Skip to content

Mooncake PG buffer#1401

Merged
UNIDY2002 merged 7 commits intomainfrom
mapc-bf
Jan 19, 2026
Merged

Mooncake PG buffer#1401
UNIDY2002 merged 7 commits intomainfrom
mapc-bf

Conversation

@ympcMark
Copy link
Copy Markdown
Collaborator

@ympcMark ympcMark commented Jan 19, 2026

Description

Part of #1225.

Avoid indexing SegmentDesc::buffers to obtain peer memory locations; transfer them through Torch's rendezvous store instead

Type of Change

  • Types
    • Bug fix
    • New feature
      • Transfer Engine
      • Mooncake Store
      • Mooncake EP
      • Integration
      • P2P Store
      • Python Wheel
    • Breaking change
    • CI/CD
    • Documentation update
    • Other

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.

@ympcMark ympcMark requested a review from UNIDY2002 as a code owner January 19, 2026 07:35
@gemini-code-assist
Copy link
Copy Markdown
Contributor

Summary of Changes

Hello @ympcMark, 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 refactors the Mooncake backend's buffer management strategy by moving away from implicitly indexing into SegmentDesc::buffers to obtain peer memory locations. Instead, it introduces a dedicated SegmentInfo struct that explicitly holds the addresses of various communication buffers. This struct is then serialized and exchanged between ranks using Torch's rendezvous store, providing a more direct and explicit mechanism for accessing remote buffer addresses. This change aims to streamline buffer management and enhance clarity in how memory regions are shared and accessed across distributed processes.

Highlights

  • New SegmentInfo Struct: Introduced a new SegmentInfo struct to explicitly store memory addresses for send, receive, synchronization, and warmup buffers, providing a clearer representation of memory locations.
  • Refactored Buffer Management: Replaced the use of std::shared_ptr<TransferMetadata::SegmentDesc> with the new SegmentInfo struct within TransferGroupMeta, allowing direct storage and access of buffer information for each rank.
  • Torch Rendezvous Store Integration: Implemented the transfer of SegmentInfo data between ranks via Torch's rendezvous store, eliminating the need for direct indexing into SegmentDesc::buffers for peer memory locations.
  • Updated Buffer Access: Modified all relevant buffer access points across mooncake_backend.cpp, mooncake_worker.cu, and mooncake_worker_thread.cpp to utilize the new SegmentInfo struct for retrieving memory addresses, simplifying the code and making memory access more explicit.

🧠 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.

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.

Footnotes

  1. 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.

Copy link
Copy Markdown
Contributor

@gemini-code-assist gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Code Review

This pull request refactors the mechanism for sharing peer memory locations by introducing a SegmentInfo struct and using Torch's rendezvous store for communication, instead of relying on indexed access into SegmentDesc::buffers. This is a good change that improves code clarity and maintainability by replacing magic indices with named fields. The implementation looks solid, but I have identified one area for improvement regarding data validation to enhance robustness.

Comment thread mooncake-pg/src/mooncake_backend.cpp Outdated
meta_.segmentDescs[pollingRank] = segment_desc;
std::string buffer_key = "buffer_" + std::to_string(backendIndex_) + "_" + std::to_string(pollingRank);
auto buffer_data = store->get(buffer_key);
memcpy(&meta_.segmentInfos[pollingRank], buffer_data.data(), sizeof(SegmentInfo));
Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

security-high high

The code copies data using memcpy without validating the size of the source buffer buffer_data. This is unsafe as it can lead to memory corruption if the store returns data of an unexpected size (e.g., due to version mismatches or data corruption). Please add a size check before this operation for robustness and security.

            TORCH_CHECK(buffer_data.size() == sizeof(SegmentInfo), "Received SegmentInfo of unexpected size from store for rank ", pollingRank);
            memcpy(&meta_.segmentInfos[pollingRank], buffer_data.data(), sizeof(SegmentInfo));

@codecov-commenter
Copy link
Copy Markdown

⚠️ Please install the 'codecov app svg image' 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!

@UNIDY2002
Copy link
Copy Markdown
Collaborator

E2E test have passed. Well done!

@UNIDY2002 UNIDY2002 merged commit e43dce6 into main Jan 19, 2026
3 of 4 checks passed
@UNIDY2002 UNIDY2002 deleted the mapc-bf branch January 22, 2026 03:08
UNIDY2002 added a commit that referenced this pull request Jan 27, 2026
XucSh pushed a commit to XucSh/Mooncake that referenced this pull request Jan 27, 2026
UNIDY2002 added a commit that referenced this pull request Jan 27, 2026
JasonZhang517 pushed a commit to JasonZhang517/Mooncake that referenced this pull request Feb 9, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants