[OPEN-7119] Add support for Google ADK

[viniciusdsmello]

Pull Request

Summary

This PR adds comprehensive tracing support for Google Agent Development Kit (ADK), enabling Openlayer users to trace agent execution, LLM calls, tool calls, and multi-agent workflows with proper hierarchy and metadata extraction.

Closes OPEN-7119

Changes

• Core Integration: Added google_adk_tracer.py with support for:

  • Agent execution tracing with sub-agent detection
  • LLM call instrumentation (Gemini models)
  • Tool execution tracking
  • Token usage extraction
  • System instruction handling
  • Session/user ID tracking

• Smart Agent Nesting: Implemented context variable system to ensure sub-agents appear as siblings of LLM calls (not nested under tool calls), providing cleaner trace hierarchies

• NoOp Tracer: Replaced ADK's built-in OpenTelemetry tracer to prevent duplicate spans

• Examples & Documentation:

  • Created interactive Jupyter notebook (google_adk_tracing.ipynb) with multi-agent workflow example
  • Added complete example code with multi-agent system (creator, list videos, choose video, extract text, script, text-to-post agents)
  • Added mock data support for YouTube API and transcript extraction (no API keys needed for demos)

• Testing: Added 16 comprehensive unit tests covering:

  • Conditional imports
  • NoOp span/tracer behavior
  • Message format conversion
  • LLM attribute extraction
  • Agent metadata extraction
  • Agent transfer handling
  • Tool execution tracing

• Export: Added trace_google_adk and unpatch_google_adk to __init__.py

• Documentation: Updated README.md with Google ADK tracing section

Context

Google Agent Development Kit (ADK) is Google's framework for building multi-agent systems with Gemini models. This integration enables Openlayer customers using ADK to:

1. Monitor Agent Performance: Track agent execution times, LLM calls, and tool usage
2. Debug Multi-Agent Workflows: Visualize agent transfers and sub-agent hierarchies
3. Analyze Token Usage: Track prompt/completion tokens across all agent interactions
4. Session Tracking: Maintain user/session context across agent executions

Key Technical Decisions:

• Used wrapt for method patching to ensure clean instrumentation
• Implemented context variables for tracking user queries and agent transfers
• Skipped tool steps for agent transfers to avoid excessive nesting
• Created sub-agent steps with explicit parent_step parameter for proper hierarchy

Testing

• Unit tests
• Manual testing
• Postman CI/CD
• Other (please specify)