Empty traces with flush race conditions

Applies to:

Plan -
Deployment -

Summary

Issue: Traces display empty fields or remain “in progress” indefinitely. Cause: Application terminates before flush() completes or spans aren’t properly ended, particularly in serverless/container environments. Resolution: Explicitly call flush() before termination and ensure all spans are ended.

Resolution Steps

Step 1: Use context managers (recommended)

Python

with logger.start_span(name="operation") as span:
    span.log(input=input_data, output=output_data, metadata=metadata)
    # span.end() called automatically

TypeScript

await logger.traced(async (span) => {
  span.log({ input: inputData, output: outputData, metadata });
  // span.end() called automatically
});

Step 2: Call flush() explicitly before process exit

Python

def cleanup_braintrust():
    try:
        project.logger.flush()  # Blocks until upload completes
    except Exception:
        logging.exception("Failed to flush Braintrust project")

TypeScript

async function cleanupBraintrust() {
  try {
    await project.logger.flush(); // Waits until upload completes
  } catch (error) {
    console.error("Failed to flush Braintrust project:", error);
  }
}

Key Points

flush() waits for completion
Context managers automatically call span.end()
Manual spans: call span.end() in finally blocks or they remain incomplete
For serverless functions, call flush() before the handler returns

⌘I

​Summary

​Resolution Steps

​Step 1: Use context managers (recommended)

​Python

​TypeScript

​Step 2: Call flush() explicitly before process exit

​Python

​TypeScript

​Key Points

Summary

Resolution Steps

Step 1: Use context managers (recommended)

Python

TypeScript

Step 2: Call flush() explicitly before process exit

Python

TypeScript

Key Points