernetes.io/version: 1.1.1
19    app.kubernetes.io/managed-by: Helm
20    app.kubernetes.io/component: controller
21  name: ingress-nginx
22  namespace: ingress-nginx
23automountServiceAccountToken: true
24---
25# Source: ingress-nginx/templates/controller-configmap.yaml
26apiVersion: v1
27kind: ConfigMap
28metadata:
29  labels:
30    helm.sh/chart: ingress-nginx-4.0.15
31    app.kubernetes.io/name: ingress-nginx
32    app.kubernetes.io/instance: ingress-nginx
33    app.kubernetes.io/version: 1.1.1
34    app.kubernetes.io/managed-by: Helm
35    app.kubernetes.io/component: controller
36  name: ingress-nginx-controller
37  namespace: ingress-nginx
38data:
39  allow-snippet-annotations: 'true'
40---
41# Source: ingress-nginx/templates/clusterrole.yaml
42apiVersion: rbac.authorization.k8s.io/v1
43kind: ClusterRole
44metadata:
45  labels:
46    helm.sh/chart: ingress-nginx-4.0.15
47    app.kubernetes.io/name: ingress-nginx
48    app.kubernetes.io/instance: ingress-nginx
49    app.kubernetes.io/version: 1.1.1
50    app.kubernetes.io/managed-by: Helm
51  name: ingress-nginx
52rules:
53  - apiGroups:
54      - ''
55    resources:
56      - configmaps
57      - endpoints
58      - nodes
59      - pods
60      - secrets
61      - namespaces
62    verbs:
63      - list
64      - watch
65  - apiGroups:
66      - ''
67    resources:
68      - nodes
69    verbs:
70      - get
71  - apiGroups:
72      - ''
73    resources:
74      - services
75    verbs:
76      - get
77      - list
78      - watch
79  - apiGroups:
80      - networking.k8s.io
81    resources:
82      - ingresses
83    verbs:
84      - get
85      - list
86      - watch
87  - apiGroups:
88      - ''
89    resources:
90      - events
91    verbs:
92      - create
93      - patch
94  - apiGroups:
95      - networking.k8s.io
96    resources:
97      - ingresses/status
98    verbs:
99      - update
100  - apiGroups:
101      - networking.k8s.io
102    resources:
103      - ingressclasses
104    verbs:
105      - get
106      - list
107      - watch
108---
109# Source: ingress-nginx/templates/clusterrolebinding.yaml
110apiVersion: rbac.authorization.k8s.io/v1
111kind: ClusterRoleBinding
112metadata:
113  labels:
114    helm.sh/chart: ingress-nginx-4.0.15
115    app.kubernetes.io/name: ingress-nginx
116    app.kubernetes.io/instance: ingress-nginx
117    app.kubernetes.io/version: 1.1.1
118    app.kubernetes.io/managed-by: Helm
119  name: ingress-nginx
120roleRef:
121  apiGroup: rbac.authorization.k8s.io
122  kind: ClusterRole
123  name: ingress-nginx
124subjects:
125  - kind: ServiceAccount
126    name: ingress-nginx
127    namespace: ingress-nginx
128---
129# Source: ingress-nginx/templates/controller-role.yaml
130apiVersion: rbac.authorization.k8s.io/v1
131kind: Role
132metadata:
133  labels:
134    helm.sh/chart: ingress-nginx-4.0.15
135    app.kubernetes.io/name: ingress-nginx
136    app.kubernetes.io/instance: ingress-nginx
137    app.kubernetes.io/version: 1.1.1
138    app.kubernetes.io/managed-by: Helm
139    app.kubernetes.io/component: controller
140  name: ingress-nginx
141  namespace: ingress-nginx
142rules:
143  - apiGroups:
144      - ''
145    resources:
146      - namespaces
147    verbs:
148      - get
149  - apiGroups:
150      - ''
151    resources:
152      - configmaps
153      - pods
154      - secrets
155      - endpoints
156    verbs:
157      - get
158      - list
159      - watch
160  - apiGroups:
161      - ''
162    resources:
163      - services
164    verbs:
165      - get
166      - list
167      - watch
168  - apiGroups:
169      - networking.k8s.io
170    resources:
171      - ingresses
172    verbs:
173      - get
174      - list
175      - watch
176  - apiGroups:
177      - networking.k8s.io
178    resources:
179      - ingresses/status
180    verbs:
181      - update
182  - apiGroups:
183      - networking.k8s.io
184    resources:
185      - ingressclasses
186    verbs:
187      - get
188      - list
189      - watch
190  - apiGroups:
191      - ''
192    resources:
193      - configmaps
194    resourceNames:
195      - ingress-controller-leader
196    verbs:
197      - get
198      - update
199  - apiGroups:
