azure-pipelines.yml

# Sample pipeline that grabs a model from MLFlow, creates a Docker image and uploads it to Azure Container Registry

# These parameters are set by the MLFlow webhook that calls this pipeline
parameters:
- name: run_id
  type: string
- name: model_version
  type: string
- name: databricks_host
  type: string
- name: databricks_token
  type: string

trigger: none

pool:
  vmImage: ubuntu-latest

steps:

- script: echo Parameters sent to pipeline ${{ parameters.run_id }} and ${{ parameters.model_version }}
  displayName: 'Display parameters'
  
- task: DockerInstaller@0
  inputs:
    dockerVersion: '17.09.0-ce'

- script: pip install mlflow
  displayName: 'Installing dependencies'
  enabled: 'true'

- script: |
    export MLFLOW_TRACKING_URI=databricks
    export DATABRICKS_HOST=${{ parameters.databricks_host }}
    export DATABRICKS_TOKEN=${{ parameters.databricks_token }}
    mlflow artifacts download -r "${{ parameters.run_id }}" -d "model_dir"
    ls -la
  displayName: 'Configure MLFlow environmental variables'
  enabled: 'true'

- script: |
    mlflow models generate-dockerfile -m "model_dir/model/" -d "mlflow_docker" --enable-mlserver
  displayName: 'Create Dockerfile'
  enabled: 'true'

- script: |
    docker build -t "solacc.azurecr.io/ml-model:v.${{ parameters.model_version }}" mlflow_docker/
  displayName: 'Create Docker Image'
  enabled: 'true'

- task: Docker@2
  displayName: 'Push Docker Image to ACR'
  inputs:
    containerRegistry: 'solacc_acr'
    repository: 'ml-model'
    command: 'push'
    tags: 'v.${{ parameters.model_version }}'