1

I am trying to create a web-service via Azure Model Management and am struggling.

I've followed the instructions and have managed to operationalize locally in a Docker container. My 'score.py' file includes a query to a SQL database using pyodbc. This functions perfectly when I test this on my local environment using the ML Workbench, however once this has been deployed in a Docker container I come across this error:

'Response Content': b'(\'01000\', "[01000] [unixODBC][Driver Manager]Can\'t open lib \'ODBC Driver 13 for SQL Server\' : file not found (0) (SQLDriverConnect)")'

I have included pyodbc in my conda_dependencies.yml.

Has anyone got any suggestions? Is there any further dependencies that I need to include?

Azure seem to have recently added the ability to customize container images using what they call a 'Docker Steps file'. I have practically no experience in Docker, but after reading this question i tried including a 'Docker Steps file' containing this:

ADD odbcinst.ini /etc/odbcinst.ini
RUN apt-get update
RUN apt-get install -y tdsodbc unixodbc-dev
RUN apt install unixodbc-bin -y
RUN apt-get clean -y

However i understand 'ADD' commands are not possible in this type of file, so this seems to have made no difference.

Hopefully this this all makes sense! Any advice would be very much appreciated! I hope I'm not the only one stumbling my way through Azure ML!

EDIT:

I'm still stuck, but making progress...

I accessed the root of the container using:

docker exec -ti -u root container_name bash

From here I ran 'odbcinst -j`, resulting in:

unixODBC 2.3.6
DRIVERS............: /etc/odbcinst.ini
SYSTEM DATA SOURCES: /etc/odbc.ini
FILE DATA SOURCES..: /etc/ODBCDataSources
USER DATA SOURCES..: /root/.odbc.ini
SQLULEN Size.......: 8
SQLLEN Size........: 8
SQLSETPOSIROW Size.: 8

I couldn't seem to actually locate odbc.ini so i followed these instructions for installing 'ODBC Driver 13' for Ubuntu 16.04. Now when I run the service i get a different error:

{'Error': MlCliError({'Error': 'Error occurred while attempting to score service myapp.', 'Response Code': 502, 'Response Content': b'<html>\r\n<head><title>502 Bad Gateway</title></head>\r\n<body bgcolor="white">\r\n<center><h1>502 Bad Gateway</h1></center>\r\n<hr><center>nginx/1.10.3 (Ubuntu)</center>\r\n</body>\r\n</html>\r\n', 'Response Headers': {'Content-Length': '182', 'Content-Type': 'text/html', 'Date': 'Wed, 18 Apr 2018 14:06:30 GMT', 'Server': 'nginx/1.10.3 (Ubuntu)', 'Connection': 'keep-alive'}},), 'Azure-cli-ml Version': '0.1.0b2'}

I have also tried altering my score.py file to return: pyodbc.drivers() this results in a blank '[]'

Andreas
  • 25
  • 4
  • Where did you add `pyodbc` in the dependencies file? I think there's a difference between adding it directly under the `dependencies` module and adding it under the `pip` module under there. – Jon Apr 13 '18 at 14:36
  • Hi @Jon, I included `pyodbc` directly under `dependencies` – Andreas Apr 13 '18 at 15:05
  • Try adding it under `pip` similar to how `matplotlib` is here - https://github.com/Azure/MachineLearningSamples-Iris/blob/master/aml_config/conda_dependencies.yml – Jon Apr 13 '18 at 15:05
  • Thanks for the suggestion! I tried, however when creating the image, an error occurred. It's a bit tricky to read through the log information but `gcc: error trying to exec 'cc1plus': execvp: No such file or directory error: command 'gcc' failed with exit status 1` – Andreas Apr 13 '18 at 15:34
  • Interesting. What image are you using for it? This may be in the `docker.compute` file – Jon Apr 13 '18 at 15:43
  • The `docker.compute` file includes the line `baseDockerImage: "microsoft/mmlspark:plus-0.9.9"` , but is this involved if I've built the image via Model Management? – Andreas Apr 13 '18 at 15:59
  • Ahhhh, you're right. You do [create the image](https://learn.microsoft.com/en-us/azure/machine-learning/preview/model-management-service-deploy#6-create-an-image). Sorry about that. The error sounds like something should be installed on the image, but I'm not sure that's the case here. – Jon Apr 14 '18 at 16:40
  • Sorry I should have made clear, with `pyodbc` installed directly under the `dependencies:` the image is built successfully, the issue only occurs when attempting to use the run command. Thanks for your suggestions anyway! I'm going to attempt to place `print(pyodbc.drivers())` in my `score.py` file to see if there are any other SQL drivers available – Andreas Apr 16 '18 at 08:40
  • Might seem obvious but `print(pyodbc.drivers())` returned an empty list. Seems like i definitely need to include the SQL drivers on the image somehow – Andreas Apr 16 '18 at 15:51
  • That is interesting. Looks like it is there, though. Unfortunately, I haven't worked with `pyodbc` so I'm not sure how to add the drivers :( – Jon Apr 16 '18 at 15:58

0 Answers0