[Bug]: vLLM rejects valid JSON when tools array contains 17+ tools (OpenAI format)

[islam-prevera]

Your current environment

Environment: - vLLM Version: 0.11.0 (from Docker image vllm/vllm-openai:latest) - Model: THUDM/glm-4-9b-chat and Qwen/Qwen2.5-14B-Instruct - Server: Docker container with default settings - API: OpenAI-compatible /v1/chat/completions

Configuration:
docker run -d --gpus all -p 8000:8000
vllm/vllm-openai:latest
--model THUDM/glm-4-9b-chat
--max-model-len 8192

🐛 Describe the bug

When sending a valid JSON request with 17 tools in the tools array, vLLM returns:
400 Bad Request
{"error": {"message": "1 validation error for list[function-wrap[log_extra_fields()]]\n Invalid JSON: EOF while parsing a string at line 57", "type":
"BadRequestError"}}

The error location varies (lines 23, 57, 84, 171, 261, 271, 346, 1024 in different requests), but always reports "EOF while parsing" errors.

The input_value in the error shows corrupted/truncated data (just newlines/whitespace).

Before submitting a new issue...

• Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.

[bbrowning]

Do you have an example request that reproduces this issue consistently that others can run against their local vLLM to reproduce and track down what's happening? Does this only happen for you with lots of tools, or are there other similarly large request bodies that don't involve lots of tools that also trigger this error?