This repo contains the release infrastructure for Kubernetes.
Live Kubernetes releases are done by the Kubernetes team at Google due to permissions and other restrictions. This may expand eventually to allow other Kubernetes contributors to generate releases.
The current tooling runs by default in mock mode and anyone should be able to run it in this mode to see exactly how the process works. In mock mode all the code paths are followed for a release, but nothing is pushed to repositories.
Sticking with the ancient Greek theme, the release script is called anago.
Anago means, in the context of navigators and shipping:
"to launch out, set sail, put to sea."
Tools in this repository includes a familiar *nix-style man page with usage, process and examples. The link shows how the self-contained doc/man page makes up the header of the script itself and the same info is available on the command-line (or get usage simply by calling the script with no options):
$ anago -man
The idea is that no external doc updates should be necessary and the tool itself contains all of the details and instructions and prerequisite checks needed for anyone to run the tool in at least mock mode.
There is a simple $USER check to ensure that no one but a certain few people can run the script with --nomock to perform a real release.
NEW
anago is entirely re-entrant with both stateful and stateless steps.
Upon initial execution, a WORKFLOW TOC (table of contents) is provided that
enumerates the stateful steps that will be followed based on context and
command-line args. NOTE: Stateless steps are not called out in the TOC.
- Completed steps will show with a checkmark
- Not-yet-run steps will show a box
- Stateful steps will show a TITLE and index (n/N) to show progress
- /tmp/anago-runstate contains the state in the following format:
- CMDLINE - stores the original critical command-line entries
- entry_points+$label - Entry points will appear as they are completed with an optional +label to distinguish those that are used with arguments
- entry_points+$label name=value - Entry points that set stateful name=value pairs will have the name=value on the same line
- Use --clean to reset the state and start over
The tool was designed to require minimal inputs.
The only information the tool needs is to know where you want to create a
release with one optional flag [--official] (used on release-* branches only).
The [--official] flag indicates that a standard patch release will be created
on a branch. Without the [--official] flag, a beta would be created.
There are two workflows you can choose from:
- Run through a complete release end to end
- Create any number of staged (--stage) releases and release from there
First try a staged alpha release:
$ anago master --stage
Later, release the staged bits: (Artifacts are cached both locally and on GCS)
# Set --buildversion based on the final output of the staging build
$ anago master --buildversion=<staging build version>
Try a beta release on a branch:
$ anago release-1.2
Try an official release on a branch:
$ anago release-1.2 --official
Try a beta release on a new branch:
$ anago release-9.9
NOTE: You can first stage and then release on master or any supported branch.
Try creating a new branch and beta for an emergency zero-day fix. See docs/branching.md for more details.
(The branch name should reflect the branch point/tag. So if branching at the v9.9.9 tag on the release-9.9 branch, create a release-9.9.9 branch):
$ anago release-9.9.9
Stage an official (patch) release on your local disk:
# add --build-at-head to force a build, otherwise rely on find_green_build
# in-line to find a build candidate
$ anago release-1.8 --stage
Release previously staged official (patch) release from your local disk:
# $buildversion will come from the output at the end of the above --stage run
# as will this command-line in its entirety
$ anago release-1.8 --buildversion=$buildversion
Anago is currently locked down to only run for a specific set of individuals.
when --nomock is specified.
Adding that flag to the command-line indicates the release will push tags and artifacts. The user is still prompted before a push occurs, however.
How can I manually specify a build to use when anago (find_green_build) can't automatically locate a green run?
The output from anago (or find_green_build) displays --buildversion in its output while trying to locate a build. The user is welcome to pass any --buildversion value to anago to create a release at a particular hash.
All standalone scripts have embedded man pages. Just use -man to view or
your favorite editor.
- prin : What tags/releases is my PR IN?
- find_green_build : Ask Jenkins for a good build to use
- script-template : Generate a script template in the kubernetes/release ecosystem
- relnotes : Scrape github for release notes (See below for more info)
- branchff : Fast-forward branching helper
- changelog-update : Update CHANGELOG.md version entries by rescanning github for text and label changes
- push-build.sh : Push a developer (or CI) build up to GCS
# get details on how to use the tool
$ relnotes -man
$ cd /kubernetes
# Show release notes from the last release on a branch to HEAD
$ relnotes
# Show release notes from the last release on a specific branch to branch HEAD
$ relnotes --branch=release-1.2
# Show release notes between two specific releases
$ relnotes v1.2.0..v1.2.1 --branch=release-1.2
Please report any issues you encounter.
You can build the deb packages in a Docker container like this:
docker build --tag=debian-packager debian
docker run --volume="$(pwd)/debian:/src" debian-packager
The build runs for a while, after it's done you will find the output in debian/bin.
You can build the rpm packages in a Docker container with:
cd rpm
./docker-build.sh
Resulting rpms, and a pre-generated yum repository will be generated in rpm/output/x86_64.