Issues with SQL Alchemy whilst deploying real time endpoint on ACI

Cameron Grimshaw-Jones 11

Hello,

I'm trying to deploy an image to an ACI using Azure Machine Learning. One of the requirements/installations is sqlalchemy.

When building the image, sqlalchemy seems to install correctly. However, when I try to import the sqlalchemy modules within the code, the deployment to the ACI fails and I can't work out why.

Anyone had any similar issues - let me know if I need to provide any more info.

Thanks,

Cam

romungi-MSFT 42,286 Reputation points Microsoft Employee

2022-01-14T06:51:10.207+00:00
@Cameron Grimshaw-Jones Have you tried setting the following parameter to true for your environment?

user_managed_dependencies=True

For example as mentioned in the document for custom environments.

dockerfile = """ FROM mcr.microsoft.com/azureml/openmpi3.1.2-ubuntu18.04:20210615.v1 RUN conda install numpy """ myenv.docker.base_image = None myenv.docker.base_dockerfile = dockerfile myenv.python.user_managed_dependencies=True myenv.python.interpreter_path = "/opt/miniconda/bin/python"

If you install some Python dependencies in your Docker image and forget to set user_managed_dependencies=True, those packages will not exist in the execution environment thus causing runtime failures. By default, Azure ML will build a Conda environment with dependencies you specified, and will execute the run in that environment instead of using any Python libraries that you installed on the base image.

1 answer

Cameron Grimshaw-Jones 11 Reputation points

2022-01-14T09:23:13.24+00:00

Hey,

Thanks for the reply. Below is currently what I'm doing - I'm using a yaml file rather than a docker file but I think what I'm currently doing should allow dependencies?

This is part of my yaml file that i'm supplying to the image

And then this is the code for the image i'm deploying.
Please sign in to rate this answer.
romungi-MSFT 42,286 Reputation points Microsoft Employee

2022-01-14T10:09:01.19+00:00

@Cameron Grimshaw-Jones I believe the containerImage class is now deprecated. Please check the reference on this page. You should switch to using the environment class for installing the required packages.

from azureml.core.webservice import AciWebservice, Webservice from azureml.core.model import Model env = Environment(name='myenv') python_packages = ['sqlalchemy','ntlk'] for package in python_packages: env.python.conda_dependencies.add_pip_package(package) inference_config = InferenceConfig(environment=env, source_directory='./source_dir', entry_script='./score.py') deployment_config = AciWebservice.deploy_configuration(cpu_cores = 1, memory_gb = 1) service = Model.deploy(ws, "aciservice", [model], inference_config, deployment_config) service.wait_for_deployment(show_output = True) print(service.state)

Cameron Grimshaw-Jones 11 Reputation points

2022-01-19T11:45:32.803+00:00

Thanks for pointing this out - I'm now using environments instead. I'm now using a docker file & conda_dependencies files to build the environment/image but I'm still getting an issue with pyodbc whatever method I end up trying.

Defining my environment:

Deploying to ACI:

Conda yml file
166367-conda-dependencies.txt

DockerFile
166403-docker-file.txt

Error message:

romungi-MSFT 42,286 Reputation points Microsoft Employee

2022-01-24T07:39:56.067+00:00

It seems like an issue with the install of one of the dependencies of pyodbc while creating your environment from the docker file config. I see you have added unixodbc-dev as mentioned in the pre-reqs though. Have you tried to use the basic image to setup an environment first and then adding your dependencies? Ubuntu 20.04 basic docker install file is here with py3.7
Sign in to comment