Skip to content

Latest commit

 

History

History
152 lines (114 loc) · 6.28 KB

release-pipeline.md

File metadata and controls

152 lines (114 loc) · 6.28 KB
Raw

Setting Release Pipeline for Model Deployment

This Azure DevOps release pipeline will deploy trained model to Azure Container Instance or Azure Kubernetes Services every time a new model is registered in the Azure Machine Learning Service workspace.

Releasepipeline

Pre-requisite Steps

Install the Azure Machine Learning extension from the marketplace to the desired Azure DevOps account.

Tasks

  1. Add Release Artifacts (GitHub and AzureML Model)
  2. Add Deployment Stage (ACI or AKS)
  3. Trigger Pipeline and view API being served

01 - Add Release Artifacts

Source Code artifact

Fork this GitHub repository if it hasn't been done already and then add the associated GitHub artifact to a new release pipeline.

GitHub Artifact

AzureML Model Artifact

AzureMLArtifact

Enabling the continuous deployment trigger on this Azure ML artifact will trigger the release pipeline every time a new version of the model is registered in the Azure Machine Learning workspace cdtrigger

02 – Add Deployment Stage

There are two deployment options for the model using the AzureML service to make it easy to package up the model file into a container and serve it via an API. After the common steps, choose either 2a to deploy to Azure Container Instances (ACI) or 2b to deploy to Azure Kubernetes Services (AKS).

Common steps in both the stages

Add the Azure CLI task and then add the following code to the inline script:

02.a – Deploy to ACI

ACI Deployment Config

containerResourceRequirements:
  cpu: 1
  memoryInGB: 1
computeType: ACI

02.b – Deploy to AKS

AKS Deployment Config

computeType: AKS
autoScaler:
    autoscaleEnabled: True
    minReplicas: 1
    maxReplicas: 3
    refreshPeriodInSeconds: 10
    targetUtilization: 70
authEnabled: True
containerResourceRequirements:
    cpu: 1
    memoryInGB: 2
appInsightsEnabled: False
scoringTimeoutMs: 5000
maxConcurrentRequestsPerContainer: 2
maxQueueWaitMs: 5000
sslEnabled: True

03 - Trigger Pipeline and view API

After the Azure CLI task has been configured either deploy a new model to the attached artifact or manually trigger a release with the latest model artifact.

Once it is complete, view the log files from the executed Azure CLI task to determine the API's URL.

With ACI it should look something like this where the scoringUri is where the API is served.

2019-07-06T00:59:41.7979279Z ACI service creation operation finished, operation "Succeeded"
2019-07-06T00:59:41.7979391Z {
2019-07-06T00:59:41.7979526Z   "computeType": "ACI",
2019-07-06T00:59:41.7979670Z   "imageId": "torchcnnnaci:3",
2019-07-06T00:59:41.7979772Z   "modelDetails": "torchcnn [Ver. 6]",
2019-07-06T00:59:41.7979920Z   "name": "torchcnnnaci",
2019-07-06T00:59:41.7980058Z   "properties": {
2019-07-06T00:59:41.7980168Z     "azureml.git.commit": "971e9eb8326afed5994e781a1c03a33221ce2092",
2019-07-06T00:59:41.7980326Z     "azureml.git.dirty": "False",
2019-07-06T00:59:41.7981217Z     "azureml.git.repository_uri": "https://github.com/Azure-Samples/MLOpsDatabricks.git",
2019-07-06T00:59:41.7981377Z     "mlflow.source.git.commit": "971e9eb8326afed5994e781a1c03a33221ce2092",
2019-07-06T00:59:41.7981836Z     "mlflow.source.git.repoURL": "https://github.com/Azure-Samples/MLOpsDatabricks.git"
2019-07-06T00:59:41.7982010Z   },
2019-07-06T00:59:41.7982142Z   "runtimeType": "Python",
2019-07-06T00:59:41.7982917Z   "scoringUri": "http://ed1a4489-1219-4196-9638-32bf81750c15.eastus.azurecontainer.io/score",
2019-07-06T00:59:41.7983100Z   "state": "Healthy",
2019-07-06T00:59:41.7983231Z   "tags": "",
2019-07-06T00:59:41.7983707Z   "updatedAt": "2019-07-06T00:58:09.752479+00:00"
2019-07-06T00:59:41.7983853Z }

With AKS it should look something like this where the scoringUri is where the API is served.

2019-07-06T00:59:41.9345485Z Creating service
2019-07-06T00:59:41.9346166Z AKS service creation operation finished, operation "Succeeded"
2019-07-06T00:59:41.9346370Z {
2019-07-06T00:59:41.9346450Z   "computeType": "AKS",
2019-07-06T00:59:41.9346612Z   "imageId": "torchcnnaks:2",
2019-07-06T00:59:41.9346709Z   "modelDetails": "torchcnn [Ver. 6]",
2019-07-06T00:59:41.9346983Z   "name": "torchcnnaks",
2019-07-06T00:59:41.9347106Z   "properties": {
2019-07-06T00:59:41.9347227Z     "azureml.git.commit": "971e9eb8326afed5994e781a1c03a33221ce2092",
2019-07-06T00:59:41.9347335Z     "azureml.git.dirty": "False",
2019-07-06T00:59:41.9348191Z     "azureml.git.repository_uri": "https://github.com/Azure-Samples/MLOpsDatabricks.git",
2019-07-06T00:59:41.9348372Z     "mlflow.source.git.commit": "971e9eb8326afed5994e781a1c03a33221ce2092",
2019-07-06T00:59:41.9348712Z     "mlflow.source.git.repoURL": "https://github.com/Azure-Samples/MLOpsDatabricks.git"
2019-07-06T00:59:41.9348824Z   },
2019-07-06T00:59:41.9348891Z   "runtimeType": "Python",
2019-07-06T00:59:41.9349198Z   "scoringUri": "http://52.165.220.48:80/api/v1/service/torchcnnaks/score",
2019-07-06T00:59:41.9349305Z   "state": "Healthy",
2019-07-06T00:59:41.9349378Z   "tags": "",
2019-07-06T00:59:41.9349716Z   "updatedAt": "2019-07-06T00:58:34.413656+00:00"
2019-07-06T00:59:41.9349822Z }

See a sample implementation for release pipeline here