Skip to content

Support symmetric memory pre-allocation to avoid fragmentation#17089

Merged
Fridge003 merged 3 commits intosgl-project:mainfrom
nvcastet:prealloc_symm_mem
Jan 23, 2026
Merged

Support symmetric memory pre-allocation to avoid fragmentation#17089
Fridge003 merged 3 commits intosgl-project:mainfrom
nvcastet:prealloc_symm_mem

Conversation

@nvcastet
Copy link
Copy Markdown
Collaborator

Motivation

Related to #17038
PyTorch mempools never de-fragment memory in OOM scenarios, so we need to pre-allocate a large chunk of memory to limit fragmentation in symmetric memory pool.

Add SGLANG_SYMM_MEM_PREALLOC_GB_SIZE to specify the prealloc size in GiB.
Solved issues found in #16327 and #13863

CC @merrymercy @Fridge003

Modifications

Accuracy Tests

Benchmarking and Profiling

Checklist

Review Process

  1. Ping Merge Oncalls to start the PR flow. See the PR Merge Process.
  2. Get approvals from CODEOWNERS and other reviewers.
  3. Trigger CI tests with comments or contact authorized users to do so.
    • /tag-run-ci-label, /rerun-failed-ci, /tag-and-rerun-ci
  4. After green CI and required approvals, ask Merge Oncalls to merge.

@nvcastet
Copy link
Copy Markdown
Collaborator Author

/tag-and-rerun-ci

@gemini-code-assist
Copy link
Copy Markdown
Contributor

Summary of Changes

Hello @nvcastet, 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 enhances memory management within the system by introducing a mechanism to pre-allocate symmetric memory. The primary goal is to combat memory fragmentation, a common issue in PyTorch's memory pools that can lead to out-of-memory errors even when sufficient total memory is available. By allowing a configurable pre-allocation size, the system can proactively reserve memory, thereby improving overall stability and performance, especially in scenarios involving distributed operations.

Highlights

  • Configurable Symmetric Memory Pre-allocation: Introduced a new environment variable, SGLANG_SYMM_MEM_PREALLOC_GB_SIZE, allowing users to specify the amount of memory (in GiB) to pre-allocate for symmetric memory pools, mitigating fragmentation issues.
  • Memory Fragmentation Mitigation: Implemented a prealloc_symmetric_memory_pool method in ModelRunner that, when enabled, pre-allocates a large chunk of memory at initialization. This addresses PyTorch's mempool behavior of not de-fragmenting memory in OOM scenarios, improving stability.
  • Refactored Forward Stream Initialization: The forward_stream is now initialized earlier within the ModelRunner's __init__ method and explicitly passed to other components like init_model_worker and get_worker_info. This ensures the stream is available for critical memory operations, such as the new pre-allocation logic.

🧠 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 introduces support for symmetric memory pre-allocation to mitigate memory fragmentation in PyTorch memory pools. A new environment variable SGLANG_SYMM_MEM_PREALLOC_GB_SIZE is added for this purpose. The changes involve refactoring the forward_stream initialization and adding the pre-allocation logic in ModelRunner.

My review identifies a critical issue in the pre-allocation logic where the memory size is incorrectly calculated, potentially leading to overallocation and Out-Of-Memory errors. I've provided a code suggestion to fix this.

Comment thread python/sglang/srt/model_executor/model_runner.py
Comment thread python/sglang/srt/environ.py
Comment thread python/sglang/srt/managers/scheduler.py
@Fridge003
Copy link
Copy Markdown
Collaborator

Also can we add a B200 test for symmetric memory. Can be a subtest of this one
https://github.com/sgl-project/sglang/blob/main/test/srt/test_deepseek_v3_fp4_4gpu.py

@github-actions github-actions Bot added documentation Improvements or additions to documentation deepseek labels Jan 20, 2026
Comment thread python/sglang/srt/managers/scheduler.py
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

deepseek documentation Improvements or additions to documentation run-ci

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants