orchestration

Dagster

Dagster is a orchestration product that asset-based scheduling for pipelines. This profile covers what it does, how it is priced, and where it stops being the right choice.

What the product actually does

Dagster is an orchestrator built around the idea that a pipeline's assets matter more than its tasks. In a task-based orchestrator you describe steps and their order; in Dagster you describe the tables and files that should exist, and the framework works out what to run to produce them. The difference shows up when something fails: an asset-based system knows which downstream tables are now stale, and a task-based system knows only that a step failed.

The second design choice is that pipelines are ordinary Python functions with type annotations, not configuration files. That makes them testable in the way the rest of a codebase is testable, which is a larger advantage than it first appears.

How pricing works once you are inside

The open-source version is free and complete enough for most teams. Dagster+ — the hosted product — prices per credit, where credits map to compute and orchestration volume. The pricing model is consumption-based, so a team that runs many small jobs frequently pays differently from one that runs few large jobs.

The self-hosted option remains viable at real scale, which is unusual for modern orchestrators. Teams that already run Kubernetes can host Dagster without buying anything, and many do.

Where it breaks down

The asset model has a learning curve that is steeper than a task model, and teams coming from Airflow consistently underestimate it. The first month is slower than the equivalent Airflow build; the payoff arrives later, when a schema change needs to propagate and the graph knows how.

The ecosystem is smaller than Airflow's. Provider packages, community operators and Stack Overflow answers are all thinner, and a niche integration may simply not exist.

What switching costs

High. Pipeline definitions are Dagster-specific Python, and the asset graph is the valuable artefact. Moving to another orchestrator means rebuilding the graph, not translating it.

Who should not use it

Teams with a handful of simple scheduled jobs. The asset model's benefits appear at scale and complexity; below that it is overhead.

Sources

Dagster public documentation, pricing page and release notes, read October 2026.