Trace Utilities

Utility functions and classes for tracing, including ThreadPoolExecutor

The trace utilities module provides helper functions and classes for advanced tracing scenarios.

ThreadPoolExecutor

A context-aware ThreadPoolExecutor that preserves trace context across thread boundaries.

from weave import ThreadPoolExecutor

@weave.op
def process_item(item: str) -> str:
    return item.upper()

@weave.op
def process_batch(items: List[str]) -> List[str]:
    with ThreadPoolExecutor(max_workers=4) as executor:
        # Each item is processed in parallel while maintaining trace context
        results = list(executor.map(process_item, items))
    return results

Key Features

• Preserves parent-child relationships in traces
• Maintains trace context across thread boundaries
• Compatible with standard ThreadPoolExecutor API

Example with Nested Tracing

@weave.op
def fetch_data(url: str) -> dict:
    # Simulated API call
    return {"data": f"from {url}"}

@weave.op
def process_urls(urls: List[str]) -> List[dict]:
    with ThreadPoolExecutor(max_workers=10) as executor:
        # All fetch_data calls will be properly nested under process_urls
        results = list(executor.map(fetch_data, urls))
    return results

# Usage
urls = ["http://api1.com", "http://api2.com", "http://api3.com"]
data = process_urls(urls)

attributes

Context manager for adding attributes to calls.

from weave import attributes

@weave.op
def my_function():
    return "result"

# Add metadata to the call
with attributes({'environment': 'production', 'version': '1.0.0'}):
    my_function()

Nested Attributes

Attributes can be nested and will merge:

with attributes({'env': 'prod'}):
    with attributes({'region': 'us-east-1'}):
        # Call will have both attributes
        my_function()

set_tracing_enabled

Context manager to temporarily disable tracing.

from weave.trace.context.call_context import set_tracing_enabled

@weave.op
def traced_function():
    return "This will be traced"

# Temporarily disable tracing
with set_tracing_enabled(False):
    traced_function()  # This call won't be traced

# Tracing is automatically re-enabled outside the context
traced_function()  # This will be traced

get_current_call

Get the currently executing call within an op.

from weave.trace.context import get_current_call

@weave.op
def my_function():
    call = get_current_call()
    if call:
        print(f"Current call ID: {call.id}")
        print(f"Current call inputs: {call.inputs}")

Summary Utilities

add_summary

Add custom summary metrics to a call.

@weave.op
def evaluate_model(prompt: str) -> str:
    result = model.generate(prompt)
    
    # Add custom metrics to the call summary
    call = get_current_call()
    if call:
        call.summary.update({
            'token_count': len(result.split()),
            'confidence': 0.95
        })
    
    return result

These utilities are designed to work seamlessly with Weave’s tracing system. Always ensure weave.init() has been called before using these features.