search

microsoft / MLOpsPython

MLOps using Azure ML Services and Azure DevOps
MIT License
1.19k stars 1.09k forks source link

Unable to execute QA (ACI) stage. Path name issue. #58

Closed KBaroni closed 4 years ago

KBaroni commented 5 years ago

According to the get started document, there are 2 path variables that need to be configured for the Azure ML Model Deploy task in QA stage:

After checking to make sure both training and QA pipeline are using ubuntu-16.04, execution fails with this message:
[error]ENOENT: no such file or directory, stat /home/vsts/work/r1/a/_ci-build/mlops-pipelines/code/scoring/deployment_config_aci.yml

In checking the logs for the training pipeline, this is where the files should have been downloaded to:

So, i'm not sure why the execution is failing to find the files in the location. Any suggestions?

Thank you.

KBaroni commented 5 years ago

I have tried multiple ways to resolve this issue including, changing...

And consistently get the same error that there is no such file /home/vsts/work/r1/a/_ci-build/mlops-pipelines/code/scoring/deployment_config_aci.yml. This is a blocker for me.

KBaroni commented 5 years ago

Resolution requires adding a 2nd artifact (which is not documented). Once the repo is added as an artifact, path names can be selected from drop down.

eedorenko commented 5 years ago

_ci-build is the first artifact and It is documented in the 8. Deploy the Model section (and it is shown on the pipeline screenshot): "The pipeline consumes two artifacts: the result of the Build Pipeline as it contains configuration files and the model trained and registered by the ML training pipeline.

Configuration of a code _ci-build artifact is similar to what we did in the previous chapter."

And if the code artifact is aliased exactly as in the Getting Started (_ci-build), then the path parameters would be exactly as it is stated in the table $(System.DefaultWorkingDirectory)/_ci-build/mlops-pipelines/code/scoring/inference_config.yml and $(System.DefaultWorkingDirectory)/_ci-build/mlops-pipelines/code/scoring/deployment_config_aci.yml which can be selected from the drop-down list.

eedorenko commented 4 years ago

So, I guess it should be closed.