200      - ''
201    resources:
202      - configmaps
203    verbs:
204      - create
205  - apiGroups:
206      - ''
207    resources:
208      - events
209    verbs:
210      - create
211      - patch
212---
213# Source: ingress-nginx/templates/controller-rolebinding.yaml
214apiVersion: rbac.authorization.k8s.io/v1
215kind: RoleBinding
216metadata:
217  labels:
218    helm.sh/chart: ingress-nginx-4.0.15
219    app.kubernetes.io/name: ingress-nginx
220    app.kubernetes.io/instance: ingress-nginx
221    app.kubernetes.io/version: 1.1.1
222    app.kubernetes.io/managed-by: Helm
223    app.kubernetes.io/component: controller
224  name: ingress-nginx
225  namespace: ingress-nginx
226roleRef:
227  apiGroup: rbac.authorization.k8s.io
228  kind: Role
229  name: ingress-nginx
230subjects:
231  - kind: ServiceAccount
232    name: ingress-nginx
233    namespace: ingress-nginx
234---
235# Source: ingress-nginx/templates/controller-service-webhook.yaml
236apiVersion: v1
237kind: Service
238metadata:
239  labels:
240    helm.sh/chart: ingress-nginx-4.0.15
241    app.kubernetes.io/name: ingress-nginx
242    app.kubernetes.io/instance: ingress-nginx
243    app.kubernetes.io/version: 1.1.1
244    app.kubernetes.io/managed-by: Helm
245    app.kubernetes.io/component: controller
246  name: ingress-nginx-controller-admission
247  namespace: ingress-nginx
248spec:
249  type: ClusterIP
250  ports:
251    - name: https-webhook
252      port: 443
253      targetPort: webhook
254      appProtocol: https
255  selector:
256    app.kubernetes.io/name: ingress-nginx
257    app.kubernetes.io/instance: ingress-nginx
258    app.kubernetes.io/component: controller
259---
260# Source: ingress-nginx/templates/controller-service.yaml
261apiVersion: v1
262kind: Service
263metadata:
264  annotations:
265  labels:
266    helm.sh/chart: ingress-nginx-4.0.15
267    app.kubernetes.io/name: ingress-nginx
268    app.kubernetes.io/instance: ingress-nginx
269    app.kubernetes.io/version: 1.1.1
270    app.kubernetes.io/managed-by: Helm
271    app.kubernetes.io/component: controller
272  name: ingress-nginx-controller
273  namespace: ingress-nginx
274spec:
275  type: NodePort
276  externalTrafficPolicy: Local
277  ipFamilyPolicy: SingleStack
278  ipFamilies:
279    - IPv4
280  ports:
281    - name: http
282      port: 80
283      protocol: TCP
284      targetPort: http
285      appProtocol: http
286      nodePort: 32591 # Ingress HTTP 服务的端口号，可根据情况调整
287    - name: https
288      port: 443
289      protocol: TCP
290      targetPort: https
291      appProtocol: https
292      nodePort: 32592 # Ingress HTTPS 服务的端口号，可根据情况调整
293  selector:
294    app.kubernetes.io/name: ingress-nginx
295    app.kubernetes.io/instance: ingress-nginx
296    app.kubernetes.io/component: controller
297---
298# Source: ingress-nginx/templates/controller-deployment.yaml
299apiVersion: apps/v1
300kind: Deployment
301metadata:
302  labels:
303    helm.sh/chart: ingress-nginx-4.0.15
304    app.kubernetes.io/name: ingress-nginx
305    app.kubernetes.io/instance: ingress-nginx
306    app.kubernetes.io/version: 1.1.1
307    app.kubernetes.io/managed-by: Helm
308    app.kubernetes.io/component: controller
309  name: ingress-nginx-controller
310  namespace: ingress-nginx
311spec:
312  selector:
313    matchLabels:
314      app.kubernetes.io/name: ingress-nginx
315      app.kubernetes.io/instance: ingress-nginx
316      app.kubernetes.io/component: controller
317  revisionHistoryLimit: 10
318  minReadySeconds: 0
319  template:
320    metadata:
321      labels:
322        app.kubernetes.io/name: ingress-nginx
323        app.kubernetes.io/instance: ingress-nginx
324        app.kubernetes.io/component: controller
325    spec:
326      dnsPolicy: ClusterFirst
327      containers:
328        - name: controller
329          image: registry.cn-hangzhou.aliyuncs.com/google_containers/nginx-ingress-controller:v1.1.1
330          imagePullPolicy: IfNotPresent
331          lifecycle:
332            preStop:
333              exec:
334                command:
335                  - /wait-shutdown
336          args:
337            - /nginx-ingress-controller
