Fix profiles do not have consistent names

[fzyzcjy]

Motivation

Modifications

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 @fzyzcjy, 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! Gemini-code-assist here, providing a summary of this pull request to help everyone get up to speed quickly.

This PR addresses an inconsistency related to profiling within the scheduler component. The primary goal is to ensure that when profiling data is exported, the output file name uses a consistent identifier (profile_id) provided by the profiling request, rather than a less predictable timestamp. This change aims to improve the organization and traceability of profiling results.

Highlights

• Profile ID Consistency: Renamed an internal variable from profiler_id to profile_id for better naming consistency.
• Pass Profile ID: Modified the profile method to correctly pass the profile_id received in the ProfileReq to the init_profile method.
• Consistent Output Filename: Updated the stop_profile method to use the stored profile_id when generating the filename for the exported trace, ensuring the filename is tied to the specific profile request rather than a timestamp.

Changelog

• python/sglang/srt/managers/scheduler.py
  • Renamed the instance variable self.profiler_id to self.profile_id in the __init__ method (L432).
  • Added recv_req.profile_id as an argument when calling self.init_profile within the profile method (L2147, L2157).
  • Added the profile_id: str parameter to the signature of the init_profile method (L2171).
  • Assigned the passed profile_id parameter to self.profile_id within the init_profile method (L2190).
  • Modified the stop_profile method to use self.profile_id instead of str(time.time()) when constructing the output filename for the trace export (L2290).

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.

---

A profile trace, a name it needs,
No random time, but planted seeds,
A consistent ID,
For all eyes to see,
Where performance truly leads.

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 aims to improve the consistency of profile names, primarily by renaming profiler_id to profile_id and using this ID for naming exported trace files. This is a positive change for clarity and traceability of profiles.

However, there's a critical issue regarding a potential TypeError due to a mismatch between the Optional[str] type of profile_id in ProfileReq and the str type expected by the init_profile method. This needs to be addressed to prevent runtime errors.

Additionally, while this PR makes the profile_id consistent, it's worth noting for future consideration that other profiler-related attributes in the Scheduler class still use a mix of profiler_ and profile_ prefixes (e.g., self.torch_profiler_output_dir, self.profiler_activities vs. self.profile_by_stage, self.profile_steps). Addressing this broader naming pattern could further enhance codebase consistency, though it's outside the scope of the current changes.

Summary of Findings

• Potential TypeError due to Optional profile_id: The init_profile method expects profile_id as str, but it can receive None from ProfileReq. If self.profile_id becomes None, it will cause a TypeError when used to construct filenames in stop_profile. This needs to be handled by either ensuring profile_id is always a string or by providing a default value within init_profile.

Merge Readiness

The pull request addresses an important consistency issue with profile naming. However, there's a critical issue identified concerning a potential TypeError if profile_id is not provided by the client. This must be addressed before the PR can be considered ready for merging. I am unable to approve pull requests, so please ensure this critical issue is resolved and the PR is reviewed and approved by other maintainers.

[gemini-code-assist]

The profile_id parameter is typed as str here, but it's derived from recv_req.profile_id which is defined as Optional[str] in the ProfileReq dataclass (in io_struct.py).

If recv_req.profile_id is None (which is its default value if not provided by the client), passing None to this parameter, which expects a str, could lead to unexpected behavior or type-related issues, especially if strict type checking were enforced or if subsequent code assumes it's always a string without further checks.

Consider changing the type hint to Optional[str] to accurately reflect that None might be passed. The handling of a None value should then be done explicitly in the method body (see related comment for line 2190).

[gemini-code-assist]

If profile_id (the parameter to init_profile) can be None (as suggested by its source ProfileReq.profile_id: Optional[str]), then assigning it directly to self.profile_id would result in self.profile_id being None.

This would subsequently cause a TypeError in the stop_profile method at line 2290, where self.profile_id is used in string concatenation to form the trace filename: self.profile_id + f"-TP-{self.tp_rank}" ....

To prevent this, it's crucial to ensure self.profile_id is always a string. A robust way to handle this is to assign a default generated ID if the input profile_id is None.

        if profile_id is None:
            # Generate a default profile ID if not provided.
            # Using a timestamp and a short random hex ensures reasonable uniqueness.
            self.profile_id = f"profile_{int(time.time())}_{os.urandom(4).hex()}"
        else:
            self.profile_id = profile_id