Packaging app

In this step, the MONAI Deploy application can now be built into a deployable Docker image using the MONAI Application Packager.

Overview on MONAI Application Packager

The MONAI Application Packager (Packager) is a utility for building an application developed with the MONAI Deploy App SDK into a structured MONAI Application Package (MAP).

The MAP produced by the Packager is a deployable and reusable docker image that can be launched locally or remotely.

Basic Usage of MONAI Application Packager

monai-deploy package <APP_PATH> --tag <TAG> [--model <MODEL_PATH>] [--log-level <LEVEL>] [-h]

Required Arguments

  • <APP_PATH>: A path to MONAI Deploy Application folder or main code.

  • --tag, -t <TAG>: A MAP name and optionally a tag in the ‘name:tag’ format.

Note

If <APP_PATH> refers to a python code (such as ./my_app.py), only the file would be packaged into the Docker image.

Please specify the application root path which includes __main__.py file if you want to package the whole application folder.

The file usually would look like below:

# Assuming that the class `App` (that inherits Application) is defined in `app.py`.
from app import App

if __name__ == "__main__":
    App(do_run=True)

Optional Arguments

The following list contains arguments that have default values or are provided through the SDK if these flags are not invoked. However, the user can choose to override these values by invoking these optional flags:

  • --model, -m <MODEL_PATH>: Path to a model file or a directory containing all models as subdirectories

  • --requirements, -r <FILE_PATH>: Optional path to requirements.txt containing package dependencies of the application

  • --log-level, -l <LEVEL>: Set the logging level ("DEBUG", "INFO", "WARN", "ERROR", or "CRITICAL").

  • --base <BASE_IMAGE>: Base Docker Image (overrides default "nvcr.io/nvidia/pytorch:22.08-py3").

  • --input-dir, -i <INPUT_DIR>: Directory mounted in container for Application Input (overrides default "input").

  • --models-dir <MODELS_DIR>: Directory mounted in container for Models Path (overrides default "/opt/monai/models").

  • --output-dir, -o <OUTPUT_DIR>: Directory mounted in container for Application Output (overrides default "output").

  • --working-dir, -w <WORKDIR>: Directory mounted in container for Application (overrides default "/var/monai").

  • --no-cache, -n: Do not use cache when building image.

  • --timeout <VALUE>: Timeout (overrides default 0. In seconds.).

  • --version: Set the version of the Application (overrides default "0.0.0").

  • -h, --help: Show the help message and exit.

Example of using MONAI Application Packager

Given an example MONAI Deploy App SDK application with its code residing in a directory ./my_app, packaging this application with the Packager to create a Docker image tagged my_app:latest would look like this:

$ monai-deploy package ./my_app -t my_app:latest --model ./model.pt

Building MONAI Application Package...
Successfully built my_app:latest

The MAP image my_app:latest will be seen in the list of container images on the user’s local machine when the command docker images is run. The MAP image my_app:latest will be able to run locally or remotely.

Note

  • The current implementation (as of 0.1.0) of the Packager ONLY supports CUDA and Pytorch images from nvcr.io as base images for the MAP. There are efforts in progress to add support for smaller images from dockerhub.

  • This release was developed and tested with CUDA 11 and Python 3.8. There are efforts in progress to add support for different versions of the CUDA Toolkit.

Next Step

See the next page to learn more on how to locally run a MONAI application package image built by the Packager.

Executing app locally Executing packaged app locally