The aws.lambda.load span shows massively inflated durations on cold starts when using AWS Lambda provisioned concurrency.

Test Results:

1. Deployed the Lambda function with provisioned concurrency
2. Invoke for the first time (cold start)
3. Result exemple:
   • aws.lambda (cold start span): 6 seconds
   • aws.lambda.load span: 19 minutes 41 seconds

Affected Spans

Only the aws.lambda.load span is affected.
The main aws.lambda cold start span correctly shows 6 seconds. Warm start invocations show ~8 seconds and have no aws.lambda.load span.

Environment

• dd-trace version: 5.73.0
• datadog-cdk-constructs-v2: 3.2.2
• Node.js version: 22.x
• AWS Lambda Runtime: Node.js with provisioned concurrency enabled
• Datadog Lambda Extension: Layer version 92

Questions

1. Is this the expected behavior or a bug?
2. Are there any recommended approaches for handling trace context in long-lived Lambda environments with provisioned concurrency?
3. Should dd-trace handle this case automatically?

Thank you!