feature(rbg): support gang scheduling in v1alpha2#195
feature(rbg): support gang scheduling in v1alpha2#195Syspretor wants to merge 3 commits intosgl-project:mainfrom
Conversation
Summary of ChangesHello, I'm Gemini Code Assist1! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed! This pull request introduces a significant feature by adding support for gang scheduling to RoleBasedGroups (RBGs) in the v1alpha2 API. It provides a flexible framework to integrate with different gang scheduling solutions like Kubernetes scheduler-plugins and Volcano. The changes involve refactoring the core logic for managing PodGroups, introducing new configuration options via annotations, and enhancing the reconciliation process to ensure proper gang scheduling behavior and consistency across various workload types. This allows RBGs to better manage co-scheduled workloads, improving resource utilization and job completion rates for distributed applications. Highlights
🧠 New Feature in Public Preview: You can now enable Memory to help Gemini Code Assist learn from your team's feedback. This makes future code reviews more consistent and personalized to your project's style. Click here to enable Memory in your admin console. Changelog
Activity
Using Gemini Code AssistThe full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips. Invoking Gemini You can request assistance from Gemini at any point by creating a comment using either
Customization To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a Limitations & Feedback Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for GitHub and other Google products, sign up here. You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension. Footnotes
|
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request introduces support for gang scheduling in v1alpha2 by adding a PodGroupManager abstraction for different scheduler plugins like scheduler-plugins and volcano. The changes are well-structured, adding a new --scheduler-name flag and propagating the configuration through the controllers. The implementation correctly handles the PodGroup lifecycle and injects necessary labels/annotations. The changes also extend to RoleInstance controllers to enforce gang scheduling semantics. Overall, this is a solid implementation, but I've found one issue where PodGroup updates for scheduler-plugins do not consider changes to ScheduleTimeoutSeconds, and I've provided a suggestion to fix it.
| if podGroup.Spec.MinMember != int32(rbg.GetGroupSize()) { | ||
| updateErr := retry.RetryOnConflict(retry.DefaultRetry, func() error { | ||
| if fetchErr := m.client.Get( | ||
| ctx, types.NamespacedName{Name: rbg.Name, Namespace: rbg.Namespace}, podGroup, | ||
| ); fetchErr != nil { | ||
| return fetchErr | ||
| } | ||
| if !utils.CheckOwnerReference(podGroup.OwnerReferences, gvk) { | ||
| podGroup.OwnerReferences = append(podGroup.OwnerReferences, *metav1.NewControllerRef(rbg, gvk)) | ||
| } | ||
| podGroup.Spec.MinMember = int32(rbg.GetGroupSize()) | ||
| return m.client.Update(ctx, podGroup) | ||
| }) | ||
| if updateErr != nil { | ||
| logger.Error(updateErr, "update pod group error") | ||
| return updateErr | ||
| } | ||
| } |
There was a problem hiding this comment.
The PodGroup is only updated if MinMember changes. If a user updates the group-gang-scheduling-timeout annotation, the ScheduleTimeoutSeconds field in the PodGroup custom resource will not be updated. The condition should also check for changes in ScheduleTimeoutSeconds.
| if podGroup.Spec.MinMember != int32(rbg.GetGroupSize()) { | |
| updateErr := retry.RetryOnConflict(retry.DefaultRetry, func() error { | |
| if fetchErr := m.client.Get( | |
| ctx, types.NamespacedName{Name: rbg.Name, Namespace: rbg.Namespace}, podGroup, | |
| ); fetchErr != nil { | |
| return fetchErr | |
| } | |
| if !utils.CheckOwnerReference(podGroup.OwnerReferences, gvk) { | |
| podGroup.OwnerReferences = append(podGroup.OwnerReferences, *metav1.NewControllerRef(rbg, gvk)) | |
| } | |
| podGroup.Spec.MinMember = int32(rbg.GetGroupSize()) | |
| return m.client.Update(ctx, podGroup) | |
| }) | |
| if updateErr != nil { | |
| logger.Error(updateErr, "update pod group error") | |
| return updateErr | |
| } | |
| } | |
| timeout := getScheduleTimeoutSeconds(rbg) | |
| if podGroup.Spec.MinMember != int32(rbg.GetGroupSize()) || podGroup.Spec.ScheduleTimeoutSeconds == nil || *podGroup.Spec.ScheduleTimeoutSeconds != *timeout { | |
| updateErr := retry.RetryOnConflict(retry.DefaultRetry, func() error { | |
| if fetchErr := m.client.Get( | |
| ctx, types.NamespacedName{Name: rbg.Name, Namespace: rbg.Namespace}, podGroup, | |
| ); fetchErr != nil { | |
| return fetchErr | |
| } | |
| if !utils.CheckOwnerReference(podGroup.OwnerReferences, gvk) { | |
| podGroup.OwnerReferences = append(podGroup.OwnerReferences, *metav1.NewControllerRef(rbg, gvk)) | |
| } | |
| podGroup.Spec.MinMember = int32(rbg.GetGroupSize()) | |
| podGroup.Spec.ScheduleTimeoutSeconds = timeout | |
| return m.client.Update(ctx, podGroup) | |
| }) | |
| if updateErr != nil { | |
| logger.Error(updateErr, "update pod group error") | |
| return updateErr | |
| } | |
| } |
| return nil | ||
| } | ||
|
|
||
| if podGroup.Spec.MinMember != int32(rbg.GetGroupSize()) { |
There was a problem hiding this comment.
if the user changes the rbg.workloads.x-k8s.io/group-gang-scheduling-timeout annotation,
the existing PodGroup will never be updated to reflect the new timeout value.
Is it by design？
|
Pls add comment/docs clarification around the gang-scheduling annotation semantics. |
2 participants
Ⅰ. Motivation
Ⅱ. Modifications
Ⅲ. Does this pull request fix one issue?
fixes #XXXX
Ⅳ. List the added test cases (unit test/integration test) if any, please explain if no tests are needed.
Ⅴ. Describe how to verify it
VI. Special notes for reviews
Checklist
make fmt.