Skip to content

SageMaker-Installation-Instructions.md

Latest commit

 

History

History
50 lines (30 loc) · 2.17 KB

SageMaker-Installation-Instructions.md

File metadata and controls

50 lines (30 loc) · 2.17 KB

Ensuring permissions for your SageMaker execution role

Make sure to add the correct permissions to the role your SageMaker Studio runs with. A good place to start is by:

  • Adding the AwsGlueDataBrewFullAccessPolicy managed policy to your SageMaker execution role, and;

  • Creating AwsGlueDataBrewSpecificS3BucketPolicy customer managed policy with s3:GetObject and s3:PutObject permissions for the bucket/s in which you would like to operate and adding this policy to your SageMaker execution role. Refer to IAM policy for S3 documentation for more details.

This grants baseline permissions to use AWS Glue DataBrew console features. Refer to DataBrew developer guide on Setting up IAM permissions for a comprehensive guide on permissions setup.

Installing the plugin

  1. Install/upgrade dependencies

Once SageMaker Studio has started, open system terminal under Launcher tab. Amazon SageMaker Studio, as of writing, comes installed with aws_jupyter_proxy v0.1.0. Upgrade aws_jupyter_proxy to v0.3.1 or later and ensure you have a compatible version.

After installation, enable the aws_jupyter_proxy as jupyter server extension.

# upgrade to latest
pip install --upgrade aws-jupyter-proxy

# verify version
pip show aws_jupyter_proxy

# enable server extension
jupyter serverextension enable --py aws_jupyter_proxy
  1. Install the extension and build Jupyter assets

Download and install the latest version of DataBrew JupyterLab extension, and build Jupyter assets.

jupyter labextension install aws_glue_databrew_jupyter --minimize=False
  1. Restart the Jupyter server

When installing server extensions, like aws_jupyter_proxy, Jupyter server needs to be restarted to run the newly installed server extension.

nohup supervisorctl -c /etc/supervisor/conf.d/supervisord.conf restart jupyterlabserver
  1. Restart browser

After running the command above, terminal will automatically close and move you to the Launcher tab. Hard refresh the page to relaunch JupyterLab assets.

  1. Launch DataBrew extension