[EP] Enable Fabric Mem only if MC_USE_NVLINK_IPC is explicitly set to zero by UNIDY2002 · Pull Request #1637 · kvcache-ai/Mooncake

UNIDY2002 · 2026-03-10T03:16:08Z

Description

Module

Type of Change

How Has This Been Tested?

Checklist

I have performed a self-review of my own code.
I have formatted my own code using ./scripts/code_format.sh before submitting.
I have updated the documentation.
I have added tests to prove my changes are effective.

… zero

gemini-code-assist · 2026-03-10T03:16:17Z

Summary of Changes

Hello, 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 addresses a bug in the Mooncake EP module where the activation of Fabric Memory was incorrectly handled based on the presence of an environment variable. The change refines the condition for enabling Fabric Memory, ensuring it is active by default and only disabled when explicitly requested, thereby improving the intended behavior of memory management within the system.

Highlights

Fabric Memory Activation Logic: Updated the logic for enabling Fabric Memory (MNNVL) to only disable it when the MC_USE_NVLINK_IPC environment variable is explicitly set to '0'. Previously, Fabric Memory was disabled if the variable was set to any value, including an empty string or non-zero values.
Environment Variable Handling: Modified the supportFabricMem function to correctly parse the MC_USE_NVLINK_IPC environment variable, ensuring that strcmp is used to check for the specific value '0' rather than just its presence.

🧠 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

mooncake-ep/src/mooncake_ep_buffer.cpp
- Fixed the logic for supportFabricMem to correctly interpret the MC_USE_NVLINK_IPC environment variable, enabling Fabric Memory unless MC_USE_NVLINK_IPC is explicitly '0'.

Using Gemini Code Assist

The 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 /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

Feature	Command	Description
Code Review	`/gemini review`	Performs a code review for the current pull request in its current state.
Pull Request Summary	`/gemini summary`	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in pull request comments and review comments.
Help	`/gemini help`	Displays a list of available commands.

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 .gemini/ folder in the base of the repository. Detailed instructions can be found here.

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.

Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

UNIDY2002 · 2026-03-10T03:16:30Z

cc @he-yufeng

gemini-code-assist

Code Review

This pull request aims to change the condition for enabling fabric memory, making it active only when MC_USE_NVLINK_IPC is explicitly set to '0'. However, the implementation in mooncake-ep/src/mooncake_ep_buffer.cpp contains a significant logic error where the newly calculated fabric_enabled flag is never used. This results in fabric memory being always enabled if hardware supports it, effectively ignoring the user's intent to disable it via the environment variable, which is a security regression as it bypasses intended access control logic for a feature that grants broad memory access across devices. Furthermore, this change creates a discrepancy with a similar check in another module, which could lead to inconsistent behavior.

Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>

codecov-commenter · 2026-03-10T05:07:08Z

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

ympcMark

LGTM

UNIDY2002 · 2026-03-10T05:27:48Z

@ympcMark Next time, please ensure you click "squash and merge"

ympcMark · 2026-03-10T05:38:35Z

Oh, I will check it next time!

he-yufeng · 2026-03-10T05:46:06Z

Thanks for the fix and the cc! This makes sense — the original getenv() check I wrote in #1629 was too broad, it would disable fabric mem even when the env var was set to a non-zero value. The explicit strcmp(nvlink_ipc, "0") check is cleaner.

One thing I noticed: the Gemini review flagged that fabric_enabled is true when MC_USE_NVLINK_IPC=0, which then enables fabric mem. The variable name reads a bit counterintuitive since MC_USE_NVLINK_IPC=0 means "don't use NVLink IPC" → "use fabric mem instead". Might be worth a comment clarifying the semantics, but not a blocker — the logic itself is correct.

