This document contains step-by-step details for how to release Habitat.
Note: this document covers what to do once you decide you want to make an official release. If you would like further details on how the overall pipeline works, please go here.
In Slack, execute the following command:
/expeditor promote habitat acceptance
This will take whatever packages are currently in the acceptance
channel and promote them into the staging
channel. This manual
action is the only way things get into the staging
channel, so you
can perform whatever manual validation you need without worrying about
new packages coming in and invalidating your efforts.
For each platform
(darwin,
linux,
linux-kernel2,
windows),
download the latest release candidate CLI from packages.chef.io
. You
must have run the /expeditor
Slack command above before
downloading the package!
Alternatively, you can either use our "curlbash" installation script
to install the above packages, or (if you're running on a system that
already has and older version of hab
installed) you can upgrade
using hab
itself, since the same CLI is available from
packages.chef.io
as well as Builder.
Run either of the following:
curl https://raw.githubusercontent.com/habitat-sh/habitat/master/components/hab/install.sh \
| sudo bash -s -- -c staging
sudo hab pkg install core/hab --binlink --force --channel=staging
Run either of the following commands:
curl https://raw.githubusercontent.com/habitat-sh/habitat/master/components/hab/install.sh \
| sudo bash -s -- -c staging -t x86_64-linux-kernel2
(Note the addition of the target option on the "curlbash" command! Without this you will end up with the modern Linux version, and things won't work properly!)
sudo hab pkg install core/hab --binlink --force --channel=staging
curl https://raw.githubusercontent.com/habitat-sh/habitat/master/components/hab/install.sh \
| sudo bash -s -- -c staging
You cannot (yet) update using hab
itself due to how the CLI is
currently installed on macOS.
iex "& { $(irm https://raw.githubusercontent.com/habitat-sh/habitat/master/components/hab/install.ps1) } -Channel staging"
There may be special behavior related to this release that you will want to validate but at the very least, you should validate the version, try running some services, and exercising the Studio.
To ensure everything is working properly, you will need to have the following environment variables set:
export HAB_INTERNAL_BLDR_CHANNEL=staging
export HAB_STUDIO_SECRET_HAB_INTERNAL_BLDR_CHANNEL=staging
On Windows, only HAB_INTERNAL_BLDR_CHANNEL
needs to be set:
$env:HAB_INTERNAL_BLDR_CHANNEL="staging"
HAB_INTERNAL_BLDR_CHANNEL
is needed outside the Studio in order to
install the correct core/hab-studio
package to begin with. If you already
have this installed when executing hab studio enter
, you can leave
this variable out.
HAB_STUDIO_SECRET_HAB_INTERNAL_BLDR_CHANNEL
is needed so that the
inside the studio can install the Launcher and Supervisor
packages for the internal Supervisor from the appropriate place.
See #4656 for further context and ideas.
Here is how you can validate the version of the cli, supervisor and the supervisor version inside the studio.
On Linux:
hab --version
hab sup --version
hab studio enter
hab --version
sup-log
hab studio enter -D
hab --version
sup-log
On Windows:
hab --version
hab sup --version
hab studio enter
hab --version
Get-Supervisorlog
hab studio enter -D
hab --version
Get-Supervisorlog
Here are examples of what you might do with a Studio.
First, set up the studio properly. We clone core-plans
to have
things to build, and we run hab studio rm
to ensure a clean slate.
mkdir testing
cd testing
git clone https://github.com/habitat-sh/core-plans
export HAB_INTERNAL_BLDR_CHANNEL=staging
export HAB_STUDIO_SECRET_HAB_INTERNAL_BLDR_CHANNEL=staging
hab studio rm
hab studio enter
Then, once inside the Studio, you could try these:
# Does the version of the cli inside the studio match staging?
hab --version
# Does the version of the supervisor inside the studio match staging?
sup-log
^C
# build the apache plan
build core-plans/httpd
source results/last_build.env
hab svc load $pkg_ident
sup-log
# Is Apache running
^C
# Is it responding?
hab pkg exec core/busybox-static wget http://localhost -S
On Linux, testing the Docker studio is identical, except you enter using the following command instead:
hab studio enter -D
Test both in the chroot and docker studios on x86 linux.
