Task SDK transition and XCom state management boundaries

Apache Airflow is transitioning toward a decoupled execution model via the Task SDK to reduce memory overhead by separating task execution logic from the core environment. This architectural shift aims to mitigate resource leaks associated with loading the full Airflow core in every worker process.

A specific uncertainty exists regarding the state management boundary between the Task SDK and the metadata database, particularly when processing complex XCom objects. While the SDK provides a lightweight runtime, it is unclear how the decoupling affects the consistency and memory footprint of large XCom retrievals compared to the legacy integrated model.

• How does the Task SDK handle the lifecycle of XCom data in memory to prevent leaks during high-volume task execution?
• What are the defined boundaries for state synchronization between the decoupled SDK and the core database?