UNIDY2002 · 2026-03-10T05:50:18Z

One thing I noticed: the Gemini review flagged that fabric_enabled is true when MC_USE_NVLINK_IPC=0, which then enables fabric mem. The variable name reads a bit counterintuitive since MC_USE_NVLINK_IPC=0 means "don't use NVLink IPC" → "use fabric mem instead". Might be worth a comment clarifying the semantics, but not a blocker — the logic itself is correct.

Yes, I agree with you. I was just mirroring the logic here:

Mooncake/mooncake-transfer-engine/src/transport/hip_transport/hip_transport.cpp

Lines 329 to 337 in 845ccc4

    
           const char *hip_ipc = getenv("MC_USE_HIP_IPC"); 
        
           const char *nvlink_ipc = getenv("MC_USE_NVLINK_IPC"); 
        
           bool fabric_enabled = (hip_ipc && strcmp(hip_ipc, "0") == 0) || 
        
                                 (nvlink_ipc && strcmp(nvlink_ipc, "0") == 0); 
        
           if (!fabric_enabled) { 
        
               return false; 
        
           }

Perhaps we have to clarify the semantics of these env vars in the future.

stmatengss · 2026-03-11T06:35:52Z

LGTM

@ympcMark Please use squash merge to consolidate duplicate commits.

[EP] Enable Fabric Mem only if MC_USE_NVLINK_IPC is explicitly set to…

ea59594

… zero

UNIDY2002 requested a review from ympcMark as a code owner March 10, 2026 03:16

github-actions Bot added run-ci Mooncake EP labels Mar 10, 2026

gemini-code-assist Bot reviewed Mar 10, 2026

View reviewed changes

Comment thread mooncake-ep/src/mooncake_ep_buffer.cpp Outdated

UNIDY2002 and others added 2 commits March 10, 2026 11:20

Update mooncake-ep/src/mooncake_ep_buffer.cpp

9349356

Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>

Format?

5a47ce0

ympcMark approved these changes Mar 10, 2026

View reviewed changes

ympcMark merged commit 9968d05 into main Mar 10, 2026
28 of 30 checks passed

UNIDY2002 deleted the sunxun/fix branch March 10, 2026 05:26

UNIDY2002 added a commit that referenced this pull request Mar 13, 2026

chore: align MC_USE_NVLINK_IPC check with PR #1637

12ee350

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[EP] Enable Fabric Mem only if MC_USE_NVLINK_IPC is explicitly set to zero#1637

[EP] Enable Fabric Mem only if MC_USE_NVLINK_IPC is explicitly set to zero#1637
ympcMark merged 3 commits intomainfrom
sunxun/fix

UNIDY2002 commented Mar 10, 2026

Uh oh!

gemini-code-assist Bot commented Mar 10, 2026

Uh oh!

UNIDY2002 commented Mar 10, 2026

Uh oh!

gemini-code-assist Bot left a comment

Uh oh!

Uh oh!

codecov-commenter commented Mar 10, 2026

Uh oh!

ympcMark left a comment

Uh oh!

Uh oh!

UNIDY2002 commented Mar 10, 2026

Uh oh!

ympcMark commented Mar 10, 2026

Uh oh!

he-yufeng commented Mar 10, 2026

Uh oh!

UNIDY2002 commented Mar 10, 2026

Uh oh!

stmatengss commented Mar 11, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

Conversation

UNIDY2002 commented Mar 10, 2026

Description

Module

Type of Change

How Has This Been Tested?

Checklist

Uh oh!

gemini-code-assist Bot commented Mar 10, 2026

Summary of Changes

Highlights

Footnotes

Uh oh!

UNIDY2002 commented Mar 10, 2026

Uh oh!

gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

Uh oh!

codecov-commenter commented Mar 10, 2026

Codecov Report

Uh oh!

ympcMark left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

UNIDY2002 commented Mar 10, 2026

Uh oh!

ympcMark commented Mar 10, 2026

Uh oh!

he-yufeng commented Mar 10, 2026

Uh oh!

UNIDY2002 commented Mar 10, 2026

Uh oh!

stmatengss commented Mar 11, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants