Skip to main content

Deploy Dedicated Computing Services

Configure Dedicated Computing Server​

Joining the Kubernetes Cluster​

  1. Deploy the Kubernetes node environment on the dedicated computing server in advance.

  2. Obtain the command to add a new node to the cluster from the Kubernetes master node:

    kubeadm token create --print-join-command

  3. Execute the join cluster command on the dedicated computing server:

    kubeadm join 192.168.1.100:6443 --token 3nwjzw.pdod3r27lnqqhi0x --discovery-token-ca-cert-hash sha256:a84445303xxxxxxxxxxxxxxxxxxx175a2

Download Images​

  1. Download the computing instance image (must be performed separately on the microservice nodes and dedicated computing nodes).

    crictl pull registry.cn-hangzhou.aliyuncs.com/mdpublic/mingdaoyun-computinginstance:7.5.4

Add Taint and Label​

  1. Obtain the node name of the dedicated computing node:

    kubectl get node

  2. Add taint and label to the dedicated computing node:

    kubectl taint node $your_node_name md=workflowcompute:NoSchedule
    kubectl label node $your_node_name md=workflowcompute
    • Note, you need to replace $your_node_name with the actual node name

  3. Add label to the dedicated computing node

Deploy Dedicated Computing Services​

note

  • Dedicated computing is supported starting from version v5.1.0. When installing the manager version >=5.1.0 for the first time, a default computinginstance.yaml file will be generated. Usually, you only need to modify the existing configuration file when configuring the dedicated computing service.

  • If installing manager version <5.1.0 during the first deployment, a new computinginstance.yaml file needs to be added and the relevant scripts need to be modified.

  1. Execute cat $KUBECONFIG to obtain the $KUBECONFIG content (corresponding to the /etc/kubernetes/admin.conf file in Kubernetes):

    cat $KUBECONFIG

    Output sample:

    apiVersion: v1
    clusters:
    - cluster:
        certificate-authority-data: LS0tLS1CRUdJTiBDRVJUSUZJQ0FURS0tLS0tCk1JSUMvakNDQWVhZ0...
        server: https://192.168.0.150:6443
      name: kubernetes
    contexts:
    - context:
        cluster: kubernetes
        user: kubernetes-admin
      name: kubernetes-admin@kubernetes
    current-context: kubernetes-admin@kubernetes
    kind: Config
    preferences: {}
    users:
    - name: kubernetes-admin
      user:
        client-certificate-data: LS0tLS1CRUdJTiBDRVJUSUZJQ0...
    • In the next step, copy the cat $KUBECONFIG output content to the computinginstance.yaml file, make sure to add four spaces before each line when copying

  2. Modify the /data/mingdao/script/kubernetes/computinginstance.yaml file:

    For the first deployment, make sure to modify the following points in the computinginstance.yaml file

    • kafka.brokers: Change to the kafka address connected by the microservice
    • syncStatus.mongoUri: Modify to the actual mongodb connection information
    • image: Change to the same computing instance version number as the microservice
    • kubeconfig.yaml: Change to the actual content of $KUBECONFIG, remember to add four spaces before each line of the original $KUBECONFIG content

    Sample Modification:

    apiVersion: v1
    kind: ConfigMap
    metadata:
      name: computinginstance
      namespace: default
    data:
      config.yaml: |-
        server:
          listen:
            host: "0.0.0.0"
            port: "9157"
        common:
          kubernetes:
            configFile: "/usr/local/computinginstance/kubeconfig.yaml"
            namespace: "default"
            configmapTemplate: "/usr/local/computinginstance/configmap-workflowcompute.yaml"
            deploymentTemplate: "/usr/local/computinginstance/deployment-workflowcompute.yaml"
          kafka:
            brokers: "192.168.0.144:9092"  # Modify to the kafka address connected by the microservice
            workflowTopicPrefix: "WorkFlow-"
            workSheetTopicPrefix: "WorkSheet-"
            workflowConsumerIdPrefix: "md-workflow-consumer-"
            replicationFactor: 1
            deleteTopic: true
          callback:
            url: "http://computingschedule:9158"
            createInterval: 120000 #ms
            deleteInterval: 120000 #ms
          syncStatus:
            mongoUri: "mongodb://mingdao:123456@192.168.0.144:27017/mdIdentification"  # Modify to the actual mongodb connection information
            interval: 30000 # ms
          model:
            10:
              replicas: 1     # Number of instances
              thread: 2       # Number of threads per type, total = 5*thread*replicas=10
              cpu: 2          # Maximum CPU cores per instance, total = cpu*replicas=2
              memory: 4       # Maximum memory in GB per instance, total = memory*replicas=4
            20:
              replicas: 1
              thread: 4
              cpu: 4
              memory: 8
            50:
              replicas: 1
              thread: 10
              cpu: 8
              memory: 16
            100:
              replicas: 1
              thread: 20
              cpu: 16
              memory: 32

      configmap-workflowcompute.yaml: |-
        apiVersion: v1
        kind: ConfigMap
        metadata:
          name: workflowcompute
        data:
          application-www-computing.properties: |-
            md.resource.consumer.config.maps={\
              'resourceId': 'CONFIGMAP_INSTANCEID', \
              'wfTopic': 'WorkFlow-CONFIGMAP_INSTANCEID', \
              'wsTopic': 'WorkSheet-CONFIGMAP_INSTANCEID', \
              'partition': 'CONFIGMAP_WORKFLOW_PARTITION' \
            }
            md.kafka.consumer.topic=WorkSheet-CONFIGMAP_INSTANCEID
            md.kafka.consumer.group.id=md-workflow-consumer-CONFIGMAP_INSTANCEID
            md.kafka.consumer.concurrency=CONFIGMAP_WORKFLOW_THREAD
            md.kafka.batch.topic=WorkSheet-CONFIGMAP_INSTANCEID
            md.kafka.batch.concurrency=CONFIGMAP_WORKFLOW_THREAD
            spring.kafka.batch.topic=WorkFlow-CONFIGMAP_INSTANCEID
            spring.kafka.batch.concurrency=CONFIGMAP_WORKFLOW_THREAD
            spring.kafka.button.topic=WorkFlow-CONFIGMAP_INSTANCEID
            spring.kafka.button.concurrency=CONFIGMAP_WORKFLOW_THREAD
            spring.kafka.process.topic=WorkFlow-CONFIGMAP_INSTANCEID
            spring.kafka.process.concurrency=CONFIGMAP_WORKFLOW_THREAD
            spring.kafka.consumer.topic=WorkFlow-CONFIGMAP_INSTANCEID
            spring.kafka.consumer.group.id=md-workflow-consumer-CONFIGMAP_INSTANCEID
            spring.kafka.consumer.concurrency=CONFIGMAP_WORKFLOW_THREAD
            spring.kafka.properties.partition=CONFIGMAP_WORKFLOW_PARTITION
            grpc.client.MDWorksheetService.address=static://127.0.0.1:9422
            spring.kafka.router.topic=WorkFlow-CONFIGMAP_INSTANCEID

      deployment-workflowcompute.yaml: |-
        apiVersion: apps/v1
        kind: Deployment
        metadata:
          name: workflowcompute-DEPLOYMENT_INSTANCEID
          labels:
            app: workflowcompute-DEPLOYMENT_INSTANCEID
            md-service: workflowcompute
        spec:
          replicas: DEPLOYMENT_REPLICAS
          selector:
            matchLabels:
              app: workflowcompute-DEPLOYMENT_INSTANCEID
          template:
            metadata:
              labels:
                app: workflowcompute-DEPLOYMENT_INSTANCEID
                md-service: workflowcompute
              annotations:
                md-update: '20231228184263'
            spec:
              imagePullSecrets:
              - name: hub.mingdao.com
              tolerations:
                - key: "md"
                  operator: "Equal"
                  value: "workflowcompute"
                  effect: "NoSchedule"
              nodeSelector:
                md: workflowcompute
              containers:
                - name: workflow-consumer
                  image: registry.cn-hangzhou.aliyuncs.com/mdpublic/mingdaoyun-computinginstance:7.5.4  # computinginstance image version
                  env:
                  - name: ENV_SERVERID
                    value: "single:workflowconsumer"
                  command:
                    - sh
                    - -c
                    - |
                      cp /usr/local/MDPrivateDeployment/workflowconsumer/application-www-computing.properties.template /usr/local/MDPrivateDeployment/workflowconsumer/application-www-computing.properties
                      sed -i s/CONFIGMAP_INSTANCEID/DEPLOYMENT_INSTANCEID/g /usr/local/MDPrivateDeployment/workflowconsumer/application-www-computing.properties
                      sed -i s/CONFIGMAP_WORKFLOW_THREAD/DEPLOYMENT_WORKFLOW_THREAD/g /usr/local/MDPrivateDeployment/workflowconsumer/application-www-computing.properties
                      sed -i s/CONFIGMAP_WORKFLOW_PARTITION/DEPLOYMENT_WORKFLOW_PARTITION/g /usr/local/MDPrivateDeployment/workflowconsumer/application-www-computing.properties
                      cat /usr/local/MDPrivateDeployment/workflowconsumer/application-www-computing.properties
                      sleep 20
                      exec /Housekeeper/main -config /Housekeeper/config.yaml
                  resources:
                    limits:
                      memory: DEPLOYMENT_MEMORYGi
                      cpu: DEPLOYMENT_CPU
                    requests:
                      memory: 1Gi
                      cpu: 0.25
                  volumeMounts:
                    - name: tz-config
                      mountPath: /etc/localtime
                    - name: workflowcompute
                      mountPath: /usr/local/MDPrivateDeployment/workflowconsumer/application-www-computing.properties.template
                      subPath: application-www-computing.properties
                - name: worksheetservice
                  image: registry.cn-hangzhou.aliyuncs.com/mdpublic/mingdaoyun-computinginstance:7.5.4  # computinginstance image version
                  env:
                  - name: ENV_SERVERID
                    value: "single:worksheet"
                  command:
                    - sh
                    - -c
                    - |
                      cat /usr/local/MDPrivateDeployment/worksheet/Config/appsettingsMain.json
                      exec /Housekeeper/main -config /Housekeeper/config.yaml
                  resources:
                    limits:
                      memory: DEPLOYMENT_MEMORYGi
                      cpu: DEPLOYMENT_CPU
                    requests:
                      memory: 1Gi
                      cpu: 0.25
                  readinessProbe:
                    tcpSocket:
                      port: 9422
                    initialDelaySeconds: 10
                    periodSeconds: 10
                  livenessProbe:
                    tcpSocket:
                      port: 9422
                    initialDelaySeconds: 60
                    periodSeconds: 10
                  volumeMounts:
                    - name: tz-config
                      mountPath: /etc/localtime
              volumes:
                - name: tz-config
                  hostPath:
                    path: /usr/share/zoneinfo/Etc/GMT-8
                - name: workflowcompute
                  configMap:
                    name: workflowcompute
                    items:
                    - key: application-www-computing.properties
                      path: application-www-computing.properties

      kubeconfig.yaml: |-    # Change to the actual content of $KUBECONFIG below, make sure to add four spaces before each line of the original $KUBECONFIG content
        apiVersion: v1
        clusters:
        - cluster:
            certificate-authority-data: LS0tLS1CRUdJTiBDRVJUSUZJQ0FURS0tLS0tCk1JSUMvakNDQWVhZ0...
            server: https://192.168.0.150:6443
          name: kubernetes
        contexts:
        - context:
            cluster: kubernetes
            user: kubernetes-admin
          name: kubernetes-admin@kubernetes
        current-context: kubernetes-admin@kubernetes
        kind: Config
        preferences: {}
        users:
        - name: kubernetes-admin
          user:
            client-certificate-data: LS0tLS1CRUdJTiBDRVJUSUZJQ0...
            client-key-data: LS0tLS1CRUdJTiBSU0EgUFJJVkFURSBLRVktLS0tLQ...
    ---