338            - --publish-service=$(POD_NAMESPACE)/ingress-nginx-controller
339            - --election-id=ingress-controller-leader
340            - --controller-class=k8s.io/ingress-nginx
341            - --configmap=$(POD_NAMESPACE)/ingress-nginx-controller
342            - --validating-webhook=:8443
343            - --validating-webhook-certificate=/usr/local/certificates/cert
344            - --validating-webhook-key=/usr/local/certificates/key
345          securityContext:
346            capabilities:
347              drop:
348                - ALL
349              add:
350                - NET_BIND_SERVICE
351            runAsUser: 101
352            allowPrivilegeEscalation: true
353          env:
354            - name: POD_NAME
355              valueFrom:
356                fieldRef:
357                  fieldPath: metadata.name
358            - name: POD_NAMESPACE
359              valueFrom:
360                fieldRef:
361                  fieldPath: metadata.namespace
362            - name: LD_PRELOAD
363              value: /usr/local/lib/libmimalloc.so
364          livenessProbe:
365            failureThreshold: 5
366            httpGet:
367              path: /healthz
368              port: 10254
369              scheme: HTTP
370            initialDelaySeconds: 10
371            periodSeconds: 10
372            successThreshold: 1
373            timeoutSeconds: 1
374          readinessProbe:
375            failureThreshold: 3
376            httpGet:
377              path: /healthz
378              port: 10254
379              scheme: HTTP
380            initialDelaySeconds: 10
381            periodSeconds: 10
382            successThreshold: 1
383            timeoutSeconds: 1
384          ports:
385            - name: http
386              containerPort: 80
387              hostPort: 80
388              protocol: TCP
389            - name: https
390              hostPort: 443
391              containerPort: 443
392              protocol: TCP
393            - name: webhook
394              containerPort: 8443
395              protocol: TCP
396          volumeMounts:
397            - name: webhook-cert
398              mountPath: /usr/local/certificates/
399              readOnly: true
400          resources:
401            requests:
402              cpu: 100m
403              memory: 90Mi
404      nodeSelector:
405        kubernetes.io/os: linux
406      serviceAccountName: ingress-nginx
407      terminationGracePeriodSeconds: 300
408      volumes:
409        - name: webhook-cert
410          secret:
411            secretName: ingress-nginx-admission
412---
413# Source: ingress-nginx/templates/controller-ingressclass.yaml
414# We don't support namespaced ingressClass yet
415# So a ClusterRole and a ClusterRoleBinding is required
416apiVersion: networking.k8s.io/v1
417kind: IngressClass
418metadata:
419  labels:
420    helm.sh/chart: ingress-nginx-4.0.15
421    app.kubernetes.io/name: ingress-nginx
422    app.kubernetes.io/instance: ingress-nginx
423    app.kubernetes.io/version: 1.1.1
424    app.kubernetes.io/managed-by: Helm
425    app.kubernetes.io/component: controller
426  name: nginx
427  namespace: ingress-nginx
428spec:
429  controller: k8s.io/ingress-nginx
430---
431# Source: ingress-nginx/templates/admission-webhooks/validating-webhook.yaml
432# before changing this value, check the required kubernetes version
433# https://kubernetes.io/docs/reference/access-authn-authz/extensible-admission-controllers/#prerequisites
434apiVersion: admissionregistration.k8s.io/v1
435kind: ValidatingWebhookConfiguration
436metadata:
437  labels:
438    helm.sh/chart: ingress-nginx-4.0.15
439    app.kubernetes.io/name: ingress-nginx
440    app.kubernetes.io/instance: ingress-nginx
441    app.kubernetes.io/version: 1.1.1
442    app.kubernetes.io/managed-by: Helm
443    app.kubernetes.io/component: admission-webhook
444  name: ingress-nginx-admission
445webhooks:
446  - name: validate.nginx.ingress.kubernetes.io
447    matchPolicy: Equivalent
448    rules:
449      - apiGroups:
450          - networking.k8s.io
451        apiVersions:
452          - v1
453        operations:
454          - CREATE
455          - UPDATE
