Signed-off-by: mingfu <mingfu@alauda.io>
11 KiB
Contributing to the catalog repo
Thank you for your interest in contributing!
This doc is about how to contribute to this repo specifically. For how to
contribute to tektoncd projects in general, see the overview in our README
and the individual CONTRIBUTING.md
files in each respective project.
All contributors must comply with the code of conduct.
PRs are welcome, and will follow the tektoncd pull request process.
How to Contribute a Task, StepAction or Pipeline
The Catalog repository is intended to serve as a location where users can find
Task
s, StepAction
s and Pipeline
s that are maintained, useful and follow established
best practices.
The process for contributing looks like this:
- Fork this repository, develop and test your
Task
s. - Create a new folder for your
Task
(s) - Ensure your Task
- Follows the guidelines
- Meets the technical requirements
- Includes OWNERS
- Submit a pull request.
How to Contribute a new version of a Task or Pipeline
If you are planning to add a new version of a Task or Pipeline make sure to separate your changes from the copied task. This makes it easy for reviewers to review the changes and not the actual copy.
For example if you have to bump the catalog task called foo
from 0.1
to
0.2
you simply first copy the old task :
% cp -a tasks/foo/0.1 tasks/foo/0.2
and then immediately commit that change :
% git add tasks/foo/0.2
% git commit -m "Copy task foo from 0.1 to 0.2
and then add your change and commit it.
This will result to a clean git log and makes it easier to only see your changes.
Guidelines
When reviewing PRs that add new Task
s, StepAction
s or Pipeline
s, maintainers will follow
the following guidelines:
- Submissions should be useful in real-world applications. While this repository is meant to be educational, its primary goal is to serve as a place users can find, share and discover useful components. This is not a samples repo to showcase Tekton features, this is a collection
- Submissions should follow established authoring recommendations
- Submissions should be well-documented.
- Coming Soon Submissions should be testable, and come with the required tests.
If you have an idea for a new submission, feel free to open an issue to discuss the idea with the catalog maintainers and community. Once you are ready to write your submission, please open a PR with the code, documentation and tests and a maintainer will review it.
Over time we hope to create a scalable ownership system where community members can be responsible for maintaining their own submissions, but we are not there yet.
Technical requirements
- Must pass the Task validation (aka
kubectl create -f task.yaml
should succeed) - Images should be published and maintained on an public image registry (gcr.io, docker.io, quay.io, …). A bonus if those images are auto-built.
- Images should not have any major security vulnerabilities
- Should follow Kubernetes best practices
- Provide as many default paramater values as possible
- Provide end to end tests
- (Nice to have) : provide versions with and without
PipelineResource
End to end Testing
There are two types of e2e tests launched on CI.
The first one would just apply the yaml files making sure they don't have any syntax issues. Pretty simple one, it just basically checks the syntax.
The second one would do some proper functional testing, making sure the task actually ran properly.
The way the functional tests works is that if you have a directory called
tests/
inside the task, CI will create a random Namespace
then apply
the task and then every yaml file in the tests/
directory.
Note that the test runner for the integration tests will only test the tasks
that have been added or modified in the submitted PR and will not run tests for
any tasks that haven't been changed unless the environment variable
TEST_RUN_ALL_TESTS
has been set.
Usually in these other yaml files you would have a yaml file for the
test resources (PipelineResource
) and a yaml files to run the tasks
(TaskRun or PipelineRun
).
Sometimes you may need to be able to run scripts before applying the tested task
or the other yaml files. For example, your tests may need pre-setup in the
Namespace
, external setup, or perhaps even manipulation of the main Task
.
For example on image builders tasks like kaniko
or jib
we want to upload
the tasks to a registry to make sure it is actually built properly. To do so, we
manipulate the Task
with a
python script (something we only want for the tests) to add a registry as a
Sidecar
and make sure that the TaskRun
sets the parameters to upload there.
This is simple and straightforward -- there is no need to upload to an external
image registry provider which would require settin up tokens and dealing with
both side effects and an external dependency.
There are two different scripts that are automatically applied if present. These
are applied using the source
bash script, so you can output environment
variables that will be applied:
- pre-apply-task-hook.sh: Script to run before applying the task
- pre-apply-taskrun-hook.sh: Script to run before applying the taskruns or other yaml files.
We have some helper functions you can use from your hook
scripts:
- add_sidecar_registry: This will add a registry as a sidecar to allow the builder tasks to upload an image directly to this sidecar registry instead of relying on an external registries.
- add_sidecar_secure_registry: This will run a secure registry as a sidecar
to allow the tasks to push to this registry using certs. It will create a
configmap
sslcert
with certificate available at keyca.crt
- add_task: Install a task into the testing namespace, the first argument is
the name of the task, the second argument is the version of the task. If the
version is equal to
latest
it will install the latest version of the task.
What can you run from those scripts is defined in the test-runner image. If you
need to have another binary available, make a PR to this Dockerfile
:
https://github.com/tektoncd/plumbing/blob/main/tekton/images/test-runner/Dockerfile
A helper script called run-test.sh
is provider in the
test directory to help the developer running the test locally. Just
specify the task name and the version as the first and the second argument i.e:
./test/run-test.sh task git-clone 0.1
and it will use your current kubernetes context to run the test and show you the outputs similar to the CI.
End to end Testing for external services
Some tasks need to be able to access some external REST api services.
There are two approaches for testing external services:
- Spin up a deployment of the service tests and expose a kubernetes service.
- Create an http rest api reflector for task that connects to a rest apis endpoint that cannot be available as a deployment (i.e: Saas services like github)
For the first approach, you can take the trigger-jenkins-build test as an example.
You will want to modify the
pre-apply-task-hook.sh
script to create the deployment and make it available to your test PipelineRun
.
Here is a rundown of the steps we are doing in trigger-jenkins-build/pre-apply-task-hook.sh
script :
- Create a deployment with the
jenkins
image - Wait until the deployment has completed.
- Expose the deployment as a service, which would then be easily available for other pods in the namespace.
- Do some shenanigans inside the jenkins pod so we can grab the jenkins apikey and create a new jenkins job.
- Create a secret with the apikey, username and other items.
The test pipelinerun for the
trigger-jenkins-build/
will then point to http://jenkins:8080
which is the
service URL where our just deployed jenkins is exposed. It uses the credentials
from the secret in the pre-apply-task-hook.sh
script.
For services where you can't spin up a new deployment of the service easily, the test runner supports the "Go Rest api test" project. The Go rest api test project is a simple service that replies back to http requests according to rules.
As an example see the github-add-comment task. For this task to be tested we need to be able to "fake" the Github REST api calls. To be able to do so, we are adding a go-rest-api-test rule inside the testing repository; the rule looks like this :
---
headers:
method: POST
path: /repos/{repo:[^/]+/[^/]+}/issues/{issue:[0-9]+}/comments
response:
status: 200
output: '{"status": 200}'
content-type: text/json
The rule is saying that for every POST requests going to this url :
/repos/${ORG}/${REPO}/issues/${issues}/comments
we will reply by a 200
status and output {"status": 200}
The Pipelinerun test for the
github-add-comment
task overrides the github host url in its param to point to
localhost:8080
:
- name: GITHUB_HOST_URL
value: http://localhost:8080
In the test runner if we find a directory called
task/${task}/${version}/tests/fixtures
we automatically spin up the
"go-rest-api-test" server as a
sidecar container with the test's fixtures yaml as the config. It will be then
available to the task locally to this URL http://localhost:8080
.
The task runs against that service instead of the github servcer and the responder replies with the right calls, we know then that the task has been properly tested.
The only requirement to use the fixtures testing facility is to enable the task to override the URL via a task parameter.
The go-rest-api-test
is a very simple service at the moment and may see other
improvements in the future to support more robust testing.
Owning and Maintaining a Task
Individual tasks should maintained by one or more users of GitHub. When someone maintains a Task, they have the access to merge changes to that Task. To have merge access to a Task, someone needs to:
- Be invited (and accept your invite) as a read-only collaborator on the tekton organization. If you need sponsors and have contributed to the chart, please reach out to the existing maintainers, or if you are having trouble connecting with them, please reach out to one of the main OWNERS of this repository.
- an
OWNERS
file needs to be added in theTask
folder. ThatOWNERS
file should list the maintainers' GitHub login names for both the reviewers and approvers sections.
OWNERS
The top-level OWNERS
file lists the Trusted
Collaborators. The process to becoming an
OWNER
is the same as other Tekton projects.