Stackable develops and maintains a growing number of operators for open source software.
The high level structure of the repository is consistent across all repositories, which in the past has led to repetitive maintenance tasks in many repositories even for small changes.
This repository is intended to help with these changes by keeping common files as templates and offering tooling to roll these out to all repositories that are under management by this tool.
The definition files for GitHub actions that in turn execute the playbooks doing the actual work are kept in .github
Everything under the top level folder template
is replicated to the target repositories.
The folder and file structure under template
is considered as source structure which will be synchronized across all repositories in the following manner:
-
Regular files and folders are simply copied
-
Files with an extension of
.j2
will be processed as jinja2 templates and copied to the target repositories with the.j2
extension removed. The default jinja2 variable delimiter has been replaced with{[ }]
since a lot of the template files contain{{ }}
and caused issues. -
File and directory names which contain
[[product]]
will have this substituted for the value of theproduct_string
variable. For example: "stackable-[[product]]-operator.service" would becomestackable-kafka-operator.service
.
To remove files or directories that already exist in the target repositories these need to be configured in repositories.yaml
under the retired_files
key.
Anything that is listed here will be deleted from the target repositories.
Note
|
Deletion is the last step that is performed, so if there is an overlap between files existing in the template folder and this setting, the files would not be rolled out, since they’d get deleted before creting the pull request. |
All user-facing configuration is kept in repositories.yaml
.
The actions in this repository are set up so that the playbooks are automatically executed whenever a commit is made to the main
branch of this repository.
So in principle it is as simple as pushing any commit to main
which will trigger a sync of everything under template
to all target repositories.
Target repositories are configured in repositories.yaml
in the following form:
- name: kafka-operator url: stackabletech/kafka-operator.git product_string: kafka pretty_string: Kafka
Field | Description |
---|---|
name |
This is only used internally to name working directories and the like. |
url |
The github repository for this operator. Need to be in the form of |
product_string |
A lower case string to use in config files, file names and the like. Should not contain whitespaces. This can sometimes be a shortened version of the full name, for example for Open Policy Agent this would be "opa" |
pretty_string |
The actual name of the product, including whitespaces and proper capitalization. This is intended to be used in doc or man files or similar things. |
include_productconfig |
Whether to include files from the Default: true |
These are the only variables currently being used on the playbooks, but can be extended easily as more are needed.
Note
|
If a new variable is introduced, it needs to be added to all repository objects! |
Additional settings can be found in playbook/group_vars/all
, but these are not intended to be freely changed and should be treated with care.
If you want to make a change that should be rolled out to all operators, make the change in the template
directory.
Consult the section above to learn more about the structure of the template.
-
Run the
test.sh
script. It will automatically delete and recreate awork
directory. -
The changes can be examined with
git status
. When the pull request is later merged into themain
branch then pull requests with these changes will be created automatically. -
Depending on the change, it makes sense to run the integration tests for all changed operators. If the tests are not run in this stage and if there is even just one integration test failing in the subsequential generated pull requests then the operator-templating must be adapted which creates again pull requests for all operators. Changes in the GitHub workflow actions cannot be tested until finally merged.
Changes are rolled out via GitHub actions.
Since this tool needs to authenticate itself in order to push commits it needs credentials of a user.
This is currently solved via a personal access token that needs to be provided as a repository (or org) secret with the name GH_ACCESS_TOKEN
.
The personal access token needs to have the following permissions for this to work:
-
repo
-
org:read
-
workflow
There is currently no synchronization with existing PRs on the target repositories whatsoever. A new pull request will be created for every commit made to this repository.
To update a PR that was created via this tool, it will have to be closed and necessary changes pushed here, which will result in a new PR.
Warning
|
The Helm Chart files that are rolled out by the templates in their current form do not include a ClusterRole object which may be needed for this to work with RBAC. |