| Description | Location |
|---|---|
| Release orchestration | Link |
| Manual release management | Link |
| Deployment tool | Link |
| Patcher tool (Private) | Link |
| Release documentation | Link |
| Release related tasks issue tracker | Link |
| Delivery team issue tracker | Link |
| Creating patch release for self-managed release | Youtube Video |
| GitLab.com auto-deploy run through | Youtube Video |
The self-managed release is a collection of many successfully deployed auto-deploy
releases on GitLab.com.
The main priority of both types of releases is GitLab availability & security as an application running on both GitLab.com and for customers running GitLab in their own infrastructure.
With these two types of releases, GitLab Inc. has to balance at the same time the workflows normally found in SaaS companies with the ones found in companies that publish packaged software.
GitLab.com receives regular updates according to our auto-deploy schedule. This means that GitLab.com lives in the now and semver versioning used by self-managed installations does not apply to it.
For example, developers might be working in the 12.3 milestone on items that
will be a part of 12.3.0 release but GitLab.com will only "know" of the commit
SHA that the developer created.
Users on GitLab.com therefore receive features and bug fixes earlier than users of self-managed installations.
The self-managed release is a semver versioned package of features that
are already released on GitLab.com through auto-deploys.
This means that the official published version of GitLab is a historical
snapshot of items that are released to users on GitLab.com. As such,
the self-managed release is created from a backport branch named by the
targeted semver version with a stable suffix, eg. 12-3-stable.
Self-managed users receive features and bug fixes only after a new semver version is created and published.
Our maintenance policy describes in detail the cadence of our major, minor and patch releases for self-managed users. The major release yearly cadence was defined after an all stakeholder discussion.
The only guaranteed date throughout the release cycle is the 22nd. On this date,
the self-managed release will be published together with the release announcement.
All other dates are a guideline only and cannot be considered a deadline when it comes to what will be included into any type of release. This includes the development month and the dates defined there as well as any promises given to customers. This is strictly because there are a lot of moving parts to be considered when preparing a release which includes working on highest priority and severity issues as well as security related issues.
Keep in mind that, if it is absolutely necessary to get a certain feature
ready for a specific version, merge the feature early in the development cycle.
Merges closer to the release date are absolutely not guaranteed to be included
in that specific monthly self-managed release.
For GitLab.com releases, timelines are different and described below.
In June 2019, GitLab.com transitioned to more frequent deploys through auto-deploys
as a first step towards continuous deployments. The auto deploy transition document
was created to serve as a transition overview document as we progress further
towards even more frequent deployments. As we transitioned, the timelines
for GitLab.com releases have changed and the current general process is
described below.
New auto-deploy branches are created daily.
Once a new branch is created, only commits that pass the CI tests are eligible for deployments ("green build"). This means that if specs are failing in gitlab-org/gitlab, the deployments cannot progress further.
Automated tasks in the release-tools project are setup to drive the next steps:
~"Pick into auto-deploy" (See Labels of importance).When a new package is built, it is automatically deployed to staging.gitlab.com.
As part of the deployment to this environment, a set of automated QA integration tests are run.
When the automated QA test pass, the deployment automatically progresses to the canary stage.
Automated deployments to any production environment (including canary), are halted during the change lock period. Currently, the change lock period is:
During the change lock period, manual deployment can be triggered through GitLab ChatOps if the deployment fixes a S1 availability or security issue.
When the release reaches the canary stage and no new exceptions or alerts are reported, release is considered to be ready for deployment to GitLab.com.
The promotion to the rest of the GitLab.com is triggered manually by the release managers and this can happen at any point in time.
The release managers will deploy any amount of code changes to GitLab.com without a manual approval when:
Each deployment will trigger a notification in the Slack channel #announcements. After each successful deployment, QA issue is created in release/tasks issue tracker. The purpose of these notifications are to inform the people who are involved in the process that their change is going through environments. This allows them to execute any manual testing or other tasks related to the release of their fix/feature.
The self-managed release timelines are concentrated around the 22nd. One week
before the release date, the release managers will start preparing for the release
by ensuring that the GitLab.com releases are in a consistent state. This means
that what gets into the self-managed release fully depends on the state of
GitLab.com releases.
The self-managed release will only contain code that is successfully running
on GitLab.com at the time the release manager decides to finalise the release.
This can be as early as 5 work days or as late as 1 day before the 22nd. This fluctuation comes from a possibility of 22nd being at or right after the weekend or instability on GitLab.com in the run-up to the 22nd. The release manager preparing the final release will post an announcement in #releases, #development and #g_delivery with a link to the final branch.
