File size: 10,087 Bytes
a1e2111 38bdec7 a1e2111 38bdec7 a1e2111 38bdec7 a1e2111 38bdec7 a1e2111 38bdec7 a1e2111 38bdec7 a1e2111 38bdec7 a1e2111 38bdec7 a1e2111 38bdec7 a1e2111 38bdec7 a1e2111 38bdec7 a1e2111 38bdec7 a1e2111 38bdec7 a1e2111 38bdec7 a1e2111 38bdec7 a1e2111 38bdec7 a1e2111 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 | """Langfuse tracking integration for GAIA Benchmark Agent.
This module provides decorators and context managers for tracking:
- Agent execution sessions
- LLM invocations
- Tool calls
- Question processing
"""
import os
import functools
import time
from typing import Any, Optional, Dict
from contextlib import contextmanager
# Langfuse will be imported conditionally
langfuse = None
try:
from langfuse import Langfuse, observe
LANGFUSE_AVAILABLE = True
except ImportError:
LANGFUSE_AVAILABLE = False
print("[INFO] Langfuse not installed. Tracking is disabled. Install with: pip install langfuse")
class LangfuseTracker:
"""Singleton class to manage Langfuse client and tracking state."""
_instance = None
_client = None
_enabled = False
def __new__(cls):
if cls._instance is None:
cls._instance = super().__new__(cls)
return cls._instance
def __init__(self):
"""Initialize Langfuse client if not already initialized."""
if self._client is None and LANGFUSE_AVAILABLE:
public_key = os.getenv("LANGFUSE_PUBLIC_KEY")
secret_key = os.getenv("LANGFUSE_SECRET_KEY")
host = os.getenv("LANGFUSE_HOST", "https://us.cloud.langfuse.com")
if public_key and secret_key:
self._client = Langfuse(
public_key=public_key,
secret_key=secret_key,
host=host
)
self._enabled = True
print(f"[LANGFUSE] Tracking enabled (host: {host})")
else:
print("[LANGFUSE] Tracking disabled. Set LANGFUSE_PUBLIC_KEY and LANGFUSE_SECRET_KEY to enable.")
@property
def enabled(self) -> bool:
"""Check if Langfuse tracking is enabled."""
return self._enabled and LANGFUSE_AVAILABLE
@property
def client(self):
"""Get Langfuse client instance."""
return self._client if self.enabled else None
# Global tracker instance
tracker = LangfuseTracker()
def track_agent_execution(agent_type: str):
"""Decorator to track agent execution lifecycle.
Args:
agent_type: Type of agent (e.g., "LangGraph", "ReAct", "LlamaIndex")
Usage:
@track_agent_execution("LangGraph")
def __call__(self, question: str, file_name: str = None) -> str:
...
"""
def decorator(func):
if not tracker.enabled:
return func
@observe(name=f"{agent_type}_Agent_Execution")
@functools.wraps(func)
def wrapper(self, question: str, file_name: str = None, *args, **kwargs):
# Add metadata to current observation
if tracker.client:
tracker.client.update_current_span(
metadata={
"agent_type": agent_type,
"has_file": file_name is not None,
"file_name": file_name or "none",
"question_length": len(question)
},
input={"question": question[:500], "file_name": file_name} # Limit question length
)
start_time = time.time()
try:
result = func(self, question, file_name, *args, **kwargs)
# Update with output and success metrics
if tracker.client:
tracker.client.update_current_span(
output={"answer": str(result)[:500]}, # Limit answer length
metadata={
"execution_time_seconds": time.time() - start_time,
"success": not result.startswith("Error:")
}
)
return result
except Exception as e:
# Track errors
if tracker.client:
tracker.client.update_current_span(
level="ERROR",
status_message=str(e),
metadata={
"execution_time_seconds": time.time() - start_time,
"error": str(e)
}
)
raise
return wrapper
return decorator
def track_llm_call(model_name: str):
"""Decorator to track LLM invocations.
