-
Notifications
You must be signed in to change notification settings - Fork 869
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Kustomize package for mxnet operator #279
Conversation
I might suggest splitting the package into 2
The reason is this way if a user doesn't have cluster admin priveleges they can just ask a cluster admin to do the first part |
Yeah, that's very reasonable. I will make the change. |
/assign @jlewi |
@Jeffwan let me know when this is ready for another look |
@jlewi Not sure if we need to put service-account.yaml to the same package as cluster roles and crds. If end user wants to deploy operator under user's namespace, they also need to have service account at the same namespace, and clusterRoleBinding need to know namespace of the service account. I fee it's kind of hard to split into two packages. If another userB wants to use operator, should this user deploy another operator? Trying to understand the best practice here. CRD is totally separate and it can be moved to another package. If cluster admin will be responsible for both CRD and cluster role and bindings, it's not that meaningful to just have crd in separate package. Any thoughts? |
good point raised by @jlewi . However, we haven't followed the same for other operators. https://github.com/kubeflow/manifests/tree/master/pytorch-job -- crds are separate |
@johnugeorge is correct that we haven't consistently followed the pattern of splitting cluster level resources into separate applications. The request to split it into packages 1 for cluster scoped resources and 1 for non cluster scoped resources comes from some users who have expressed that request given the way permissions are structured across different teams. If you don't want to split it into I can live with that. |
@jlewi I think split into two is a good idea, meet some issues in this way. I am not sure where should I put the service account. SA is reference by both cluster-role-binding (cluster-scoped) and operator (non cluster scoped). Operator was designed as a cluster scoped controller which listen job requests from all namespaces.. |
I don't have a good answer so my suggestion is pick which ever you think best and we can change it later if that doesn't work. |
Let's merge it and if anyone have requests on split, we have a refactor for all training operators and make structure unified. |
/lgtm |
[APPROVALNOTIFIER] This PR is APPROVED This pull-request has been approved by: jlewi The full list of commands accepted by this bot can be found here. The pull request process is described here
Needs approval from an approver in each of these files:
Approvers can indicate their approval by writing |
…Tekton (kubeflow#279) * casestudy-pipeline * casestudy-pipeline
Which issue is resolved by this Pull Request:
Resolves #228
Description of your changes:
Generate packages using ksonnet and migrate to kustomize .
Checklist:
cd manifests/tests
make generate
make test
e2e test works as expected
This change is