feat(binder): GPU_MEMORY_LIMIT binder plugin for shared GPU workloads

[FouoF]

Description

As the issue described.

Related Issues

Fixes #1367

Checklist

Note: Ensure your PR title follows the Conventional Commits format (e.g., feat(scheduler): add new feature)

• Self-reviewed
• Added/updated tests (if needed)
• Updated documentation (if needed)

Breaking Changes

No

Additional Notes

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: c271df49-fb9e-471c-9bb9-ea776822d8b7

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

Tip

💬 Introducing Slack Agent: The best way for teams to turn conversations into code.

Slack Agent is built on CodeRabbit's deep understanding of your code, so your team can collaborate across the entire SDLC without losing context.

• Generate code and open pull requests
• Plan features and break down work
• Investigate incidents and troubleshoot customer tickets together
• Automate recurring tasks and respond to alerts with triggers
• Summarize progress and report instantly

Built for teams:

• Shared memory across your entire org—no repeating context
• Per-thread sandboxes to safely plan and execute work
• Governance built-in—scoped access, auditability, and budget controls

One agent for your entire SDLC. Right inside Slack.

👉 Get started

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[enoodle]

@FouoF Can you please rebase this on top of origin/main removing the commits already merged?

[enoodle]

@FouoF Can you please implement it as a new plugin to the binder that will be "opt-in" - ie not enabled by default?
Currently it is added as part of the gpu sharing plugin, but I would like to keep all the GPU virtualization separate to avoid a mess in the future when more are implemented.

[github-actions]

Merging this branch will decrease overall coverage

Impacted Packages	Coverage Δ	🤖
github.com/kai-scheduler/KAI-scheduler/pkg/apis/kai/v1/binder	32.73% (ø)
github.com/kai-scheduler/KAI-scheduler/pkg/binder/binding	72.92% (ø)
github.com/kai-scheduler/KAI-scheduler/pkg/binder/common	13.27% (-1.68%)	👎
github.com/kai-scheduler/KAI-scheduler/pkg/binder/plugins	64.89% (-0.40%)	👎
github.com/kai-scheduler/KAI-scheduler/pkg/binder/plugins/gpusharing	30.23% (-13.10%)	💀
github.com/kai-scheduler/KAI-scheduler/pkg/operator/operands/binder	74.03% (ø)

---

Coverage by file

Changed files (no unit tests)

Changed File	Coverage Δ	Total	Covered	Missed	🤖
github.com/kai-scheduler/KAI-scheduler/pkg/apis/kai/v1/binder/binder.go	94.74% (ø)	57	54	3
github.com/kai-scheduler/KAI-scheduler/pkg/binder/common/constants.go	0.00% (ø)	0	0	0
github.com/kai-scheduler/KAI-scheduler/pkg/binder/common/gpu_access.go	0.00% (ø)	76 (+11)	0	76 (+11)
github.com/kai-scheduler/KAI-scheduler/pkg/binder/plugins/config.go	98.04% (ø)	51	50	1
github.com/kai-scheduler/KAI-scheduler/pkg/binder/plugins/factory.go	53.23% (+1.30%)	62 (+10)	33 (+6)	29 (+4)	👍
github.com/kai-scheduler/KAI-scheduler/pkg/binder/plugins/gpusharing/gpu_sharing.go	30.23% (-13.10%)	86 (+26)	26	60 (+26)	💀

Please note that the "Total", "Covered", and "Missed" counts above refer to code statements instead of lines of code. The value in brackets refers to the test coverage of that file in the old version of the code.

Changed unit test files

• github.com/kai-scheduler/KAI-scheduler/pkg/apis/kai/v1/binder/binder_test.go
• github.com/kai-scheduler/KAI-scheduler/pkg/binder/binding/default_binder_test.go
• github.com/kai-scheduler/KAI-scheduler/pkg/binder/binding/fraction_binder_test.go
• github.com/kai-scheduler/KAI-scheduler/pkg/binder/plugins/config_test.go
• github.com/kai-scheduler/KAI-scheduler/pkg/binder/plugins/gpusharing/gpu_sharing_test.go
• github.com/kai-scheduler/KAI-scheduler/pkg/operator/operands/binder/binder_test.go

[FouoF]

@FouoF Can you please implement it as a new plugin to the binder that will be "opt-in" - ie not enabled by default? Currently it is added as part of the gpu sharing plugin, but I would like to keep all the GPU virtualization separate to avoid a mess in the future when more are implemented.

Thanks, I agree with the goal of keeping GPU virtualization separate from the GPU sharing plugin.

After looking deeper, I don't think making this a binder-only standalone plugin is actually independent in the current architecture. The virtualization logic still relies on several pieces owned by GPU sharing: the admission mutation that injects the env/configmap references, the runai/shared-gpu-configmap annotation, the GPU sharing ConfigMap lifecycle, and the binder plugin ordering where GPU sharing must create the ConfigMap before virtualization can update it.

So an opt-in binder plugin would currently be a hidden extension of gpusharing, not a truly separate virtualization plugin. That could be misleading and still leave us with tight coupling.

I think there are two reasonable options:

1. Keep this behavior explicitly under GPU sharing for now.
2. Do a larger follow-up refactor that separates the admission mutation, ConfigMap/env management, and binder logic for GPU virtualization as its own opt-in path.

Given that, I’d prefer not to present this as an independent binder plugin in this PR unless we also split the admission/configmap pieces.

[enoodle]

@FouoF Lets present it as a dependent binder plugin then - it will only function well if the gpusharing plugin is also enabled.

[FouoF]

@enoodle I have made it an independent plugin and it will check if the gpusharing plugin is enabled.