Closed Jeffwan closed 3 years ago
/hold
/cc @Thor-wl @shinytang6 for extra review due to blocked PR. https://github.com/kubeflow/common/pull/123 /cc @kubeflow/wg-training-leads
@Jeffwan: GitHub didn't allow me to request PR reviews from the following users: Thor-wl, shinytang6.
Note that only kubeflow members and repo collaborators can review this PR, and authors cannot review their own PRs.
I think the main purpose of https://github.com/kubeflow/common/pull/123 is to avoid the direct dependency to k8s. Then I am wondering if we should upgrade the version to 1.19.
Do you want to add context first, then we can eliminate the k8s import packages?
I'm both OK for avoiding the direct depend dependency to k8s first or upgrade to k8s to v1.19 first. You can make a decision @Jeffwan @gaocegege
I think the main purpose of #123 is to avoid the direct dependency to k8s. Then I am wondering if we should upgrade the version to 1.19.
Do you want to add context first, then we can eliminate the k8s import packages?
@gaocegege I think your concern is reasonable. community is releasing 1.21. It's better to keep < 3 version changes with latest version and we will update to higher version sometime soon.
1.18 brings in context changes, I am ok to bump to 1.18. The major reason I choose 1.19 is because volcano/api now is using 1.19 https://github.com/volcano-sh/apis/blob/cfacea273e7da6b53170cc00ece9b2caf019316f/go.mod#L6-L10
Another possible solution is to ask volcano/api to downgrade to version >=1.16 <=1.17 and we don't need client-go related changes. But we will still need to upgrade dependencies later.
I think the main purpose of #123 is to avoid the direct dependency to k8s. Then I am wondering if we should upgrade the version to 1.19. Do you want to add context first, then we can eliminate the k8s import packages?
@gaocegege I think your concern is reasonable. community is releasing 1.21. It's better to keep < 3 version changes with latest version and we will update to higher version sometime soon.
1.18 brings in context changes, I am ok to bump to 1.18. The major reason I choose 1.19 is because volcano/api now is using 1.19 https://github.com/volcano-sh/apis/blob/cfacea273e7da6b53170cc00ece9b2caf019316f/go.mod#L6-L10
Another possible solution is to ask volcano/api to downgrade to 1.17 and we don't need client-go related changes. But we will still need to upgrade dependencies later.
It's OK for me to provide volcano.sh/apis release-1.17 to match k8s.io v1.17. Look forward to your decision.
volcano.sh/apis v1.2.0-k8s1.16.15
this makes sense. I think @gaocegege will like it. :D
In this case, we will not bump k8s version at this moment, this PR can be left for future usage.
SGTM! I think we will update to a higher version soon.
volcano.sh/apis v1.2.0-k8s1.16.15
this makes sense. I think @gaocegege will like it. :D In this case, we will not bump k8s version at this moment, this PR can be left for future usage.SGTM! I think we will update to a higher version soon.
@Jeffwan @gaocegege So what's the version shall i provide?
/cc @kubeflow/wg-training-leads /hold cancel
This unblocks user's request @chxk in https://github.com/kubeflow/common/issues/132
[APPROVALNOTIFIER] This PR is APPROVED
This pull-request has been approved by: andreyvelich, Jeffwan, terrytangyuan
The full list of commands accepted by this bot can be found here.
The pull request process is described here
Signed-off-by: Jiaxin Shan seedjeffwan@gmail.com