pydolphinscheduler

pydolphinscheduler is python API for Apache DolphinScheduler, which allow you definition your workflow by python code, aka workflow-as-codes.

Quick Start

Notice: For now, due to pydolphinscheduler without release to any binary tarball or PyPI, you have to clone Apache DolphinScheduler code from GitHub to ensure quick start setup

Here we show you how to install and run a simple example of pydolphinscheduler

Prepare

# Clone code from github
git clone git@github.com:apache/dolphinscheduler.git

# Install pydolphinscheduler from source
cd dolphinscheduler-python/pydolphinscheduler
pip install -e .

Start Server And Run Example

Before you run an example, you have to start backend server. You could follow development setup section "DolphinScheduler Standalone Quick Start" to set up developer environment. You have to start backend and frontend server in this step, which mean that you could view DolphinScheduler UI in your browser with URL http://localhost:12345/dolphinscheduler

After backend server is being start, all requests from pydolphinscheduler would be sent to backend server. And for now we could run a simple example by:

cd dolphinscheduler-python/pydolphinscheduler
python example/tutorial.py

NOTICE: Since Apache DolphinScheduler's tenant is requests while running command, you might need to change tenant value in example/tutorial.py. For now the value is tenant_exists, please change it to username exists in you environment.

After command execute, you could see a new project with single process definition named tutorial in the UI.

Until now, we finish quick start by an example of pydolphinscheduler and run it. If you want to inspect or join pydolphinscheduler develop, you could take a look at develop

Develop

pydolphinscheduler is python API for Apache DolphinScheduler, it just defines what workflow look like instead of store or execute it. We here use py4j to dynamically access Java Virtual Machine.

Setup Develop Environment

We already clone the code in quick start, so next step we have to open pydolphinscheduler project in you editor. We recommend you use pycharm instead of IntelliJ IDEA to open it. And you could just open directory dolphinscheduler-python/pydolphinscheduler instead of dolphinscheduler-python.

Then you should add developer dependence to make sure you could run test and check code style locally

pip install -r requirements_dev.txt

Brief Concept

Apache DolphinScheduler is design to define workflow by UI, and pydolphinscheduler try to define it by code. When define by code, user usually do not care user, tenant, or queue exists or not. All user care about is created a new workflow by the code his/her definition. So we have some side object in pydolphinscheduler/side directory, their only check object exists or not, and create them if not exists.

Process Definition

pydolphinscheduler workflow object name, process definition is also same name as Java object(maybe would be change to other word for more simple).

Tasks

pydolphinscheduler tasks object, we use tasks to define exact job we want DolphinScheduler do for us. For now, we only support shell task to execute shell task. This link list all tasks support in DolphinScheduler and would be implemented in the further.

Code Style

We use isort to automatically keep Python imports alphabetically, and use Black for code formatter and Flake8 for pep8 checker. If you use pycharmor IntelliJ IDEA, maybe you could follow Black-integration to configure them in your environment.

Our Python API CI would automatically run code style checker and unittest when you submit pull request in GitHub, you could also run static check locally.

# We recommend you run isort and Black before Flake8, because Black could auto fix some code style issue
# but Flake8 just hint when code style not match pep8

# Run Isort
isort .

# Run Black
black .

# Run Flake8
flake8

Testing

To test locally, you could directly run pytest after set PYTHONPATH

PYTHONPATH=src/ pytest

We try to keep pydolphinscheduler usable through unit test coverage. 90% test coverage is our target, but for now, we require test coverage up to 85%, and each pull request leas than 85% would fail our CI step Tests coverage. We use coverage to check our test coverage, and you could check it locally by run command.

coverage run && coverage report

It would not only run unit test but also show each file coverage which cover rate less than 100%, and TOTAL line show you total coverage of you code. If your CI failed with coverage you could go and find some reason by this command output.

7.2 KiB Raw Blame History