# Directed Acyclic Graph (DAG)

> A directed graph with no path that loops back to its starting node, useful for representing dependencies and executable plans.

- HTML version: https://robbiepalmer.me/ideas/directed-acyclic-graph
- Source: https://dvc.org/doc/command-reference/dag

A directed acyclic graph connects nodes with one-way edges and forbids cycles. If `B` depends on
`A`, the edge fixes their order. Because no chain of dependencies can return to its starting point,
the nodes can always be placed in a topological order.

That makes a DAG a useful model for work with prerequisites. Build systems, data pipelines, stream
processors, infrastructure planners, and workflow engines can find what is ready, run independent
branches concurrently, and avoid repeating an unchanged branch. A node may have several parents or
children, so a DAG is more general than a tree.

The graph is only as good as its declared edges. An omitted dependency can produce stale output,
while a cycle means the proposed ordering is impossible. I find DAGs most useful when the graph is
inspectable and the execution state remains separate from the dependency model.

## Questions it prompts

* What must exist before this node can run?
* Which branches can proceed concurrently?
* What should a change invalidate downstream?
* Does a cycle reveal a modelling mistake or two concerns that need separating?

## Where it appears

- Technology: [dbt](https://robbiepalmer.me/technologies/dbt.md)
- Technology: [DVC](https://robbiepalmer.me/technologies/dvc.md)
- Technology: [Flink](https://robbiepalmer.me/technologies/flink.md)
- Technology: [ksqlDB](https://robbiepalmer.me/technologies/ksqldb.md)
- Technology: [Quix Streams](https://robbiepalmer.me/technologies/quix-streams.md)
- Technology: [Terraform](https://robbiepalmer.me/technologies/terraform.md)

---

Markdown index of this site: https://robbiepalmer.me/llms.txt