nginx代理ambassador，再转到mlfow-tracking服务

这个服务的代理，相对于服务网关来说，有些典型，

今天调通了，作个记录。

一，nginx配置

upstream ai_ambassador {
    ip_hash;
    server 1.2.3.4:30080;
}

server {
    listen       8080;
    server_name  localhost;
    client_max_body_size 500m;
    proxy_connect_timeout    600;
    proxy_read_timeout       600;
    proxy_send_timeout       600;

    add_header 'Access-Control-Allow-Origin' '*';
    proxy_http_version 1.1;
    proxy_set_header Connection "";

    location / {
        proxy_pass  http://ai_ambassador;
        proxy_set_header Host $host;
        proxy_set_header X-Real-Scheme $scheme;
        proxy_set_header X-Real-IP $remote_addr;
        proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
    }

}

二， zipkin

记住红线部署，不然，nginx会抱怨zipkin重定义次数太多，因为/zipkin本身的服务里使用了302跳转。

---
apiVersion: v1
kind: Service
metadata:
  name: zipkin
  annotations:
    getambassador.io/config: |
      ---
      apiVersion: ambassador/v1
      kind: TracingService
      name: tracing
      service: zipkin:9411
      driver: zipkin
      ---
      apiVersion: ambassador/v1
      kind: Mapping
      name: zipkin_mapping
      prefix: /zipkin
      rewrite: /zipkin/
      service: zipkin:9411
spec:
  selector:
    app: zipkin
  ports:
  - port: 9411
    name: http
    targetPort: 9411
    # nodePort: 32764
  # type: NodePort
---
apiVersion: extensions/v1beta1
kind: Deployment
metadata:
  name: zipkin
spec:
  replicas: 1
  strategy:
    type: RollingUpdate
  template:
    metadata:
      labels:
        app: zipkin
    spec:
      containers:
      - name: zipkin
        image: harbor.xxx.cn/3rd_part/openzipkin/zipkin:2.16
        imagePullPolicy: IfNotPresent
        ports:
        - name: http
          containerPort: 9411

三，指定ambassador的nodeport端口。

---
apiVersion: v1
kind: Service
metadata:
  name: ambassador
spec:
  type: NodePort
  ports:
  - name: http
    port: 80
    targetPort: 8080
    protocol: TCP
    nodePort: 30080
  selector:
    service: ambassador

四，由于要将mlfow tracking的数据放到mysql中，先建一个mysql服务(嘿嘿，由于拼写错误，ml flow tracking和ml flow tracing没分清)。由于是测试，密码随意 ，没有将数据，配置之类挂载出来。

apiVersion: apps/v1
kind: Deployment
metadata:
  name: mlflow-tracing-mysql
spec:
  replicas: 1
  selector:
    matchLabels:
      name: mlflow-tracing-mysql
  template:
    metadata:
      labels:
        name: mlflow-tracing-mysql
    spec:
      containers:
      - name: mysql
        image: harbor.xxx.cn/3rd_part/mysql:5.7.24
        imagePullPolicy: IfNotPresent
        ports:
        - containerPort: 3306
        env:
        - name: MYSQL_ROOT_PASSWORD
          value: "xxxx"
---
kind: Service
apiVersion: v1
metadata:
  name: mlflow-tracing-mysql
spec:
  type: NodePort
  ports:
  - name: mlflow-tracing-mysql
    port: 3306
    targetPort: 3306
    protocol: TCP
  selector:
    name: mlflow-tracing-mysql

这一步完了，好像可能，要确认root用户可远程访问，我好像还自己建好了mlflow库，可能不建不行吧，毕竟人家是直接使用的。

如果更新了hdfs这些配置，最好也要重建数据库，因为这些配置是第一次连接时，写进数据库了，或是自己进数据库修改吧。

五，mlflow tracking配置，这是重头戏。

apiVersion: apps/v1
kind: Deployment
metadata:
  name: mlflow-tracking
spec:
  replicas: 1
  selector:
    matchLabels:
      name: mlflow-tracking
  template:
    metadata:
      labels:
        name: mlflow-tracking
    spec:
      imagePullSecrets:
      - name: harborsecret
      containers:
      - name: mlflow-tracing
        image: harbor.xxx.cn/mlflow/mlflow-tracing:v1.2
        imagePullPolicy: IfNotPresent
        # command: ['sh', '-c', 'echo Hello Kubernetes! && sleep 360000']
        command: ['sh', '-c', 'mlflow server --backend-store-uri mysql+pymysql://root:xxx@mlflow-tracing-mysql:3306/mlflow --default-artifact-root hdfs://xxxx:8020/ml_model/mlflow --host 0.0.0.0']
        ports:
        - containerPort: 5000
        env:
        - name: MLFRACKING_TOKEN
          value: "no use"
---
kind: Service
apiVersion: v1
metadata:
  name: mlflow-tracking
  annotations:
    getambassador.io/config: |
      ---
      apiVersion: ambassador/v1
      kind: Mapping
      name: mlflow_tracking_mapping
      prefix: /mlflow-tracking/
      # rewrite: /
      service: mlflow-tracking:5000
spec:
  type: NodePort
  ports:
  - name: mlflow-tracking
    port: 5000
    targetPort: 5000
    protocol: TCP
    # nodePort: 30080
  selector:
    name: mlflow-tracking

六，mlflow tracking的dockerfile肿么写的呢？look,如果条件所限，请使用http代码，如果国外慢，则pip时，使用国内镜像。

FROM harbor.xxx.cn/3rd_part/continuumio/miniconda3:4.7.10

RUN export http_proxy=http://xxx.local:8080 \
    && export https_proxy=http://xxx.local:8080 \
    && export ftp_proxy=xxx.local:8080 \
    && pip install pymysql mlflow  -i http://pypi.douban.com/simple --trusted-host pypi.douban.com \
    && conda install hdfs3 -c conda-forge \
    && conda clean -y -all \
    && rm -rf ~/.cache/pip

ENV MLFLOW_HDFS_DRIVER=libhdfs3

EXPOSE 5000:5000

七，七星连珠之后，即可访问nginx上代理的mlflow tracking服务啦。。