Args:
model_name: Name of the LLM model being called
Usage:
@track_llm_call("gemini-1.5-flash")
def _assistant(self, state):
response = self.llm_client_with_tools.invoke(state["messages"])
...
"""
def decorator(func):
if not tracker.enabled:
return func
@observe(as_type="generation", name=f"LLM_Call_{model_name}")
@functools.wraps(func)
def wrapper(*args, **kwargs):
start_time = time.time()
try:
result = func(*args, **kwargs)
# Update generation with model info
if tracker.client:
tracker.client.update_current_generation(
model=model_name,
metadata={
"latency_seconds": time.time() - start_time,
}
)
return result
except Exception as e:
if tracker.client:
tracker.client.update_current_generation(
level="ERROR",
status_message=str(e),
metadata={
"latency_seconds": time.time() - start_time,
"error": str(e)
}
)
raise
return wrapper
return decorator
def track_tool_call(tool_name: str):
"""Decorator to track tool/function calls.
Args:
tool_name: Name of the tool being called
Usage:
@track_tool_call("websearch")
def websearch(query: str, num_results: int = 5):
...
"""
def decorator(func):
if not tracker.enabled:
return func
@observe(name=f"Tool_{tool_name}")
@functools.wraps(func)
def wrapper(*args, **kwargs):
# Capture input parameters
if tracker.client:
tracker.client.update_current_span(
input={
"tool": tool_name,
"args": args[:3] if args else [], # Limit args
"kwargs": {k: str(v)[:100] for k, v in list(kwargs.items())[:5]} # Limit kwargs
},
metadata={"tool_name": tool_name}
)
start_time = time.time()
try:
result = func(*args, **kwargs)
# Track output
result_str = str(result)
if tracker.client:
tracker.client.update_current_span(
output={
"result_preview": result_str[:500],
"result_length": len(result_str)
},
metadata={
"execution_time_seconds": time.time() - start_time,
"success": True
}
)
return result
except Exception as e:
if tracker.client:
tracker.client.update_current_span(
level="ERROR",
status_message=str(e),
metadata={
"execution_time_seconds": time.time() - start_time,
"error": str(e)
}
)
raise
return wrapper
return decorator
@contextmanager
def track_session(session_name: str, metadata: Optional[Dict[str, Any]] = None):
"""Context manager to track a complete session (batch processing).
Args:
session_name: Name of the session (e.g., "Test_Run", "Full_Submission")
metadata: Optional metadata dict
Usage:
with track_session("Test_Run", {"agent": "LangGraph", "questions": 20}):
# Run agent on questions
...
"""
if not tracker.enabled:
yield
return
# Use start_as_current_span to create a root span/trace for the session
with tracker.client.start_as_current_span(
name=session_name,
metadata=metadata or {}
) as span:
try:
yield span
finally:
# Flush to ensure data is sent
if tracker.client:
tracker.client.flush()
@contextmanager
def track_question_processing(task_id: str, question: str):
"""Context manager to track individual question processing.
Args:
task_id: Unique task identifier
question: Question text
Usage:
with track_question_processing(task_id, question_text) as span:
answer = agent(question_text)
span.update(output={"answer": answer})
"""
if not tracker.enabled:
yield None
return
with tracker.client.start_as_current_span(
name=f"Question_{task_id[:8]}",
input={"task_id": task_id, "question": question[:300]},
metadata={"task_id": task_id}
) as span:
yield span
# Convenience function for manual span creation
def create_span(name: str, input_data: Optional[Dict] = None, metadata: Optional[Dict] = None):
"""Create a manual span for tracking custom operations.
Args:
name: Span name
input_data: Input data dict
metadata: Metadata dict
Returns:
Span context manager or None if tracking disabled
"""
if not tracker.enabled:
return None
return tracker.client.start_span(
name=name,
input=input_data,
metadata=metadata
)
|