[Bugfix] Fix ChatCompletion endpoint of mini_lb when stream is set

[ShangmingCai]

Motivation

Another bugfix for ChatCompletion endpoint, which addresses the bug reported in https://sgl-fru7574.slack.com/archives/C08QGMU93GX/p1748418994376989 .

CC: @zhyncs @hnyls2002

Checklist

• Format your code according to the Code Formatting with Pre-Commit.
• Add unit tests as outlined in the Running Unit Tests.
• Update documentation / docstrings / example tutorials as needed, according to Writing Documentation.
• Provide throughput / latency benchmark results and accuracy evaluation results as needed, according to Benchmark and Profiling and Accuracy Results.
• For reviewers: If you haven't made any contributions to this PR and are only assisting with merging the main branch, please remove yourself as a co-author when merging the PR.
• Please feel free to join our Slack channel at https://slack.sglang.ai to discuss your PR.

[gemini-code-assist]

Hello @ShangmingCai, 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!

Summary of Changes

Hello team, gemini-code-assist here to provide a summary of this pull request for a quick overview.

This PR, authored by ShangmingCai, addresses a bug in the mini_lb component related to handling ChatCompletion requests when streaming is enabled. Specifically, it fixes an issue where the load balancer was using an incorrect hardcoded endpoint path (/generate) when forwarding streaming requests to the backend prefill and decode servers. The change ensures that the correct endpoint path, determined by the request itself, is used instead.

Highlights

• Bugfix: Fixes a bug affecting the ChatCompletion endpoint in the mini_lb when streaming is active.
• Endpoint Correction: Corrects the endpoint path used by the load balancer when forwarding streaming requests to backend servers, changing from a hardcoded /generate to a dynamic variable {endpoint}.

Changelog

• python/sglang/srt/disaggregation/mini_lb.py
  • Modified the stream_results function to use the variable endpoint instead of the hardcoded /generate path for POST requests to prefill_server and decode_server (lines 120-121).

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 in your pull request via creating an issue comment (i.e. comment on the pull request page) using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands.

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 issue 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 is currently in preview and 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 to provide feedback.

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.

---

Streaming data flows,
Endpoint path was set wrong,
Bug fixed, now it goes.

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. ↩

[gemini-code-assist]

Code Review

This pull request addresses a bug in the ChatCompletion endpoint for the mini load balancer when streaming is enabled. The change correctly uses the endpoint parameter when making POST requests to the prefill and decode servers, which resolves the issue where it was previously hardcoded to /generate.

The fix itself is clear, concise, and directly addresses the problem. Well done on identifying and correcting this!

One area for improvement is regarding testing. For bugfixes, it's highly beneficial to add a regression test that specifically covers the scenario that caused the bug. This ensures the bug doesn't reappear in the future. I notice the "Add unit tests" item in the PR checklist is currently unchecked. Adding such a test would increase confidence in the fix and improve long-term maintainability. I've classified this as a medium-severity concern.

Summary of Findings

• Regression Testing for Bugfix: This PR fixes a bug in the ChatCompletion streaming endpoint. To prevent regressions and ensure the fix is robust, it's recommended to add a unit or integration test that specifically covers the scenario addressed by this bugfix. This is particularly important as the 'Add unit tests' item in the PR checklist is currently unchecked. This is considered a medium-severity concern for maintainability.

Merge Readiness

The code change in this PR correctly fixes the identified bug in the ChatCompletion streaming endpoint. The fix is well-implemented. However, to ensure the long-term stability and prevent regressions, I recommend adding a specific test case for this bugfix before merging. As a reviewer, I am not authorized to approve pull requests, but addressing the testing concern would make this PR ready for merge by those with approval permissions. Due to the medium-severity concern regarding testing, changes are requested.