Contributions are welcome and are greatly appreciated! Every little bit helps, and credit will always be given.
Table of Contents generated with DocToc
Report bugs through Apache Jira
Please report relevant information and preferably code that exhibits the problem.
Look through the Jira issues for bugs. Anything is open to whoever wants to implement it.
Look through the Apache Jira for features. Any unassigned “Improvement” issue is open to whoever wants to implement it.
We've created the operators, hooks, macros and executors we needed, but we made sure that this part of Airflow is extensible. New operators, hooks, macros and executors are very welcomed!
Airflow could always use better documentation, whether as part of the official Airflow docs, in docstrings, docs/*.rst
or even on the web as blog posts or articles.
The best way to send feedback is to open an issue on Apache Jira
If you are proposing a feature:
The latest API documentation is usually available here. To generate a local version, you need to have set up an Airflow development environment (see below). Also install the doc
extra.
pip install -e '.[doc]'
Generate and serve the documentation by running:
cd docs ./build.sh ./start_doc_server.sh
When you develop Airflow you can create local virtualenv with all requirements required by Airflow.
Advantage of local installation is that everything works locally, you do not have to enter Docker/container environment and you can easily debug the code locally. You can also have access to python virtualenv that contains all the necessary requirements and use it in your local IDE - this aids autocompletion, and running tests directly from within the IDE.
It is STRONGLY encouraged to also install and use Pre commit hooks for your local development environment. They will speed up your development cycle speed a lot.
The disadvantage is that you have to maintain your dependencies and local environment consistent with other development environments that you have on your local machine.
Another disadvantage is that you you cannot run tests that require external components - mysql, postgres database, hadoop, mongo, cassandra, redis etc.. The tests in Airflow are a mixture of unit and integration tests and some of them require those components to be setup. Only real unit tests can be run by default in local environment.
If you want to run integration tests, you need to configure and install the dependencies on your own.
It‘s also very difficult to make sure that your local environment is consistent with other environments. This can often lead to “works for me” syndrome. It’s better to use the Docker Compose integration test environment in case you want reproducible environment consistent with other people.
Install Python (3.5 or 3.6), MySQL, and libxml by using system-level package managers like yum, apt-get for Linux, or Homebrew for Mac OS at first. Refer to the Dockerfile for a comprehensive list of required packages.
In order to use your IDE you need you can use the virtual environment. Ideally you should setup virtualenv for all python versions that Airflow supports (3.5, 3.6). An easy way to create the virtualenv is to use virtualenvwrapper - it allows you to easily switch between virtualenvs using workon
command and mange your virtual environments more easily. Typically creating the environment can be done by:
mkvirtualenv <ENV_NAME> --python=python<VERSION>
Then you need to install python PIP requirements. Typically it can be done with: pip install -e ".[devel]"
.
Note - if you have trouble installing mysql client on MacOS and you have an error similar to
ld: library not found for -lssl
you should set LIBRARY_PATH before running pip install
:
export LIBRARY_PATH=$LIBRARY_PATH:/usr/local/opt/openssl/lib/
After creating the virtualenv, run this command to create the Airflow sqlite database:
airflow db init
This can be automated if you do it within a virtualenv. The ./breeze script has a flag (-e or --initialize-local-virtualenv) that automatically installs dependencies in the virtualenv you are logged in and resets the sqlite database as described below.
After the virtualenv is created, you must initialize it. Simply enter the environment (using workon
) and once you are in it run:
./breeze --initialize-local-virtualenv
Once initialization is done, you should select the virtualenv you initialized as the project's default virtualenv in your IDE and run tests efficiently.
After setting it up - you can use the usual “Run Test” option of the IDE and have the autocomplete and documentation support from IDE as well as you can debug and view the sources of Airflow - which is very helpful during development.
Once you activate virtualenv (or enter docker container) as described below you should be able to run run-tests
at will (it is in the path in Docker environment but you need to prepend it with ./
in local virtualenv (./run-tests
).
Note that this script has several flags that can be useful for your testing.
Usage: run-tests [FLAGS] [TESTS_TO_RUN] -- <EXTRA_NOSETEST_ARGS> Runs tests specified (or all tests if no tests are specified) Flags: -h, --help Shows this help message. -i, --with-db-init Forces database initialization before tests -s, --nocapture Don't capture stdout when running the tests. This is useful if you are debugging with ipdb and want to drop into console with it by adding this line to source code: import ipdb; ipdb.set_trace() -v, --verbose Verbose output showing coloured output of tests being run and summary of the tests - in a manner similar to the tests run in the CI environment.
You can pass extra parameters to nose, by adding nose arguments after --
For example, in order to just execute the “core” unit tests and add ipdb set_trace method, you can run the following command:
./run-tests tests.core:TestCore --nocapture --verbose
or a single test method without colors or debug logs:
./run-tests tests.core:TestCore.test_check_operators
Note that ./run_tests
script runs tests but the first time it runs, it performs database initialisation. If you run further tests without leaving the environment, the database will not be initialized, but you can always force database initialization with --with-db-init
(-i
) switch. The scripts will inform you what you can do when they are run.
Once you configure your tests to use the virtualenv you created. running tests from IDE is as simple as:
Note that while most of the tests are typical “unit” tests that do not require external components, there are a number of tests that are more of “integration” or even “system” tests (depending on the convention you use). Those tests interact with external components. For those tests you need to run complete Docker Compose - base environment below.
This is the environment that is used during CI builds on Travis CI. We have scripts to reproduce the Travis environment and you can enter the environment and run it locally.
The scripts used by Travis CI run also image builds which make the images contain all the sources. You can see which scripts are used in .travis.yml file.
Docker
You need to have Docker CE installed.
IMPORTANT!!! : Mac OS Docker default Disk size settings
When you develop on Mac OS you usually have not enough disk space for Docker if you start using it seriously. You should increase disk space available before starting to work with the environment. Usually you have weird problems with docker containers when you run out of Disk space. It might not be obvious that space is an issue. If you get into weird behaviour try Cleaning Up Docker
See Docker for Mac - Space for details of increasing disk space available for Docker on Mac.
At least 128 GB of Disk space is recommended. You can also get by with smaller space but you should more often clean the docker disk space periodically.
Getopt and coreutils
If you are on MacOS:
brew install gnu-getopt coreutils
(if you use brew, or use equivalent command for ports)If you use bash, you should run this command:
echo 'export PATH="/usr/local/opt/gnu-getopt/bin:$PATH"' >> ~/.bash_profile . ~/.bash_profile
If you use zsh, you should run this command:
echo 'export PATH="/usr/local/opt/gnu-getopt/bin:$PATH"' >> ~/.zprofile . ~/.zprofile
If you are on Linux:
apt install util-linux coreutils
or equivalent if your system is not Debian-based.Airflow has a super-easy-to-use integration test environment managed via Docker Compose and used by Airflow's CI Travis tests.
It‘s called Airflow Breeze as in "_It’s a breeze to develop Airflow_"
All details about using and running Airflow Breeze can be found in BREEZE.rst
The advantage of the Airflow Breeze Integration Tests environment is that it is a full environment including external components - mysql database, hadoop, mongo, cassandra, redis etc. Some of the tests in Airflow require those external components. Integration test environment provides preconfigured environment where all those services are running and can be used by tests automatically.
Another advantage is that the Airflow Breeze environment is pretty much the same as used in Travis CI automated builds, and if the tests run in your local environment they will most likely work on Travis as well.
The disadvantage of Airflow Breeze is that it is fairly complex and requires time to setup. However it is all automated and easy to setup. Another disadvantage is that it takes a lot of space in your local Docker cache. There is a separate environment for different python versions and airflow versions and each of the images take around 3GB in total. Building and preparing the environment by default uses pre-built images from DockerHub (requires time to download and extract those GB of images) and less than 10 minutes per python version to build.
Note that those images are not supposed to be used in production environments. They are optimised for repeatability of tests, maintainability and speed of building rather than performance
Once you are inside the environment you can run individual tests as described in Running individual tests.
We have a number of static code checks that are run in Travis CI but you can run them locally as well. All the scripts are available in scripts/ci folder.
All these tests run in python3.6 environment. Note that the first time you run the checks it might take some time to rebuild the docker images required to run the tests, but all subsequent runs will be much faster - the build phase will just check if your code has changed and rebuild as needed.
The checks below are run in a docker environment, which means that if you run them locally, they should give the same results as the tests run in TravisCI without special environment preparation.
You can trigger the static checks from the host environment, without entering Docker container. You do that by running appropriate scripts (The same is done in TravisCI)
The scripts will fail by default when image rebuild is needed (for example when dependencies change) and provide instruction on how to rebuild the images. You can control the default behaviour as explained in Default behaviour for user interaction
You can force rebuilding of the images by deleting .build directory. This directory keeps cached information about the images already built and you can safely delete it if you want to start from the scratch.
After Documentation is built, the html results are available in docs/_build/html folder. This folder is mounted from the host so you can access those files in your host as well.
If you are already in the Docker Compose Environment you can also run the same static checks from within container:
./scripts/ci/in_container/run_mypy.sh airflow tests
./scripts/ci/in_container/run_flake8.sh
./scripts/ci/in_container/run_check_licence.sh
./scripts/ci/in_container/run_docs_build.sh
In all static check scripts - both in container and in the host you can also pass module/file path as parameters of the scripts to only check selected modules or files. For example:
In container:
./scripts/ci/in_container/run_mypy.sh ./airflow/example_dags/
or
./scripts/ci/in_container/run_mypy.sh ./airflow/example_dags/test_utils.py
In host:
./scripts/ci/ci_mypy.sh ./airflow/example_dags/
or
./scripts/ci/ci_mypy.sh ./airflow/example_dags/test_utils.py
And similarly for other scripts.
For all development tasks related integration tests and static code checks we are using Docker images that are maintained in Dockerhub under apache/airflow
repository.
There are three images that we currently manage:
<BRANCH>-python<PYTHON_VERSION>-ci-slim
(for example master-python3.6-ci-slim
). The image is built using the Dockerfile dockerfile.<BRANCH>-python<PYTHON_VERSION>-ci
(for example master-python3.6-ci
). The image is built using the Dockerfile dockerfile.checklicence
image.We also use a very small Dockerfile-context dockerfile in order to fix file permissions for an obscure permission problem with Docker caching but it is not stored in apache/airflow
registry.
Before you run tests or enter environment or run local static checks, the necessary local images should be pulled and built from DockerHub. This happens automatically for the test environment but you need to manually trigger it for static checks as described in Building the images and Force pulling and building the images). The static checks will fail and inform what to do if the image is not yet built.
Note that building image first time pulls the pre-built version of images from Dockerhub might take a bit of time - but this wait-time will not repeat for any subsequent source code change. However, changes to sensitive files like setup.py or Dockerfile will trigger a rebuild that might take more time (but it is highly optimised to only rebuild what's needed)
In most cases re-building an image requires connectivity to network (for example to download new dependencies). In case you work offline and do not want to rebuild the images when needed - you might set ASSUME_NO_TO_ALL_QUESTIONS
variable to true
as described in the Default behaviour for user interaction chapter.
See Troubleshooting section for steps you can make to clean the environment.
Sometimes during the build user is asked whether to perform an action, skip it, or quit. This happens in case of image rebuilding and image removal - they can take a lot of time and they are potentially destructive. For automation scripts, you can export one of the three variables to control the default behaviour.
export ASSUME_YES_TO_ALL_QUESTIONS="true"
If ASSUME_YES_TO_ALL_QUESTIONS
is set to true
, the images will automatically rebuild when needed. Images are deleted without asking.
export ASSUME_NO_TO_ALL_QUESTIONS="true"
If ASSUME_NO_TO_ALL_QUESTIONS
is set to true
, the old images are used even if re-building is needed. This is useful when you work offline. Deleting images is aborted.
export ASSUME_QUIT_TO_ALL_QUESTIONS="true"
If ASSUME_QUIT_TO_ALL_QUESTIONS
is set to true
, the whole script is aborted. Deleting images is aborted.
If more than one variable is set, YES takes precedence over NO which take precedence over QUIT.
For your convenience, there are scripts that can be used in local development
Running all tests with default settings (python 3.6, sqlite backend, docker environment):
./scripts/ci/local_ci_run_airflow_testing.sh
Selecting python version, backend, docker environment:
PYTHON_VERSION=3.5 BACKEND=postgres ENV=docker ./scripts/ci/local_ci_run_airflow_testing.sh
Running kubernetes tests:
KUBERNETES_VERSION==v1.13.0 KUBERNETES_MODE=persistent_mode BACKEND=postgres ENV=kubernetes \ ./scripts/ci/local_ci_run_airflow_testing.sh
The following environments are possible:
docker
environment (default): starts all dependencies required by full integration test-suite (postgres, mysql, celery, etc.). This option is resource intensive so do not forget to Stop environment when you are finished. This option is also RAM intensive and can slow down your machine.kubernetes
environment: Runs airflow tests within a kubernetes cluster (requires KUBERNETES_VERSION and KUBERNETES_MODE variables).bare
environment: runs airflow in docker without any external dependencies. It will only work for non-dependent tests. You can only run it with sqlite backend. You can only enter the bare environment with local_ci_enter_environment.sh
and run tests manually, you cannot execute local_ci_run_airflow_testing.sh
with it.Note: The Kubernetes environment will require setting up minikube/kubernetes so it might require some host-network configuration.
Docker-compose environment starts a number of docker containers and keep them running. You can tear them down by running /scripts/ci/local_ci_stop_environment.sh
On Linux there is a problem with propagating ownership of created files (known Docker problem). Basically files and directories created in container are not owned by the host user (but by the root user in our case). This might prevent you from switching branches for example if files owned by root user are created within your sources. In case you are on Linux host and haa some files in your sources created by the root user, you can fix the ownership of those files by running scripts/ci/local_ci_fix_ownership.sh script.
You can manually trigger building of the local images using scripts/ci/local_ci_build.sh.
The scripts that build the images are optimised to minimise the time needed to rebuild the image when the source code of Airflow evolves. This means that if you already had the image locally downloaded and built, the scripts will determine, the rebuild is needed in the first place. Then it will make sure that minimal number of steps are executed to rebuild the parts of image (for example PIP dependencies) that will give you an image consistent with the one used during Continuous Integration.
You can also force-pull the images before building them locally so that you are sure that you download latest images from DockerHub repository before building. This can be done with scripts/ci/local_ci_pull_and_build.sh script.
Note that you might need to cleanup your Docker environment occasionally. The images are quite big (1.5GB for both images needed for static code analysis and CI tests). And if you often rebuild/update images you might end up with some unused image data.
Cleanup can be performed with docker system prune
command.
If you run into disk space errors, we recommend you prune your docker images using the docker system prune --all
command. You might need to Stop the environment or restart the docker engine before running this command.
You can check if your docker is clean by running docker images --all
and docker ps --all
- both should return an empty list of images and containers respectively.
If you are on Mac OS and you end up with not enough disk space for Docker you should increase disk space available for Docker. See Docker for Mac - Space for details.
If you are having problems with the Docker Compose environment - try the following (after each step you can check if your problem is fixed)
In case the problems are not solved, you can set VERBOSE variable to “true” (export VERBOSE="true"
) and rerun failing command, and copy & paste the output from your terminal, describe the problem and post it in Airflow Slack #troubleshooting channel.
Pre-commit hooks are fantastic way of speeding up your local development cycle. Those pre-commit checks will only check the files that you are currently working on which make them fast. Yet they are using exactly the same environment as the CI checks are using, so you can be pretty sure your modifications will be ok for CI if they pass pre-commit checks.
You are STRONGLY encouraged to install pre-commit hooks as they speed up your development and place less burden on the CI infrastructure.
We have integrated the fantastic pre-commit framework in our development workflow. You need to have python 3.6 installed in your host in order to install and use it. It's best to run your commits when you have your local virtualenv for Airflow activated (then pre-commit and other dependencies are automatically installed). You can also install pre-commit manually using pip install
.
The pre-commit hooks require Docker Engine to be configured as the static checks static checks are executed in docker environment. You should build the images locally before installing pre-commit checks as described in Building the images. In case you do not have your local images built the pre-commit hooks fail and provide instructions on what needs to be done.
pre-commit install
Running the command by default turns on pre-commit checks for commit
operations in git.
You can also decide to install the checks also for pre-push
operation:
pre-commit install -t pre-push
You can see advanced usage of the install method via
pre-commit install --help
Before running the pre-commit hooks you must first build the docker images locally as described in Building the images chapter.
Sometimes your image is outdated (when dependencies change) and needs to be rebuilt because some dependencies have been changed. In such case the docker build pre-commit will fail and inform you that you should rebuild the image with REBUILD=“true” environment variable set.
The pre-commit hooks use several external linters that need to be installed before pre-commit are run. Most of the linters are installed by running pip install -e .[devel]
in the airflow sources as they are python-only, however there are some that should be installed locally using different methods. In Linux you typically install them with sudo apt install
on MacOS with brew install
.
The current list of prerequisites:
sudo apt install xmllint
, MacOS - install via brew install xmllint
pip install yamllint
In airflow we have the following checks:
check-hooks-apply Check hooks apply to the repository check-apache-license Checks compatibility with Apache License requirements check-merge-conflict Checks if merge conflict is being committed check-executables-have-shebangs Check that executables have shebang check-xml Check XML files with xmllint doctoc Refresh table-of-contents for md files detect-private-key Detects if private key is added to the repository end-of-file-fixer Make sure that there is an empty line at the end flake8 Run flake8 forbid-tabs Fails if tabs are used in the project insert-license Add licences for most file types lint-dockerfile Lint dockerfile mixed-line-ending Detects if mixed line ending is used (\r vs. \r\n) mypy Run mypy shellcheck Check shell files with shellcheck yamllint Check yaml files with yamllint
The check-apache-licence check is normally skipped for commits unless .pre-commit-config.yaml
file is changed. This check always run for the full set of files and if you want to run it locally you need to specify --all-files
flag of pre-commit. For example:
pre-commit run check-apache-licenses --all-files
After installing pre-commit hooks are run automatically when you commit the code, but you can run pre-commit hooks manually as needed.
You can run all checks on your staged files by running: pre-commit run
You can run only one mypy check on your staged files by running: pre-commit run mypy
You can run only one mypy checks manually on all files by running: pre-commit run mypy --all-files
You can run all checks manually on all files by running: pre-commit run --all-files
You can skip one or more of the checks by specifying comma-separated list of checks to skip in SKIP variable: SKIP=flake8,mypy pre-commit run --all-files
You can always skip running the tests by providing --no-verify
flag to git commit
command.
You can check other usages of pre-commit framework at Pre-commit website
Before you submit a pull request from your forked repo, check that it meets these guidelines:
We currently rely heavily on Travis CI for running the full Airflow test suite as running all of the tests locally requires significant setup. You can setup Travis CI in your fork of Airflow by following the Travis CI Getting Started guide.
There are two different options available for running Travis CI which are setup as separate components on GitHub:
Once installed, you can configure the Travis CI GitHub App at https://github.com/settings/installations -> Configure Travis CI.
For the Travis CI GitHub App, you can set repository access to either “All repositories” for convenience, or “Only select repositories” and choose <username>/airflow
in the dropdown.
You can access Travis CI for your fork at https://travis-ci.com/<username>/airflow
.
The Travis CI GitHub Services versions uses an Authorized OAuth App. Note that apache/airflow
is currently still using the legacy version.
Once installed, you can configure the Travis CI Authorized OAuth App at https://github.com/settings/connections/applications/88c5b97de2dbfc50f3ac.
If you are a GitHub admin, click the “Grant” button next to your organization; otherwise, click the “Request” button.
For the Travis CI Authorized OAuth App, you may have to grant access to the forked <organization>/airflow
repo even though it is public.
You can access Travis CI for your fork at https://travis-ci.org/<organization>/airflow
.
The travis-ci.org site for open source projects is now legacy and new projects should instead be created on travis-ci.com for both private repos and open source.
Note that there is a second Authorized OAuth App available called “Travis CI for Open Source” used for the legacy travis-ci.org service. It should not be used for new projects.
More information:
When developing features the need may arise to persist information to the the metadata database. Airflow has Alembic built-in to handle all schema changes. Alembic must be installed on your development machine before continuing.
# starting at the root of the project $ pwd ~/airflow # change to the airflow directory $ cd airflow $ alembic revision -m "add new field to db" Generating ~/airflow/airflow/migrations/versions/12341123_add_new_field_to_db.py
airflow/www_rbac/
contains all npm-managed, front end assets. Flask-Appbuilder itself comes bundled with jQuery and bootstrap. While these may be phased out over time, these packages are currently not managed with npm.
Make sure you are using recent versions of node and npm. No problems have been found with node>=8.11.3 and npm>=6.1.3
First, npm must be available in your environment. If you are on Mac and it is not installed, you can run the following commands (taken from this source):
brew install node --without-npm echo prefix=~/.npm-packages >> ~/.npmrc curl -L https://www.npmjs.com/install.sh | sh
The final step is to add ~/.npm-packages/bin
to your PATH
so commands you install globally are usable. Add something like this to your .bashrc
file, then source ~/.bashrc
to reflect the change.
export PATH="$HOME/.npm-packages/bin:$PATH"
You can also follow the general npm installation instructions.
To install third party libraries defined in package.json
, run the following within the airflow/www_rbac/
directory which will install them in a new node_modules/
folder within www/
.
# from the root of the repository, move to where our JS package.json lives cd airflow/www_rbac/ # run npm install to fetch all the dependencies npm install
To parse and generate bundled files for airflow, run either of the following commands. The dev
flag will keep the npm script running and re-run it upon any changes within the assets directory.
# Compiles the production / optimized js & css npm run prod # Start a web server that manages and updates your assets as you modify them npm run dev
Should you add or upgrade an npm package, which involves changing package.json
, you'll need to re-run npm install
and push the newly generated package-lock.json
file so we get the reproducible build.
We try to enforce a more consistent style and try to follow the JS community guidelines. Once you add or modify any javascript code in the project, please make sure it follows the guidelines defined in Airbnb JavaScript Style Guide. Apache Airflow uses ESLint as a tool for identifying and reporting on patterns in JavaScript, which can be used by running any of the following commands.
# Check JS code in .js and .html files, and report any errors/warnings npm run lint # Check JS code in .js and .html files, report any errors/warnings and fix them if possible npm run lint:fix