Pipelines WG and Kubeflow 1.6 release

[annajung]

Hello @kubeflow/wg-pipeline-leads @zijianjoy

The 1.6 release team will use this tracking issue to coordinate the integration of the Pipelines with the Kubeflow 1.6 release.

Heads up that the feature freeze is scheduled for Wednesday, Jun 1st, 2022. For the full release schedule, refer to the 1.6 release timeline.

(note: links are pending kubeflow/community#558 and kubeflow/community#560)

To prepare for the 1.6 release, please provide answers to the following questions:

1. What version of Pipelines would you like to include for the 1.6 release?
2. Could you provide a branch/tag for this version? (note: It doesn't have to be final)
3. Can you share a link to all open / work in progress issues that you're expecting to include in the 1.6?
4. If you're using the following dependencies, could you provide the versions you'll be supporting for each dependency?

• Knative
• Istio
• Kubernetes
• Cert-Manager
• Kustomize

The release team WG liaison for Pipelines WG will be @DomFleischmann and you can keep track of high-level release updates in the 1.6 tracking issue.

[annajung]

Due to a blocker that was identified last week, we are delaying the release for 2 weeks, with the possibility of another extension. The new feature freeze date is Wednesday, Jun 15th, 2022. If another extension is needed, another announcement will be made to notify the community.

• Tracking issue: Alternative solution to removal of test on optional-test-infra testing#1006
• Email announcement: https://groups.google.com/u/2/g/kubeflow-discuss
• New schedule PR: Update release timeline community#561

[zijianjoy]

Thank you @annajung for the update! If we cannot find testing support, does that mean we need to further delay the release?

[annajung]

Hi @zijianjoy, from today's release team meeting, it looks like all other WG leads have decided to migrate to GitHub Actions as a short-term solution and will be able to meet the new feature freeze deadline of June 15th.

I need to confirm with Training Operators / AutoML WG as they were the only ones not present in the meeting. You can find a detailed update of the meeting here

[annajung]

After checking with all WGs, we are delaying the release for 2 more weeks. The new feature freeze date is Wednesday, Jun 29th, 2022.

Email announcement: https://groups.google.com/g/kubeflow-discuss/c/I4l97HvrGEA/m/227aCe_mCgAJ
New schedule PR: kubeflow/community#562

The release team is tracking Pipeline 2.0.0-alpha for the 1.6 release. If there are changes to the branch/tag, please let us know before the new feature freeze deadline.

[zijianjoy]

Thank you @annajung , what criteria do we use to determine whether to extend release schedule again on Jun 29th?

[annajung]

Hi @zijianjoy, great question! I actually had the same question when evaluating the second extension since the requirements for release extension are not defined in the release handbook.

I reached out to the previous release manager, @kimwnasptd for clarification and to learn about how extension requests were evaluated in the past releases. Based on our conversation, a release extension has been granted based on a discussion between WGs and the release manager.

As for this specific case, Katib, Training Operator, and Notebook WGs needed more time and I thought it was better to be flexible due to the unpredicted work that each WG had to do with the testing infrastructure issue. In addition, the first extension was made without clarity on how long each WG would need to resolve the testing infrastructure blocker and the first extension announcement did include a note that there is a possibility of another extension until a solution is found for all affected WGs.

I hope that answers your question! If you have any feedback on the release extension process and requirements, let me know here or in the release retro doc. This is a great topic to discuss during the release retro and define the requirements we want to set for the future release extension.

[zijianjoy]

@annajung Can we make Notebook supporting Kubernetes 1.22 as the criteria for another extension? It is rather urgent that Kubeflow is able to support Kubernetes 1.22 than introducing new features. We can include the same individual kubeflow component versions as Kubeflow 1.5 except Notebook. With this isolation, it gives Kubeflow components more time to figure out the testing infrastructure solution, while unblocking users to continue using Kubeflow. We can always include all new cool features in Kubeflow 1.7.

[kimwnasptd]

Hey @zijianjoy, we are almost there for K8s 1.22 in the current master branch kubeflow/kubeflow#6353.

The only missing part is the Profile Controller, but we have an inflight PR that we are reviewing kubeflow/kubeflow#6491. This will be merged either right before the feature freeze, or immediately within the next week. As you mentioned we do really want to ensure we support 1.22.

So for a delay, I'm confident we should be good regarding Notebooks and K8s 1.22. But let's discuss if it's something else

[zijianjoy]

Thank you @kimwnasptd for the information! Sounds good on waiting Notebook to be ready.

[ryansteakley]

@zijianjoy @kimwnasptd Please take a look at this short pr kubeflow/pipelines#7946 which simply updates the minio-go versioning.

[annajung]

xposting

Hi @kubeflow/wg-pipeline-leads @zijianjoy @james-jwu pipeline leads, after various attempts to clarify the version for KF 1.6, the release team hasn't received a response yet on what tag to use for 1.6 since previously mentioned tag v2.0.0 is not available.

With the feature freeze being yesterday, the release team will go ahead with the latest stable version available for pipelines as of June 30th, which is 1.8.2.

Please reach out to us if you have any concerns! thank you!

[zijianjoy]

Hello @annajung , we have communicated that KFP is going to use 2.0.0-alpha.2 for the upcoming Kubeflow 1.6 release. Looks like you are not able to find the tag, so here it is: https://github.com/kubeflow/pipelines/releases/tag/2.0.0-alpha.2

[juliusvonkohout]

@zijianjoy there is a severe regression in 1.6 regarding pipelines kubeflow/pipelines#8256

[annajung]

With KF 1.6 released, closing the issue - 1.6.1 related work is tracked at #2194