Skip to content

Latest commit

 

History

History
92 lines (66 loc) · 4.58 KB

RELEASE.md

File metadata and controls

92 lines (66 loc) · 4.58 KB
Raw

Release Procedure :shipit:

Branching Strategy

dbt has three types of branches:

  • Trunks track the latest release of a minor version of dbt. Historically, we used the master branch as the trunk. Each minor version release has a corresponding trunk. For example, the 0.11.x series of releases has a branch called 0.11.latest. This allows us to release new patch versions under 0.11 without necessarily needing to pull them into the latest version of dbt.
  • Release Branches track a specific, not yet complete release of dbt. These releases are codenamed since we don't always know what their semantic version will be. Example: dev/lucretia-mott became 0.11.1.
  • Feature Branches track individual features and fixes. On completion they should be merged into a release branch.

Git & PyPI

  1. Update CHANGELOG.md with the most recent changes
  2. If this is a release candidate, you want to create it off of your release branch. If it's an actual release, you must first merge to a master branch. Open a Pull Request in Github to merge it into the appropriate trunk (X.X.latest)
  3. Bump the version using bumpversion:
  • Dry run first by running bumpversion --new-version <desired-version> <part> and checking the diff. If it looks correct, clean up the chanages and move on:
  • Alpha releases: bumpversion --commit --tag --new-version 0.10.2a1 num
  • Patch releases: bumpversion --commit --tag --new-version 0.10.2 patch
  • Minor releases: bumpversion --commit --tag --new-version 0.11.0 minor
  • Major releases: bumpversion --commit --tag --new-version 1.0.0 major
  1. (If this is a not a release candidate) Merge to x.x.latest and (optionally) master.
  2. Update the default branch to the next dev release branch.
  3. Build source distributions for all packages by running ./scripts/build-sdists.sh. Note that this will clean out your dist/ folder, so if you have important stuff in there, don't run it!!!
  4. Deploy to pypi
  • twine upload dist/*
  1. Deploy to homebrew (see below)
  2. Deploy to conda-forge (see below)
  3. Git release notes (points to changelog)
  4. Post to slack (point to changelog)

After releasing a new version, it's important to merge the changes back into the other outstanding release branches. This avoids merge conflicts moving forward.

In some cases, where the branches have diverged wildly, it's ok to skip this step. But this means that the changes you just released won't be included in future releases.

Homebrew Release Process

  1. Clone the homebrew-dbt repository:
git clone [email protected]:fishtown-analytics/homebrew-dbt.git
  1. For ALL releases (prereleases and version releases), copy the relevant formula. To copy from the latest version release of dbt, do:
cp Formula/dbt.rb Formula/dbt@{NEW-VERSION}.rb

To copy from a different version, simply copy the corresponding file.

  1. Open the file, and edit the following:
  • the name of the ruby class: this is important, homebrew won't function properly if the class name is wrong. Check historical versions to figure out the right name.
  • under the bottle section, remove all of the hashes (lines starting with sha256)
  1. Create a Python 3.7 virtualenv, activate it, and then install two packages: homebrew-pypi-poet, and the version of dbt you are preparing. I use:
pyenv virtualenv 3.7.0 homebrew-dbt-{VERSION}
pyenv activate homebrew-dbt-{VERSION}
pip install dbt=={VERSION} homebrew-pypi-poet

homebrew-pypi-poet is a program that generates a valid homebrew formula for an installed pip package. You want to use it to generate a diff against the existing formula. Then you want to apply the diff for the dependency packages only -- e.g. it will tell you that google-api-core has been updated and that you need to use the latest version.

  1. reinstall, test, and audit dbt. if the test or audit fails, fix the formula with step 1.
brew uninstall --force Formula/{YOUR-FILE}.rb
brew install Formula/{YOUR-FILE}.rb
brew test dbt
brew audit --strict dbt
  1. Ask Connor to bottle the change (only his laptop can do it!)

Conda Forge Release Process

  1. Clone the fork of conda-forge/dbt-feedstock here
git clone [email protected]:fishtown-analytics/dbt-feedstock.git
  1. Update the version and sha256 in recipe/meta.yml. To calculate the sha256, run:
wget https://github.com/fishtown-analytics/dbt/archive/v{version}.tar.gz
openssl sha256 v{version}.tar.gz

  1. Push the changes and create a PR against conda-forge/dbt-feedstock

  2. Confirm that all automated conda-forge tests are passing