Fix: LLM telemetry logging

[enyst]

Summary

• Extend telemetry _safe_json to serialize any Pydantic BaseModel via model_dump(mode="json", exclude_none=True) to avoid circular-reference and Pydantic serialization warnings when logging complex objects (e.g., tools/specs) that may appear in the log context for remote agent/server scenarios
• Add targeted test ensuring logging succeeds when the log context includes a self-referencing Pydantic model with excluded runtime-only fields

Problem
When running the agent remotely (Docker/remote runtime), enabling log_completions caused telemetry logging to fail:

• If the default log directory wasn't present inside the container: log_dir does not exist: logs/completions
• After creating a directory, log files were created but empty, with warnings like Circular reference detected.
  This happened because Telemetry.log_llm_call attempted to JSON-serialize arbitrary objects from the logging context (e.g., Pydantic-based tool/config objects) using generic fallbacks (dict), which can include self-references or non-serializable fields (like executors), leading to circular references and failed writes.

Changes

• telemetry.py: update _safe_json
  • Handle ModelResponse and ResponsesAPIResponse (unchanged) via model_dump(mode="json", exclude_none=True)
  • NEW: handle any Pydantic BaseModel via model_dump(mode="json", exclude_none=True) to ensure safe, cycle-free serialization respecting field exclusions
  • Fallback to dict or str(obj) remains for plain objects
• tests/sdk/llm/test_llm_telemetry.py: add test_log_completion_with_pydantic_objects_in_context
  • Simulates a Pydantic model with a self-referencing, excluded field (executor)
  • Verifies no circular reference warnings and that a valid JSON log is produced

Compatibility / Impact

• Backward compatible: only improves robustness of telemetry serialization. Existing behavior for non-Pydantic objects remains unchanged.
• No public API changes.

Testing

• Ran pre-commit hooks on changed files
• Ran telemetry tests (35/35 pass) and the new test passes
• Existing integration tests for log_completions also pass locally

Docs

• No docs changes required; this is an internal robustness fix

Fixes #1027

@enyst can click here to continue refining the PR

---

Agent Server images for this PR

• GHCR package: https://github.com/OpenHands/agent-sdk/pkgs/container/agent-server

Variants & Base Images

Variant	Architectures	Base Image	Docs / Tags
java	amd64, arm64	eclipse-temurin:17-jdk	Link
python	amd64, arm64	nikolaik/python-nodejs:python3.12-nodejs22	Link
golang	amd64, arm64	golang:1.21-bookworm	Link

Pull (multi-arch manifest)

# Each variant is a multi-arch manifest supporting both amd64 and arm64
docker pull ghcr.io/openhands/agent-server:523e655-python

Run

docker run -it --rm \
  -p 8000:8000 \
  --name agent-server-523e655-python \
  ghcr.io/openhands/agent-server:523e655-python

All tags pushed for this build

ghcr.io/openhands/agent-server:523e655-golang-amd64
ghcr.io/openhands/agent-server:v1.0.0a5_golang_tag_1.21-bookworm_binary-amd64
ghcr.io/openhands/agent-server:523e655-golang-arm64
ghcr.io/openhands/agent-server:v1.0.0a5_golang_tag_1.21-bookworm_binary-arm64
ghcr.io/openhands/agent-server:523e655-java-amd64
ghcr.io/openhands/agent-server:v1.0.0a5_eclipse-temurin_tag_17-jdk_binary-amd64
ghcr.io/openhands/agent-server:523e655-java-arm64
ghcr.io/openhands/agent-server:v1.0.0a5_eclipse-temurin_tag_17-jdk_binary-arm64
ghcr.io/openhands/agent-server:523e655-python-amd64
ghcr.io/openhands/agent-server:v1.0.0a5_nikolaik_s_python-nodejs_tag_python3.12-nodejs22_binary-amd64
ghcr.io/openhands/agent-server:523e655-python-arm64
ghcr.io/openhands/agent-server:v1.0.0a5_nikolaik_s_python-nodejs_tag_python3.12-nodejs22_binary-arm64
ghcr.io/openhands/agent-server:523e655-golang
ghcr.io/openhands/agent-server:523e655-java
ghcr.io/openhands/agent-server:523e655-python

About Multi-Architecture Support

• Each variant tag (e.g., 523e655-python) is a multi-arch manifest supporting both amd64 and arm64
• Docker automatically pulls the correct architecture for your platform
• Individual architecture tags (e.g., 523e655-python-amd64) are also available if needed

[github-actions]

Coverage Report •

File	Stmts	Miss	Cover	Missing
openhands-sdk/openhands/sdk/llm/utils
telemetry.py	135	57	57%	87, 93, 102, 106, 109, 114–115, 150, 157, 179, 183–184, 192–194, 200, 217–219, 222–225, 227, 235–236, 240–242, 245–248, 253, 258, 261, 266, 269, 275, 281, 283, 286–287, 292, 297–301, 308–309, 312, 314, 317–320
TOTAL	11664	5407	53%

[blacksmith-sh]

[Automatic Post]: I have assigned @xingyaoww as a reviewer based on git blame information. Thanks in advance for the help!