Using Amazon EFS with AWS Lambda to load a ML model

In this project, we will attach an Amazon EFS to an AWS Lambda. In the EFS, there is an XGBoost model (model.tar.gz), which will be loaded at runtime by the AWS Lambda. Since this does not require a download from Amazon S3, it has lower latency with respect to S3.

To deploy

1. Run sh build_and_push <name-of-your-liking> in the container folder
   1. in my case, the image name is xgboost-lambda. If you change it, change line 30 in stack.py
2. You can use an existing EC2 / Cloud9 / SageMaker Notebook to mount the EFS partition and upload the model. Instead, if you want the Lambda to load the model to the EFS:
   1. upload the model to S3 with aws cp model.tar.gz s3://{bucket}/{prefix}/
   2. uncomment line 9 in the predict.py file adding your bucket and prefix pointing towards the .tar.gz file containing the model
   3. change the bucket line 22 in stack.py with your bucket
3. Run pip install -r requirements.txt then cdk deploy in the stack folder
   1. if you don't have CDK installed, run npm install -g aws-cdk@latest
4. Run test.py from the test folder
5. Remember to change back the predict.py file and then redeploy your stack