456        resources:
457          - ingresses
458    failurePolicy: Fail
459    sideEffects: None
460    admissionReviewVersions:
461      - v1
462    clientConfig:
463      service:
464        namespace: ingress-nginx
465        name: ingress-nginx-controller-admission
466        path: /networking/v1/ingresses
467---
468# Source: ingress-nginx/templates/admission-webhooks/job-patch/serviceaccount.yaml
469apiVersion: v1
470kind: ServiceAccount
471metadata:
472  name: ingress-nginx-admission
473  namespace: ingress-nginx
474  annotations:
475    helm.sh/hook: pre-install,pre-upgrade,post-install,post-upgrade
476    helm.sh/hook-delete-policy: before-hook-creation,hook-succeeded
477  labels:
478    helm.sh/chart: ingress-nginx-4.0.15
479    app.kubernetes.io/name: ingress-nginx
480    app.kubernetes.io/instance: ingress-nginx
481    app.kubernetes.io/version: 1.1.1
482    app.kubernetes.io/managed-by: Helm
483    app.kubernetes.io/component: admission-webhook
484---
485# Source: ingress-nginx/templates/admission-webhooks/job-patch/clusterrole.yaml
486apiVersion: rbac.authorization.k8s.io/v1
487kind: ClusterRole
488metadata:
489  name: ingress-nginx-admission
490  annotations:
491    helm.sh/hook: pre-install,pre-upgrade,post-install,post-upgrade
492    helm.sh/hook-delete-policy: before-hook-creation,hook-succeeded
493  labels:
494    helm.sh/chart: ingress-nginx-4.0.15
495    app.kubernetes.io/name: ingress-nginx
496    app.kubernetes.io/instance: ingress-nginx
497    app.kubernetes.io/version: 1.1.1
498    app.kubernetes.io/managed-by: Helm
499    app.kubernetes.io/component: admission-webhook
500rules:
501  - apiGroups:
502      - admissionregistration.k8s.io
503    resources:
504      - validatingwebhookconfigurations
505    verbs:
506      - get
507      - update
508---
509# Source: ingress-nginx/templates/admission-webhooks/job-patch/clusterrolebinding.yaml
510apiVersion: rbac.authorization.k8s.io/v1
511kind: ClusterRoleBinding
512metadata:
513  name: ingress-nginx-admission
514  annotations:
515    helm.sh/hook: pre-install,pre-upgrade,post-install,post-upgrade
516    helm.sh/hook-delete-policy: before-hook-creation,hook-succeeded
517  labels:
518    helm.sh/chart: ingress-nginx-4.0.15
519    app.kubernetes.io/name: ingress-nginx
520    app.kubernetes.io/instance: ingress-nginx
521    app.kubernetes.io/version: 1.1.1
522    app.kubernetes.io/managed-by: Helm
523    app.kubernetes.io/component: admission-webhook
524roleRef:
525  apiGroup: rbac.authorization.k8s.io
526  kind: ClusterRole
527  name: ingress-nginx-admission
528subjects:
529  - kind: ServiceAccount
530    name: ingress-nginx-admission
531    namespace: ingress-nginx
532---
533# Source: ingress-nginx/templates/admission-webhooks/job-patch/role.yaml
534apiVersion: rbac.authorization.k8s.io/v1
535kind: Role
536metadata:
537  name: ingress-nginx-admission
538  namespace: ingress-nginx
539  annotations:
540    helm.sh/hook: pre-install,pre-upgrade,post-install,post-upgrade
541    helm.sh/hook-delete-policy: before-hook-creation,hook-succeeded
542  labels:
543    helm.sh/chart: ingress-nginx-4.0.15
544    app.kubernetes.io/name: ingress-nginx
545    app.kubernetes.io/instance: ingress-nginx
546    app.kubernetes.io/version: 1.1.1
547    app.kubernetes.io/managed-by: Helm
548    app.kubernetes.io/component: admission-webhook
549rules:
550  - apiGroups:
551      - ''
552    resources:
553      - secrets
554    verbs:
555      - get
556      - create
557---
558# Source: ingress-nginx/templates/admission-webhooks/job-patch/rolebinding.yaml
559apiVersion: rbac.authorization.k8s.io/v1
560kind: RoleBinding
561metadata:
562  name: ingress-nginx-admission
563  namespace: ingress-nginx
564  annotations:
565    helm.sh/hook: pre-install,pre-upgrade,post-install,post-upgrade
566    helm.sh/hook-delete-policy: before-hook-creation,hook-succeeded
567  labels:
568    helm.sh/chart: ingress-nginx-4.0.15
569    app.kubernetes.io/name: ingress-nginx
570    app.kubernetes.io/instance: ingress-nginx
571    app.kubernetes.io/version: 1.1.1
572    app.kubernetes.io/managed-by: Helm
573    app.kubernetes.io/component: admission-webhook
574roleRef:
575  apiGroup: rbac.authorization.k8s.io
576  kind: Role
577  name: ingress-nginx-admission
578subjects:
579  - kind: ServiceAccount
580    name: ingress-nginx-admission
581    namespace: ingress-nginx
582---
583# Source: ingress-nginx/templates/admission-webhooks/job-patch/job-createSecret.yaml
584apiVersion: batch/v1
585kind: Job
586metadata:
587  name: ingress-nginx-admission-create
588  namespace: ingress-nginx
589  annotations:
590    helm.sh/hook: pre-install,pre-upgrade
591    helm.sh/hook-delete-policy: before-hook-creation,hook-succeeded
592  labels:
593    helm.sh/chart: ingress-nginx-4.0.15
594    app.kubernetes.io/name: ingress-nginx
595    app.kubernetes.io/instance: ingress-nginx
596    app.kubernetes.io/version: 1.1.1
597    app.kubernetes.io/managed-by: Helm
598    app.kubernetes.io/component: admission-webhook
599spec:
600  template:
601    metadata:
602      name: ingress-nginx-admission-create
603      labels:
604        helm.sh/chart: ingress-nginx-4.0.15
605        app.kubernetes.io/name: ingress-nginx
606        app.kubernetes.io/instance: ingress-nginx
607        app.kubernetes.io/version: 1.1.1
608        app.kubernetes.io/managed-by: Helm
609        app.kubernetes.io/component: admission-webhook
610    spec:
611      containers:
612        - name: create
613          image: registry.cn-hangzhou.aliyuncs.com/google_containers/kube-webhook-certgen:v1.1.1
614          imagePullPolicy: IfNotPresent
615          args:
616            - create
617            - --host=ingress-nginx-controller-admission,ingress-nginx-controller-admission.$(POD_NAMESPACE).svc
618            - --namespace=$(POD_NAMESPACE)
619            - --secret-name=ingress-nginx-admission
620          env:
621            - name: POD_NAMESPACE
622              valueFrom:
623                fieldRef:
624                  fieldPath: metadata.namespace
625          securityContext:
626            allowPrivilegeEscalation: false
627      restartPolicy: OnFailure
628      serviceAccountName: ingress-nginx-admission
629      nodeSelector:
630        kubernetes.io/os: linux
631      securityContext:
632        runAsNonRoot: true
633        runAsUser: 2000
634---
635# Source: ingress-nginx/templates/admission-webhooks/job-patch/job-patchWebhook.yaml
636apiVersion: batch/v1
637kind: Job
638metadata:
639  name: ingress-nginx-admission-patch
640  namespace: ingress-nginx
641  annotations:
642    helm.sh/hook: post-install,post-upgrade
643    helm.sh/hook-delete-policy: before-hook-creation,hook-succeeded
644  labels:
645    helm.sh/chart: ingress-nginx-4.0.15
646    app.kubernetes.io/name: ingress-nginx
647    app.kubernetes.io/instance: ingress-nginx
648    app.kubernetes.io/version: 1.1.1
649    app.kubernetes.io/managed-by: Helm
650    app.kubernetes.io/component: admission-webhook
651spec:
652  template:
653    metadata:
654      name: ingress-nginx-admission-patch
655      labels:
656        helm.sh/chart: ingress-nginx-4.0.15
657        app.kubernetes.io/name: ingress-nginx
658        app.kubernetes.io/instance: ingress-nginx
659        app.kubernetes.io/version: 1.1.1
660        app.kubernetes.io/managed-by: Helm
661        app.kubernetes.io/component: admission-webhook
662    spec:
663      containers:
664        - name: patch
665          image: registry.cn-hangzhou.aliyuncs.com/google_containers/kube-webhook-certgen:v1.1.1
666          imagePullPolicy: IfNotPresent
667          args:
668            - patch
669            - --webhook-name=ingress-nginx-admission
670            - --namespace=$(POD_NAMESPACE)
671            - --patch-mutating=false
672            - --secret-name=ingress-nginx-admission
673            - --patch-failure-policy=Fail
674          env:
675            - name: POD_NAMESPACE
676              valueFrom:
677                fieldRef:
678                  fieldPath: metadata.namespace
679          securityContext:
680            allowPrivilegeEscalation: false
681      restartPolicy: OnFailure
682      serviceAccountName: ingress-nginx-admission
683      nodeSelector:
684        kubernetes.io/os: linux
685      securityContext:
686        runAsNonRoot: true
687        runAsUser: 2000

