Fix MCP stdio communication issues #326
Conversation
Kylie-dot-s
commented
Dec 26, 2025
Kylie-dot-s
commented
- Fix rule_resume.py: Replace invalid eval_details/eval_status with proper EvalDetail fields (status, label, reason)
- Fix mcp_server.py: Add environment variables to force ThreadPool usage and prevent ProcessPool fd inheritance
- Set LOCAL_DEPLOYMENT_MODE=true to use ThreadPoolExecutor instead of ProcessPoolExecutor
- Set TQDM_DISABLE=1 to prevent progress bar output pollution
- Add stdout/stderr redirection with StringIO for better error handling
- Fix rule name extraction logic to handle both class and instance types
- Add defensive checks for Model.prompt_name_map access
- Change default rule group fallback from 'default' to 'sft' to avoid buggy Resume rules
- Fix rule_resume.py: Replace invalid eval_details/eval_status with proper EvalDetail fields (status, label, reason) - Fix mcp_server.py: Add environment variables to force ThreadPool usage and prevent ProcessPool fd inheritance * Set LOCAL_DEPLOYMENT_MODE=true to use ThreadPoolExecutor instead of ProcessPoolExecutor * Set TQDM_DISABLE=1 to prevent progress bar output pollution * Add stdout/stderr redirection with StringIO for better error handling * Fix rule name extraction logic to handle both class and instance types * Add defensive checks for Model.prompt_name_map access * Change default rule group fallback from 'default' to 'sft' to avoid buggy Resume rules
Summary of ChangesHello @Kylie-dot-s, 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 focuses on enhancing the stability and reliability of the MCP server's communication and Dingo evaluation processes. It addresses critical issues related to Highlights
Using Gemini Code AssistThe 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
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 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
|
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request effectively addresses stdio communication issues in mcp_server.py by forcing single-threaded execution and redirecting output streams to prevent interference. It also introduces several valuable refactorings and robustness improvements, such as adding defensive checks, improving LLM configuration handling, and making rule processing more flexible. The changes align well with the PR description and significantly enhance the server's stability and maintainability. I have a few suggestions to further improve the code, mainly regarding a performance concern and some minor code duplication.
| input_data["max_workers"] = 1 | ||
| log.info(f"Forced max_workers=1 to use ThreadPool mode (avoiding ProcessPool fd inheritance)") |
There was a problem hiding this comment.
This change unconditionally forces max_workers to 1 to solve stdio communication issues. While this is effective for stdio mode, it also impacts other transport modes like sse, potentially limiting performance unnecessarily.
Consider making this change conditional on the transport mode. This would solve the stdio issue without affecting other deployment scenarios.
| input_data["max_workers"] = 1 | |
| log.info(f"Forced max_workers=1 to use ThreadPool mode (avoiding ProcessPool fd inheritance)") | |
| if os.environ.get("MCP_TRANSPORT", "stdio") == "stdio": | |
| input_data["max_workers"] = 1 | |
| log.info(f"Forced max_workers=1 to use ThreadPool mode (avoiding ProcessPool fd inheritance)") |
| if isinstance(rule_group_data, list): | ||
| rules_iterable = rule_group_data | ||
| elif hasattr(rule_group_data, 'rules'): | ||
| rules_iterable = rule_group_data.rules | ||
| else: | ||
| rules_iterable = [] | ||
| log.warning(f"Unknown structure for rule group '{rule_group}'") | ||
|
|
||
| # 2. 智能提取名字 (修复 'type' 问题) | ||
| evals_list = [] | ||
| for rule in rules_iterable: | ||
| # 如果 rule 是个类 (type)，直接取 __name__ | ||
| # 如果 rule 是个实例，取 __class__.__name__ | ||
| if isinstance(rule, type): | ||
| r_name = rule.__name__ | ||
| else: | ||
| r_name = rule.__class__.__name__ | ||
|
|
||
| evals_list.append({"name": r_name}) |
There was a problem hiding this comment.
The logic here for iterating through rule_group_data and handling both class and instance types is very similar to the logic in the _get_rule_details_logic function (lines 877-907). To improve maintainability and follow the DRY (Don't Repeat Yourself) principle, consider extracting this shared iteration logic into a common helper function. This would centralize the handling of different rule group structures.
| log.info(f"Found {len(evals_list)} rules in group '{rule_group}'") | ||
| else: | ||
| # Fallback to common rules | ||
| evals_list = [{"name": "RuleColonEnd"}, {"name": "RuleContentNull"}, {"name": "RuleDocRepeat"}] |
There was a problem hiding this comment.
The list of fallback rules [{\"name\": \"RuleColonEnd\"}, {\"name\": \"RuleContentNull\"}, {\"name\": \"RuleDocRepeat\"}] is hardcoded here and also in the except block on line 681. To improve maintainability and avoid duplication, consider defining this list as a constant at the top of the file, for example: FALLBACK_RULES = [...].
- Add noqa: E402 comments to imports that must stay after os.environ setup - Fix E261: Add proper spacing before inline comments - Fix F841: Remove unused variable 'inner_e' in exception handler - Fix E115: Correct indentation of FIX END comment
- Keep noqa: E402 comments for imports after os.environ setup - Maintain proper import order and formatting
2 participants