For this PackageTarget it is important that you perform validation on a Linux system running a 2.6 series kernel. CentOS 6 is recommended because it ships with a kernel of the appropriate age, but any distro with a Kernel between 2.6.32 and 3.0.0 can be used. Included in the support/validation/x86_64-linux-kernel2
directory in this repository is a Vagrantfile that will create a CentOS-6 VM to perform the validation. You can also run a VM in EC2.
The Vagrantfile is configured to grab the
core-plans repository (to
give you something to build), as well as grab the secret key for your
HAB_ORIGIN
(using the HAB_ORIGIN
and HAB_AUTH_TOKEN
variables in
your environment). Additionally, it will automatically install the
release candidate hab
binary from the staging
channel unless you
explicitly override that with the INSTALL_CHANNEL
variable (see below).
export HAB_ORIGIN=...
export HAB_AUTH_TOKEN=...
# Only if you *don't* want the staging artifact, for some reason
export INSTALL_CHANNEL=...
vagrant up
vagrant ssh
Once inside the VM, set your override environment variables (as above) and experiment. For example:
export HAB_INTERNAL_BLDR_CHANNEL=staging
export HAB_STUDIO_SECRET_HAB_INTERNAL_BLDR_CHANNEL=staging
export HAB_ORIGIN=<my_origin>
hab studio enter
Then, once inside the Studio, you could try these:
# Does the version of the cli inside the studio match staging?
hab --version
# Does the version of the supervisor inside the studio match staging?
sup-log
^C
# build the apache plan
build core-plans/redis
source results/last_build.env
hab svc load $pkg_ident
sup-log
# Is Redis running and reporting "Ready to accept connections"
^C
If all your manual validation works out, you should promote the
release candidate from the staging
channel to the current
channel. Once again, we use the Slack command:
/expeditor promote habitat staging
This will result in the Supervisors in our Production Builder updating themselves to the release candidate.
The Pending Release Notes
on the habitat githib repo's wiki is a placeholder for content that is published to the chef.io release notes and the discource release announcement. The finish_release
pipeline which runs after promoting habitat to stable in expeditor automates the process of extracting the pending notes and publishing to chef.io and discource. It is important to make sure that the wiki page is updated before promoting to stable.
Once you're satisfied with the new Supervisors in Production (you can access and navigate around https://bldr.habitat.sh/), you can
finish the release process by promoting from the current
channel to
the stable
channel.
/expeditor promote habitat current
This places all the release candidates into the stable
channel,
making them "officially" available to the world, and thus "released".
The Buildkite release is fairly-well automated at this point, but once it is complete, there are still a few remaining manual tasks to perform. In time, these will be automated as well.
The Habitat Homebrew repository will automatically generate a PR to update the Homebrew tap after a release is promoted to stable. This PR will be tested to ensure the updated version is able to install. If the PR checks are green, it is safe to merge and this will update our brew install
ed version to the current stable release.
We currently use Expeditor (an internal tool) to partially manage our changelog. It adds items to CHANGELOG.md
for every PR that is merged, based on certain labels that are attached to the PR. This is all well and good.
However, due to how our pipeline is structured, we can't really use Expeditor's built-in changelog rollup functionaltiy.
(The rollup is something Expeditor does at the current head of master; when we release, we are almost always going to be dealing with artifacts that trail the current head of master by one or more commits. If we were to use Expeditor, we would either be doing a rollup for a "release" that hadn't been fully validated yet (and might indeed never be formally released), or doing a rollup for a "release" that claims to include features that were only merged after the validated release candidates were built!)
All this means that we must manually manage the rollup headers to
match the contents of the release candidate. In practice, this should
amount to moving around a few of Expeditor's HTML comments, and
creating a new header for the just-released version. Take a look at
the contents of the CHANGELOG.md
file for the pattern. You should
also consult Expeditor's CHANGELOG
documentation
for additional details.
