nerdd-backend is the central component of NERDD - a platform providing machine learning models for cheminformatics. The backend

• accepts HTTP / websocket requests,
• communicates with all components of the platform via a message broker, and
• stores persistent information on the file system or in a database.

# clone repository
git clone https://github.com/molinfo-vienna/nerdd-backend

# create conda environment
cd nerdd-backend
conda env create -f environment.yml
conda activate nerdd_backend

# install dependencies
pip install .

# run the server using development config
# (this works without any prerequisites; it uses a toy communication channel, 
# a fake database backend and a basic computational module for demonstration)
python -m nerdd_backend.main

# use localhost:8000 for api requests
curl localhost:8000/modules

# see all endpoints at API docs at http://localhost:8000/docs
xdg-open http://localhost:8000/docs

Custom options can be passed by providing a predefined configuration (see nerdd_backend/config) or overwriting individual options using ++. By default, the development config ( nerdd_backend/config/development.yaml) is loaded.

# switch to production config
python -m nerdd_backend.main --config-name production

# change port
python -m nerdd_backend.main ++port=7999

# turn off fake data
python -m nerdd_backend.main ++mock_infra=false

# set storage directory
python -m nerdd_backend.main ++media_root=./media

# run on existing kafka cluster as communication channel and rethinkdb database backend
# (see all options in config files in config folder)
python -m nerdd_backend.main --config-name production \
  ++db.host=localhost ++db.port=31562 \
  ++channel.broker=localhost:31624

• The subpackage nerdd_backend.routers contains all FastAPI routes accessible by the user.
• All routes have access to the FastAPI application state using request.app.state containing
  • state.config: custom settings (e.g. max_num_molecules_per_job) provided by the user when starting the server,
  • state.repository: the database access layer,
  • state.channel: an object for sending messages to the message broker, and
  • state.filesystem: an object for storing and retrieving files.
• Application settings are managed using Hydra. Predefined configurations are defined in /config, but individual options may be overridden when starting the server application.
• All schemas of requests, responses and records persisted to the database are declared as Pydantic models (deriving from pydantic.BaseModel) in nerdd_backend.models.
• The nerdd_backend.data.Repository class represents the database access layer specifying all database interactions. For a concrete database, a subclass of Repository is defined and all required methods are implemented. This abstraction allows the replacement of the underlying database technology without changes to the application logic.
• Communication with the NERDD infrastructure is handled via a message broker. The package nerdd-link defines message schemas and available topics. Specifically, the nerdd_link.Action class allows iterating over all messages in a topic and processing them using custom logic. All files in the nerdd_backend.actions subpackage contain subclasses of nerdd-link.Action that react to messages coming from the message broker.
• Actions are executed asynchronously in the global FastAPI lifespan, i.e. they run in parallel to FastAPI's handlers that serve HTTP requests.

# create conda environment
conda env create -f environment.yml

# install package in development mode
pip install -e .[test]

# run tests
pytest

# run tests with automatic reload on code changes
ptw