Modular, transparent, batteries (half) included, lightning fast web server. Features a functional, isolated business layer with an imperative decoupled shell.
This is intendend as both to serve as a scaffold for our internal projects as to improve and give back to our community as an efficient bullet-proof backend design, leveraging Python's expressability.
- Lightning fast ASGI server via
uvicorn. - GraphQL support via
ariadne. - Full GraphQL compliant query pagination support.
- JWT authentication.
- Resource-based authorization integrated using GraphQL directives.
- Efficient dependency management via
poetry - Database migration systems using
alembic. - Event-driven architecture:
- Internal message bus that injects adapters dependencies into service-handlers functions.
- External message bus for background workers integrated w/ AWS Lambda.
- Sober test pyramid: units, integrations and e2e tests.
- Decoupled service layer that responds only to commands and events.
- Aggregate's atomic services consistency guaranteed using
postgresisolation levels locks. - Isolated and pure domain layer that has no dependencies (no, not even with ORM).
This is project's in its early stages, and should receive a big WIP tag. We should track progress using GitHub features:
- Discussions for brainstorming & prioritizing
- Milestones for planned features
- Issues for ongoing tasks
There are two ways to run this: i. Locally, through a virtual pyenv; ii. Using docker-compose.
We like Makefile interface and while we also don't deliver an appropriate and fancy CLI, commands will be handled there.
To find out which commands are available, cat Makefile.
This projects uses poetry to manage dependencies. Even though poetry is way too slow to run poetry install, we find
that for managing dependencies' version compatibility is a valuable tool.
While we don't have a fully automated build pipeline, we agree to poetry export -f requirements.txt > requirements.txt.
Make sure docker daemon is running.
make buildmake service
make testmake cleanHaving said that, how you instantiate your virtual environment is up to you. You can do that now.
Inside your blank python virtual environment:
pip install -r requirements.txt As there aren't any containerization being done for now, you'd need postgres up and running in your local machine.
psql -c "create database template"Right now you should be able to run the entire test-suite properly.
make test-local Why not use django? Or flask? Or FastAPI? Even though these are great frameworks they're (mostly heavily) opiniated. At T10, we have a need to implement and deliver maintainable software that we really know what's happening under the (at least conceptual Pythonic-layer) hood. As a software house, we've also come to find that by using such frameworks programmers are more likely to be inhibited from practicing and improving their software design skills.
We're (obviously) not alone here. pca have touched base a few years ago.
We are committed to these non-negotiables principle:
- Modularity, high cohesion and low coupling
- Transparency, ensuring readable, debugabble, maintainable software
- Orthogonality, which makes us sane by dilligently avoiding code that emmits side-effects
- Testability, we need code that can be (as easily as possibly) tested
We don't claim to have created everything from scratch. Quite the opposite, the work here is a direct fork from ideas we really identify with that were hard earned throughout the past two decades.
Specifically:
- Architecture Patterns with Python from Bob Gregory & Harry Percival,
- Python Clean Architecture, from
pcah - Functional Core, Imperative Shell from Destroy All Software,
- Hexagonal Architecture aka Ports & Adapters by Alistair Cockburn
- Domain-Driven-Design by Eric Evans & Martin Fowler