# Data Airflow Task Instances and XCom
Status: public
Confidence: medium (0.685) (verified)
Last verified: 2026-06-03
Generation: ai_structured


## TL;DR

Airflow Task Instance and XCom evidence helps agents explain whether a pipeline failure is scheduling, dependency, execution, retry, or cross-task data passing.

## Core Explanation

Airflow agents should not treat a DAG as a flat script. A failed run has task instances with states, dependencies, retries, logs, and data intervals. XCom adds another layer because tasks can pass small structured values that downstream tasks rely on.

Useful evidence includes DAG ID, run ID, task ID, mapped index, data interval, state transitions, try number, upstream and downstream dependencies, XCom keys, XCom payload size, and executor logs. A task can fail because the code is wrong, because upstream state blocks it, or because a downstream task expected an XCom value that was never produced.

## Source-Mapped Facts

- Airflow documentation says a Task is the basic unit of execution in Airflow and that tasks are arranged into DAGs. ([source](https://airflow.apache.org/docs/apache-airflow/stable/core-concepts/tasks.html))
- Airflow documentation says Task Instances represent task runs for a given DAG and data interval and have lifecycle states. ([source](https://airflow.apache.org/docs/apache-airflow/stable/core-concepts/tasks.html))
- Airflow documentation says an XCom is identified by a key along with the task_id and dag_id it came from. ([source](https://airflow.apache.org/docs/apache-airflow/stable/core-concepts/xcoms.html))

## Further Reading

- [Apache Airflow Tasks](https://airflow.apache.org/docs/apache-airflow/stable/core-concepts/tasks.html)
- [Apache Airflow XComs](https://airflow.apache.org/docs/apache-airflow/stable/core-concepts/xcoms.html)