As mentioned earlier, if it is absolutely necessary to include a certain feature/fix
in that months self-managed release, merge it early in the development cycle or
no later than a couple of weeks before.
The same applies for items that need to go to the next
release; if it is absolutely necessary to include it in the next months release, merge
it after the self-managed release has been prepared.
Just because a merge request has a milestone of the current release XX.YY set,
does not mean that it will end up in the same release. Milestone describes
the target release, but only guarantee that it will be included in the same
release is if it is running on GitLab.com around the time release preparation starts.
For both types of releases, there are a few labels of specific importance.
For code that needs to be deployed to GitLab.com with higher priority than the
regular cadence, we have ~"Pick into auto-deploy" label.
The automated systems that create a new GitLab.com release will look for this label specifically, and any merge request with this label and an S1/S2 severity label will be automatically cherry-picked into the active auto-deploy branch. In case the merge request cannot be picked, which can happen if there is a conflict in the files being picked, the message will be posted in the merge request asking the author to create a new merge request targeting the currently active release branch.
The label should be only used under the following circumstances, when the merge request:
If this label is added because a merge request is blocking further deploys, consider leaving a note in #releases Slack channel to raise awareness of the status.
For new features or non-urgent fixes, the label should not be used because the new release is only days or hours away.
Directions on how to how to know whether a MR is deployed to GitLab.com are in the release/docs.
Similar to the above mentioned label, each self managed release has a label
to highlight that a certain merge request should be backported to the targeted
release. For example, releases in the 12.3 series will have ~"Pick into 12.3"
label which will signal that this merge request should be included in one of the
next patch releases being created. The patch releases are created as needed but only according to our maintenance policy.
The label should be applied in the following situations:
The label should not be applied to merge requests:
You can find this out by taking a look at the GitLab Release Managers schedule.
A release candidate (RC) is a GitLab package that includes the changes that will make it into the final self-managed release, except for the rare case where a change may need to be reverted. RCs are only created for the monthly self-managed release, not patch releases. The amount of RCs created per month will vary per release.
There is no fixed point in time where a release manager creates a release candidate. Instead, this is based entirely on how the release process has been going, what the state is of GitLab.com, etc.
Release candidates are created whenever possible, and as such there are no guarantees on creation timing. This will depend on factors such as:
In other words, if you want to know when a release candidate is created your best option is to join one of the following Slack channels:
It is up to a release manager to decide when to create a release candidate, taking into account the state of deployments and GitLab.com.
Please do not message a release manager in private about release related questions or requests. Instead, post your request/question in the #releases channel.
When we create release candidates, the Release Tools project will post a comment on the merge requests included in the release candidate (unless they were included in a prior release candidate). When you see this comment you will know the merge request will be included in the release, unless it is reverted in a later release candidate.
For more information, refer to the Auto-deploy status command.
In the runup to the 22nd, release managers will also start announcing what commit will at the very least make it into the release. Such notifications are shared in Slack #releases channel and will look something like this (format is defined in the release-tools monthly template):
This is the candidate commit to be released on the 22nd. https://gitlab.com/gitlab-org/gitlab/commits/4033cb0ed66de77a4b5c1936e33de918edef558e
The last commit to make it into the release will have a message similar to this:
:mega: Barring any show-stopping issues, this is the final commit to be released on the 22nd. https://gitlab.com/gitlab-org/gitlab/-/commits/13-1-stable-ee
The earlier in the monthly cycle your MR is merged, the higher the chances are for it to be included in that months release.
There is no guaranteed "cut-off", "freeze" or any other date defined under which the MR will be included.
Availability, security and performance of GitLab.com is a pre-requisite for any monthly self-managed release. If GitLab.com is not experiencing any issues, MR's merged as late as the 20th of the month were included in the release. On the opposite side, when GitLab.com stability was lower, MR's merged as early as 15th of the month were not included.
In other words:
The quality and stability of what is delivered by everyone defines the final MR that will be included in the monthly release.
For more detailed answer, see self-managed release timelines.
We currently create an auto-deployment branch at a specific timeline.
For a merge request with a specific label, the process is different.
See this guide for more information.
If you've found a regression while checking the QA issue created in the release/tasks project, check the issue description for more details on what to do next.
If you believe that the regression has a high severity, immediately inform Release Managers in the QA issue by mentioning @gitlab-org/release/managers.
In addition to that, engage with the releases managers in the #releases channel and confirm with them whether deployments need to be halted.
If a regression is found in a new feature, and only that feature is affected, follow the directions in the QA issue for a regular regression.
The different processes are documented here:
