We'd love to accept your patches and contributions to this project. There are just a few small guidelines you need to follow.
Kubeflow uses Developer Certificate of Origin (DCO).
Please see https://github.com/kubeflow/community/tree/master/dco-signoff-hook#signing-off-commits to learn how to sign off your commits.
To propose a new feature or a change that alters some existing user experience or creates a new user experience, follow these steps:
Search on KFP GitHub issues list to see if the same or similar proposal has been made in the past. The historical context can help you draft a better proposal. Sometimes you will find a very similar proposal was already presented, discussed thoroughly, and that it is either awaiting contribution (in active development) or was rejected (often due to timing or conflicting scope with other plans). To avoid confusion and conflicts, where possible, please contribute to existing issues before creating new ones.
Create a new issue using the “Feature Request” template if no existing issue is found. Fill in answers to the template questions. To avoid delays, provide as much information as needed for initial review. Keep in mind that new features should comply with backward-compatibility and platform-portability requirements.
Wait for a member from the Kubeflow Pipelines team (under orgs/kubeflow/teams/pipelines/ in org.yaml) to comment on the issue. The team aims for triaging new issues on a weekly basis, but cannot at this time provide a guarantee on when your issue will be reviewed. The team will work with you to determine if your change is trivial and can proceed or whether it is nontrivial and needs a more detailed design document and review.
If the team agreed with the overall proposal, you would be asked to write a design documentation, explaining why you want to make a change, what changes are you proposing, and how do you plan to implement it. The design review process would be required by default unless the team agreed that the change is too trivial. It is recommended that you use this Google doc template (You need to join kubeflow-discuss google group to get access) for your design, and share it with kubeflow-discuss@googlegroups.com for commenting. After sharing the design documentation, you could optionally join a session of the bi-weekly Kubeflow Pipelines community meetings [agenda] to present or further discuss your proposal. A proposal may still get rejected at this stage if it comes with unresolved drawbacks or if it does not align with the long term plans for the project.
After you get formal approval from a Kubeflow Pipelines team member, you can implement your design and send a pull request. Make sure existing tests are all passing and new tests are added when applicable. Remember to link to the feature request issue to help reviewers catch up on the context.
Kubeflow Pipelines consists of multiple components. Before you begin, learn how to build the Kubeflow Pipelines component container images. To get started, see the development guides:
See the SDK-specific Contribution Guidelines for contributing to the kfp
SDK.
The frontend part of the project uses prettier for formatting, read frontend/README.md#code-style for more details.
Use gofmt package to format your .go source files. Read backend/README.md#code-style for more details.
- Testing via Public APIs
- Put your tests in a different package: Moving your test code out of the package allows you to write tests as though you were a real user of the package. You cannot fiddle around with the internals,
instead you focus on the exposed interface and are always thinking about any noise that you might be adding to your API. Usually the test code will be put under the same folder
but with a package suffix of
_test
. https://golang.org/src/go/ast/example_test.go (example) - Internal tests go in a different file: If you do need to unit test some internals, create another file with
_internal_test.go
as the suffix. - Write table driven tests: https://github.com/golang/go/wiki/TableDrivenTests (example)
All submissions, including submissions by project members, require review. We use GitHub pull requests for this purpose. Consult GitHub Help for more information on using pull requests.
We enforce a pull request (PR) title convention to quickly indicate the type and scope of a PR. PR titles become commit messages when PRs are merged. We also parse PR titles to generate the changelog.
PR titles should:
- Provide a user-friendly description of the change.
- Follow the Conventional Commits specification.
- Specifies issue(s) fixed, or worked on at the end of the title.
Examples:
fix(ui): fixes empty page. Fixes #1234
feat(backend): configurable service account. Fixes #1234, fixes #1235
chore: refactor some files
test: fix CI failure. Part of #1234
The following sections describe the details of the PR title convention.
PR titles should use the following structure.
<type>[optional scope]: <description>[ Fixes #<issue-number>]
Replace the following:
<type>
: The PR type describes the reason for the change, such asfix
to indicate that the PR fixes a bug. More information about PR types is available in the next section.[optional scope]
: (Optional.) The PR scope describes the part of Kubeflow Pipelines that this PR changes, such asfrontend
to indicate that the change affects the user interface. Choose a scope according to PR Scope section.<description>
: A user friendly description of this change.[ Fixes #<issues-number>]
: (Optional.) Specifies the issues fixed by this PR.
Type can be one of the following:
- feat: A new feature.
- fix: A bug fix. However, a PR that fixes test infrastructure is not user facing, so it should use the test type instead.
- docs: Documentation changes.
- chore: Anything else that does not need to be user facing.
- test: Adding or updating tests only. Please note, feat and fix PRs should have related tests too.
- refactor: A code change that neither fixes a bug nor adds a feature.
- perf: A code change that improves performance.
Note, only feature, fix and perf type PRs will be included in CHANGELOG, because they are user facing.
If you think the PR contains multiple types, you can choose the major one or split the PR to focused sub-PRs.
If you are not sure which type your PR is and it does not have user impact,
use chore
as the fallback.
Scope is optional, it can be one of the following:
- frontend: user interface or frontend server related, folder
frontend
,frontend/server
- backend: Backend, folder
backend
- sdk:
kfp
python package, foldersdk
- sdk/client:
kfp-server-api
python package, folderbackend/api/python_http_client
- components: Pipeline components, folder
components
- deployment: Kustomize or gcp marketplace manifests, folder
manifests
- metadata: Related to machine learning metadata (MLMD), folder
backend/metadata_writer
- cache: Caching, folder
backend/src/cache
- swf: Scheduled workflow, folder
backend/src/crd/controller/scheduledworkflow
- viewer: Tensorboard viewer, folder
backend/src/crd/controller/viewer
If you think the PR is related to multiple scopes, you can choose the major one or split the PR to focused sub-PRs. Note, splitting large PRs that affect multiple scopes can help make it easier to get your PR reviewed, since different scopes usually have different reviewers.
If you are not sure, or the PR doesn't fit into above scopes. You can either omit the scope because it's optional, or propose an additional scope here.
This project follows Google's Open Source Community Guidelines.