Scargo is Argo for Scientific Workflows.
A Scientific Workflow is a single-use workflow which takes statically defined inputs and outputs results. This is simpler than the CI/CD and event-driven workflows which Argo implements.
This reduced scope allows Scargo to be implemented as a subset of Python. This Python subset can be run/tested locally, before transpiling to Argo Workflow's YAML to run on a cluster.
Using Python instead of YAML also provides additional benefits, such as as:
- Access to Python tooling, such as linting, auto-formatting and auto-completion
- Less nesting and more brevity than YAML
- Easier testing
- Improved modularity
Using Scargo, scientists should be able to prototype and modify simple sequential Argo workflows without engineering intervention.
Almost all of these features have not been implemented yet. See examples/ for practical demonstrations of features.
- Run Python functions locally, allowing for unit testing before running with Argo
- Run steps defined as Bash scripts
- Define global workflow parameters such they can separated out in the compiled Argo workflow
- Import common Argo functionality (iterating through a CSV, iterating through all files in an S3 folder)
- Manage artifacts and parameters for individual steps with minimal boilerplate
- Provide a YAML escape hatch
- Allows for quickly implementing weird corner cases, such as adding volume mounts (MOE licenses), retry strategies and environment variables (MongoDB)
- Adding Python-only dependencies to a Docker Image. All other dependencies that are not pip-installable will require research-engineer intervention.
- Running a workflow locally in the same environment as it would run in the cloud.
- A user wants to specify resource requirements for each workflow step/Python function
- Automate switching between local root and S3 root
Complete implementation of the Argo API, including:
- Recursion
- Artifacts from anywhere other than S3 and Minio
We rely on invoke to install our project code with Poetry.
You can install the scargo project in development mode by running:
inv installRunning the full test suite is as easy as running:
inv testIf you're interested in the code coverage of this repo you may run:
inv test --coverage --html-reportand inspect the resulting interactive HTML report by opening the index.html in the htmlcov/ folder in your favourite browser.