    apiVersion: apps/v1
    kind: Deployment
    metadata:
      name: computinginstance
      namespace: default
    spec:
      replicas: 1
      selector:
        matchLabels:
          app: computinginstance
      template:
        metadata:
          labels:
            app: computinginstance
            dir: grpc
          annotations:
            md-update: "20231228184309"
        spec:
          imagePullSecrets: 
          - name: hub.mingdao.com
          containers:
          - name: computinginstance 
            image: registry.cn-hangzhou.aliyuncs.com/mdpublic/mingdaoyun-computinginstance:7.5.4  # computinginstance image version
            resources:
              limits:
                cpu: "2"
                memory: 2G
              requests:
                cpu: "0.01"
                memory: 128M
            readinessProbe:
              tcpSocket:
                port: 9157
              initialDelaySeconds: 3
              periodSeconds: 3
            livenessProbe:
              tcpSocket:
                port: 9157
              initialDelaySeconds: 20
              periodSeconds: 10
            volumeMounts:
            - name: configfile
              mountPath: /usr/local/computinginstance/config.yaml
              subPath: config.yaml
            - name: configfile
              mountPath: /usr/local/computinginstance/configmap-workflowcompute.yaml
              subPath: configmap-workflowcompute.yaml
            - name: configfile
              mountPath: /usr/local/computinginstance/deployment-workflowcompute.yaml
              subPath: deployment-workflowcompute.yaml
            - name: configfile
              mountPath: /usr/local/computinginstance/kubeconfig.yaml
              subPath: kubeconfig.yaml
            - name: tz-config
              mountPath: /etc/localtime
          volumes:
          - name: tz-config
            hostPath:
              path: /usr/share/zoneinfo/Asia/Shanghai
          - name: configfile
            configMap:
              name: computinginstance
              items:
              - key: config.yaml
                path: config.yaml
              - key: configmap-workflowcompute.yaml
                path: configmap-workflowcompute.yaml
              - key: deployment-workflowcompute.yaml
                path: deployment-workflowcompute.yaml
              - key: kubeconfig.yaml
                path: kubeconfig.yaml

    ---

    apiVersion: v1
    kind: Service
    metadata:
      name: computinginstance
      namespace: default
    spec:
      selector:
        app: computinginstance
      ports:
      - name: http-computinginstance
        port: 9157
        targetPort: 9157
        nodePort: 9157
      type: NodePort

  3. Restart the service:

    cd /data/mingdao/script/kubernetes/
    bash restart.sh

Verification​

In the default namespace, fetch all deployments whose names start with workflowcompute- and computinginstance

ns="default"

deployments=$(kubectl -n $ns get deployments -o jsonpath='{.items[*].metadata.name}' | tr ' ' '\n' | grep '^workflowcompute-\|computinginstance')

for deploy in $deployments
do
  echo "Namespace: $ns, Deployment: $deploy"
  
  # Retrieve and print the deployment images
  kubectl -n $ns get deployment $deploy -o jsonpath='{.spec.template.spec.containers[*].image}'
  echo
done

Output example:

Namespace: default, Deployment: computinginstance
registry.cn-hangzhou.aliyuncs.com/mdpublic/mingdaoyun-computinginstance:7.5.4
Namespace: default, Deployment: workflowcompute-7e4wf0fea4ho
registry.cn-hangzhou.aliyuncs.com/mdpublic/mingdaoyun-computinginstance:7.5.4
  • Deployments starting with workflowcompute- exist only after a dedicated compute resource has been created in the UI

Check whether taints have taken effect

kubectl get pod -o wide | grep workflowcompute-
  • Normally, the containers should be running on the dedicated compute nodes

Check Kafka

/usr/local/kafka/bin/kafka-consumer-groups.sh --bootstrap-server ${ENV_KAFKA_ENDPOINTS:=127.0.0.1:9092} --describe --group md-workflow-consumer-7e4wf0fea4ho
  • Here, resource ID 7e4wf0fea4ho is used as an example