这个文件中将 Ingress 的 HTTP 服务暴露在 32591 端口，将 HTTPS 服务暴露在 32592 端口。可根据实际情况修改，文件的其余部分基本不需要修改。

然后在该文件所在目录执行：

Plain Text

1kubectl apply -f ingress-nginx.yaml

通过下面命令查看节点状态：

Plain Text

1kubectl get pods -A

等待 ingress controller 节点运行状态变为 Ready 即可：

2、配置训练服务在 Ingress 中的转发规则。

新建一个ingress-http.yaml文件，并用文本编辑器打开，复制以下的内容并填写相应部分：

YAML

1apiVersion: networking.k8s.io/v1
2kind: Ingress
3metadata:
4  name: sugar-ml-ingress
5  namespace: sugar-ml
6  annotations:
7    kubernetes.io/ingress.class: 'nginx'
8    nginx.ingress.kubernetes.io/proxy-body-size: 2048m
9spec:
10  rules:
11    - http:
12        paths:
13          - path: /
14            pathType: Prefix
15            backend:
16              service:
17                name: sugar-ml-service
18                port:
19                  number: 54321

然后在该文件所在目录执行：

Plain Text

1kubectl apply -f ingress-http.yaml

部署训练服务

部署训练服务，新建一个train.yaml文件，并用文本编辑器打开，复制以下的内容并填写相应部分：

YAML

1apiVersion: apps/v1
2kind: StatefulSet
3metadata:
4  name: sugar-ml-stateful-set
5  namespace: sugar-ml
6spec:
7  serviceName: sugar-ml-service
8  podManagementPolicy: 'Parallel'
9  replicas: 3 # pod 数目，需要与 node 数目一致，一个 node 上部署一个 pod
10  selector:
11    matchLabels:
12      app: sugar-ml-k8s
13  template:
14    metadata:
15      labels:
16        app: sugar-ml-k8s
17    spec:
18      affinity:
19        podAntiAffinity:
20          requiredDuringSchedulingIgnoredDuringExecution:
21            - labelSelector:
22                matchExpressions:
23                  - key: app
24                    operator: In
25                    values:
26                      - sugar-ml-k8s
27              topologyKey: 'kubernetes.io/hostname'
28      containers:
29        - name: sugar-ml-k8s
30          image: 'registry.baidubce.com/sugarbi/sugar-ml-train:1.1.1'
31          resources:
32            requests:
33              memory: '4Gi'
34          ports:
35            - containerPort: 54321
36              protocol: TCP
37          readinessProbe:
38            httpGet:
39              path: /kubernetes/isLeaderNode
40              port: 8080
41            initialDelaySeconds: 5
42            periodSeconds: 5
43            failureThreshold: 1
44          livenessProbe:
