Table of Contents generated with DocToc
Date: 2021-11-28
Draft
The Breeze is a command line development environment for Apache Airflow that makes it easy to setup Airflow development and test environment easily (< 10 minutes is the goal) and enable contributors to run any subset of tests that are executed in our CI environment easily.
The environment has proven to be very useful (it has successfully onboarded a number of new contributors, and it makes the development environment of even seasoned contributors much easier as it provides a very easy replication of the CI environment as well as very easy to setup test environment that can be used to run:
It also serves as a base for our CI execution environment. The same scripts and tools are used in our CI (based on GitHub actions). A lot of common code and function between CI and Breeze are shared between the CI and Breeze. All those tools are held in “ci” package.
Unfortunately, Breeze is largely based on Bash code - for which very few people (except maybe the Breeze creator - Jarek Potiuk, the author of this document) have any other feeling that uneasiness, disgust and fear of it :). Since Airflow is largely based on Python, the common consensus is that Breeze should be rewritten in Python.
In November 2021, Outreachy sponsored two internship for two interns: @Bowrna and @edithturn were assigned to the projects:
Breeze
- from Bash-based to Python-basedWith @potiuk, @eladkal and @xurror as mentors.
The long-standing issues about those two projects are (and we hope to close the projects during the three months internship - December 2021 - March 2022):
There are a number of problems with Bash scripts:
On the contrary Python - after dropping Python 2 end of life in January 2020, has become much more appealing as a common scripting language that can be cross-platform and ubiquitous.
This is the current state of lines of code in the project (generated by sloccount
):
SLOC Directory SLOC-by-Language (Sorted) 144905 tests python=144761,xml=132,sh=12 130115 airflow python=127249,javascript=2827,sh=39 12052 docs javascript=8977,python=2931,sh=144 9073 scripts sh=7457,python=1616 6314 chart python=6218,sh=96 3665 top_dir sh=2896,python=769 3102 dev python=2938,sh=164 1723 kubernetes_tests python=1723 280 docker_tests python=280 140 metastore_browser python=140 109 clients sh=109 28 images sh=28 Totals grouped by language (dominant language first): python: 288625 (92.65%) javascript: 11804 (3.79%) sh: 10945 (3.51%) xml: 132 (0.04%)
We have now >10K lines of shell code now. We'd announce the success of the project if the shell number is less than 300
lines of code or so, constituting less than 0.1%
of the code base.
The main decision is:
Vast majority of both Breeze and our CI scripts should be Python-based
There are likely a number of scripts that will remain in Bash, but they should contain no sophisticated logic, they should not haave common code in form of libraries and only used to execute simple tasks inside Docker containers. No Bash should ever be used in the host environment.
The “working” name of the new Breeze is “Breeze2”. We might come up with a better name in the future. In order to distinguish from the Bash version of Breeze we will always use capitalized form of Breeze as opposed to lower-case often used for the Bash version.
There are a few properties of Breeze/CI scripts that should be maintained though
It should be possible to start Breeze and run any of the CI scripts without having a specially prepared virtualenv. If the virtualenv is needed - such environment should be prepared and maintained automatically by the script being run. The idea is that new person starting their adventure with Airflow can simply run a command and get everything done with the least number of prerequisites
The prerequisites for Breeze and CI are:
There are some basic assumptions that result from our common patterns across other components we use:
rich
library for colouring terminal output. Using wisely terminal colours is an essential part of the developer experience. We will have to standardize color usage in a follow-up adrclick
library to provide command line parsing and autocompletion (in the future). Click is a comprehensive library with clean, decorator-based interface and provides rich customisation optionspytest
to run automated tests for our codeBreeze
script resides in dev/Breeze
folder, without yet linking it from main directory of Airflow. Later we will link to it from the main directory likely as Breeze2
script (in some environments where filesystem is case-insensitive (MacOS) you cannot really put two files differing only by case in the same folder.dev/Breeze
will also become a home for all the CI scripts that will be used in GitHubActions in CI.The consequences of the change should be largely invisible to the current users of Breeze. They should be able to perform the same actions and operations as in the Bash version (with possible later decision of deprecating or removing of some commands). The biggest consequence should be to the whole development community of Airflow - for them, modifying and extending and fixing Breeze and CI environment should become much more appealing.
The old script should remain and be maintained until the most important functionality of the original Breeze script has been rewritten enough.