Azure Machine Learning
An Azure machine learning service for building and deploying models.
3,335 questions
Azure ML Environment build fails
![](data:image/svg+xml, %3Csvg xmlns='http://www.w3.org/2000/svg' height='64' class='font-weight-bold' style='font: 600 30.11764705882353px "SegoeUI", Arial' width='64'%3E%3Ccircle fill='hsl(115.19999999999999, 85%, 21%)' cx='32' cy='32' r='32' /%3E%3Ctext x='50%25' y='55%25' dominant-baseline='middle' text-anchor='middle' fill='%23FFF' %3EMJ%3C/text%3E%3C/svg%3E)
Marvin Joseph Guico Agor
67
Reputation points
I am trying to construct a custom environment to be used in a scheduled pipeline. I used the following script to create the custom environment:
from azure.ai.ml.entities import Environment
env_docker_conda = Environment(
image = "AzureML-sklearn-0.24-ubuntu18.04-py37-cpu", # Building on top of this pre-constructed environment
conda_file = './src/conda-env.yml',
name = "custom-env-trending-terms-01",
description = "Custom Environment for Trending Terms"
)
ml_client.environments.create_or_update(env_docker_conda) # registering custom environment
Where the conda-env.yml file is defined as follows
name: required-env-cpu
channels:
- conda-forge
dependencies:
- python=3.80
- nlp-rake
- nltk
- langdetect
- pandas
- numpy
I ran the code chunk (on kernel Python 3.10 V2). Checking the Environment's build status, I found that it failed.
Checking the build log, I found that it failed due to "Connection to container registry docker.io failed."
Would like to ask for any and all help: how could I resolve this issue?
Azure Machine Learning
{count} votes
-
![](data:image/svg+xml, %3Csvg xmlns='http://www.w3.org/2000/svg' height='64' class='font-weight-bold' style='font: 600 30.11764705882353px "SegoeUI", Arial' width='64'%3E%3Ccircle fill='hsl(278.4, 1%, 36%)' cx='32' cy='32' r='32' /%3E%3Ctext x='50%25' y='55%25' dominant-baseline='middle' text-anchor='middle' fill='%23FFF' %3EYM%3C/text%3E%3C/svg%3E)
YutongTie-MSFT 53,966 Reputation points Moderator2023-05-31T02:54:07.54+00:00 Hello @Marvin Joseph Guico Agor
Thanks for reaching out to us for this issue.
It seems like the issue is related to the connection to the Docker registry. The error message suggests that the connection to the Docker registry
docker.iofailed. This could be due to a network issue or authentication issue.Here are a few things you can try to resolve the issue:
Check your network connection: Make sure that your network connection is stable and that you are able to connect to the internet. You can try pinging
docker.ioto see if you are able to connect to it.Check your Docker credentials: Make sure that you have the correct Docker credentials to access the Docker registry. You can try logging in to Docker using the
docker logincommand and then try building the environment again.Check your Azure Machine Learning workspace configuration**: Make sure that your Azure Machine Learning workspace is configured correctly to use Docker**. You can check the workspace configuration by running the following code:
from azureml.core import Workspace ws = Workspace.from_config() print(ws.get_details()['containerRegistry'])This should return the details of the container registry that is configured for your workspace.
If none of these solutions work, please let me know. Thanks
Regards,
Yutong
-
Marvin Joseph Guico Agor 67 Reputation points
2023-05-31T04:22:09.1366667+00:00 Unfortunately, I'm unable to see the sources you are linking as I don't have access to CloudGPT
Further, I'm using a modified version of your script as I'm using Python SDK V2. Below is the script I used
# Getting a handle on the workspace ml_client = MLClient.from_config(DefaultAzureCredential()) for ret in ml_client.registries.list(): print(ret)However, this yields an empty response.
-
YutongTie-MSFT 53,966 Reputation points Moderator
2023-05-31T07:28:45.5666667+00:00 Sorry, let me send another link for you, will update it later.
-
YutongTie-MSFT 53,966 Reputation points Moderator
2023-06-02T19:18:44.0866667+00:00 Hello @Marvin Joseph Guico Agor Are you still suffering from this issue? If yes, I am happy to enable you a free support ticket for this issue, please let me know.
-
Marvin Joseph Guico Agor 67 Reputation points
2023-06-05T04:15:52.1333333+00:00 Hello @YutongTie-MSFT . Yes, I still am. However, I've been experimenting, trying other options. I'm still attempting to create an Environment, but I'm now encountering a different error.
-
Marvin Joseph Guico Agor 67 Reputation points
2023-06-07T05:05:07.52+00:00 Hi @YutongTie-MSFT , if this offer is still open, I'd like to ask for help about this error I'm encountering:
Overall, I'm planning to use AzureML's Pipeline (with the Python SDK V2) to create a scheduled pipeline that executes a preprocessing script. Currently, I'm at the step where I'm trying to define the Environment to be used by the pipeline. I've defined the YML file for this environment's dependencies as follows:
Unfortunately, upon checking the Environment build status in the studio, the build fails because "conda can't find the nlp-rake package from the channels".
I followed the instructions in the trouble shooting page and have confirmed that the package spelling and the channel (where nlp-rake can be found) is correct. Below is a screenshot of Pypi's site showing that nlp-rake can indeed be found in that channel.
Further, I've managed to install this on my Compute Instance. It confirms that the channel this package is from is Pypi. The screenshot below shows the results when I run conda list in the terminal in Azure ML.
As such, I'm unsure as to why the build is failing.
As an alternative route, I attempted to specify these as pip modules (they can all be found in pypi):
It would still fail, and this time at the parsing level. I'm not sure why, but I'm getting the following ValidationException when the yaml file is being parsed:
Sign in to comment
Sign in to answer