If there were any problems with the release process that may benefit from changes to code (including release automation itself), write up a brief description of what happened and schedule a meeting with appropriate stakeholders to scope and prioritize the work. This isn't a full-blown postmortem as described in https://github.com/chef/oc_post_mortems, so it should be quick and relatively informal, but the fundamental goals are the same:
- While context is fresh, help the team understand the what happened and why without placing blame or speaking in counter-factuals (🙆: "I did…/I thought…", 🙅♂️: "I should've…/I would've…")
- Agree on, assign and prioritize remediation items to ensure continuous improvement of our release process and codebase more generally
If the release truly had no problems at all, add a "Yay!" to the retro board and celebrate our success as a team.
In general, if an escaped defect is discovered in a Habitat release, the first solution should be to "roll forward". That is, a patch that addresses the defect (either by fixing it or reverting the code responsible) should be merged to master, and then a new release is built from that. This is the desired solution because it requires no extraordinary measures by end-users to take advantage of the fix; they only need to update their software as they would for any "normal" release.
In the rare case that that is not possible (e.g., the cause of the defect is not readily apparent), you can remove or "yank" the offending release, and fall back to the previous release. This is not a perfect solution, however. The process and drawbacks are detailed here.
It should be noted that yanking a release is a last resort action; all efforts should be made to roll forward first. Here is a (non-exhaustive) list of possible situations that may call for yanking a release:
- The cause of the defect isn't apparent, but definitely isn't present in the previous release.
- The defect is severe enough that waiting for a new release isn't acceptable (becomes less an issue as our release automation matures).
- A simple revert wouldn't work due to infrastructure dependencies such as schema changes and properly testing the forward fix will be too time-consuming.
-
Demote all packages produced during a release from the
stable
channel in Builder.This includes not only the
hab
CLI and Supervisor-related binaries, but everything, including Studio and exporters, and for all supported platforms. The list of specific package releases can be found in themanifest.json
file located athttps://chef-automate-artifacts.s3-us-west-2.amazonaws.com/files/habitat/${VERSION}/manifest.json
-
Put the last-known-good artifacts into the
stable
directory in S3. This amounts to copying the contents ofhttps://chef-automate-artifacts.s3-us-west-2.amazonaws.com/files/habitat/${LAST_GOOD_VERSION}/
into
https://chef-automate-artifacts.s3-us-west-2.amazonaws.com/stable/latest/habitat/
This makes it so our "curlbash" installer will pull the correct packages.
-
Revert the version's change for the Homebrew tap
Clone the repository from https://github.com/habitat-sh/homebrew-habitat and revert the commit for the offending release (see this example).
-
Unlist the Chocolatey version for Windows
Chocolatey versions cannot, strictly speaking, be deleted, but only unlisted. However, since the Chocolatey release ultimately pulls binaries from Bintray currently, it ends up being a distinction without a difference.
To unlist a version, go to https://chocolatey.org/packages/habitat/ and scroll down to the
Version History
listing. Find the offending version and click the link for it in theListed
column (the link text will be "yes"). This will take you to a page where you can choose to unlist the package. -
Container Images
There are no steps currently needed to yank any container images, since the binaries pull them in specifially by version tag.
-
User Remediation Steps
Though the above steps will prevent future users from installing the offending release, it does nothing to fix users that have already downloaded the release. These are the steps they will need to take to "roll back" to the last stable release. These steps should be included in any communications that are sent out about the release.
Rolling forward is preferable to yanking a release, because rolling forward does not require these manual remediation steps.
(These examples are for the 0.80.0 release; change the versions as required.)
On Linux:
hab pkg uninstall core/hab/0.80.0 hab pkg uninstall core/hab-sup/0.80.0 rm /hab/cache/artifacts/core-hab-0.80.0-* rm /hab/cache/artifacts/core-hab-sup-0.80.0-*
If a Launcher release must be pulled, make note of the version and delete any installed versions of it:
hab pkg uninstall core/hab-launcher/${VERSION} rm /hab/cache/artifacts/core-hab-launcher-${VERSION}-*
On Windows (Chocolatey) :
hab pkg exec core/windows-service uninstall hab pkg uninstall core/windows-service choco uninstall habitat --version 0.80.0 hab pkg uninstall core/hab-sup/0.80.0 choco install habitat
On macOS (Homebrew):
brew uninstall hab brew install hab