Directed acyclic graph (DAG)

A directed acyclic graph, almost always called a DAG, is a set of nodes connected by one-directional edges in which it is impossible to start at any node and follow the edges back to itself. In data engineering, nodes represent tasks, extract a table, run a transformation, send an alert, and edges represent dependencies within a data pipeline: a task only starts once every task pointing into it has finished.

The "acyclic" property is what makes DAGs useful for scheduling: because there are no loops, a valid execution order always exists, and an orchestrator can compute it automatically from the graph rather than requiring someone to hand-sequence every task. This differs from a general dependency graph or flowchart, which may contain cycles and therefore cannot be resolved into a single run order without extra rules.

Workflow orchestration tools such as Airflow, Dagster and Prefect represent every pipeline as a DAG, which is why practitioners use the term as shorthand for "the pipeline definition" itself. The main pitfall is treating the DAG purely as a diagram rather than as a real constraint: tasks that are secretly interdependent but not declared as such in the graph can run out of order or in parallel when they should not, producing subtly wrong results with no obvious error, and reprocessing history through a backfill follows the same graph, so an incomplete DAG produces the same wrong order every time it reruns.