45            exec:
46              command:
47                - /bin/sh
48                - -c
49                - sh liveness.sh
50            failureThreshold: 5
51            initialDelaySeconds: 200
52            periodSeconds: 10
53            successThreshold: 1
54            timeoutSeconds: 10
55          env:
56            - name: sugar_ml_service_dns
57              value: sugar-ml-service.sugar-ml.svc.cluster.local
58            - name: sugar_ml_lookup_timeout
59              value: '1800'
60            - name: sugar_ml_node_count
61              value: '3' # 与上面 replicas 值对应
62            - name: sugar_ml_service_endpoint
63              value: http://ingress-nginx-controller.ingress-nginx # 访问上一节中配置的 Ingress http 端口的地址。也可以是 http://XXX.XXX.XX.XX:32591，将 XXX 替换为机器 IP 即可
64---
65apiVersion: v1
66kind: Service
67metadata:
68  name: sugar-ml-service
69  namespace: sugar-ml
70spec:
71  type: ClusterIP
72  clusterIP: None
73  selector:
74    app: sugar-ml-k8s
75  ports:
76    - protocol: TCP
77      port: 54321

然后在该文件所在目录执行：

Plain Text

1kubectl apply -f train.yaml

通过下面命令查看节点状态：

Plain Text

1kubectl get pods -A

等待全部节点运行状态都变为 Running，其中 1 个节点 Ready 即可。例如这里部署了 3 个 POD：

在 Sugar BI 中开启智能预测功能

配置环境变量并重启

训练和推理服务运行成功之后，就可以在 Sugar BI 中来启用该功能了，需要在 Sugar BI 本身的 env 配置文件中增加以下配置（同理，Swarm、Kubernetes 部署方式也是同样的添加相应的环境变量配置即可）。

Plain Text

1# 训练服务的地址，XXXXXX 需要替换为训练服务所在的机器地址
2# 如果是按照上面文档的 Docker 方式部署，端口号一般为 8090
3# 如果是按照上面文档的 Kubernetes 方式部署，端口号一般为 32591
4sugar_ml_train_endpoint=http://XXXXXX:8090
5# 推理服务的地址，XXXXXX 需要替换为推理服务所在的机器地址
6# 如果是按照上面文档的 Docker 方式部署，端口号一般为 8091
7# 如果是按照上面文档的 Kubernetes 方式部署，端口号一般为 32590
8sugar_ml_predict_endpoint=http://XXXXXX:8091/predict
9
10# 预测模型在对象存储中的 bucket，注意和上面推理服务中的 sugar_ml_model_bucket 对应，一般为 sugar-ml
11sugar_ml_model_bucket=sugar-ml
12# 同时进行训练的任务的最大数目，根据机器配置和训练数据量配置
13export sugar_ml_max_running_train_num=10

env 文件更新之后需要重启 Sugar BI

验证部署

重启完成后，可进入「组织管理」->「外部模块部署验证」进行验证。

功能使用

智能预测功能的使用可参考模型训练

外观主题设置

openAPI

百度数据可视化 SUGAR

百度数据可视化 SUGAR

开启智能预测功能

前期准备

安装环境

软件环境

硬件环境

下载

1、百度智能云的 docker 镜像服务

2、直接在线下载

运行训练和推理服务

一、Docker 单机运行

运行训练服务镜像

运行推理服务镜像

二、kubernetes 集群部署

部署推理服务

部署 Ingress

部署训练服务

在 Sugar BI 中开启智能预测功能

配置环境变量并重启

验证部署

功能使用