ExplanationApplies to: Unified engine deployments
Turn a query into a repeatable workflow. Define tasks, connect their dependencies, choose compute, and decide when each run should happen.
A workload can run manually, once at a chosen time, on an interval, or on a cron schedule. Add sensors for external conditions, retry policies for eligible failures, and notifications for important events. Use data windows and backfills to process time periods explicitly.
Start with a working example
- Your first workload: run one SQL task with six inline orders.
- Build a multi-task workflow: pass rows, run parallel tasks, and combine results.
- Parameters and query context: reuse the same definition with different inputs.
- Schedules and time zones: turn a manual workflow into recurring work.
- Publish a table result: write to Iceberg and verify the committed output.
The first examples use WITH … VALUES. They need a running engine, workload access, and available compute. They do not need files, a catalog connection, credentials, or external data. Publication and external integrations list their own setup requirements.
Understand the moving parts
The scheduler creates run requests. The controller manages run progress and task dependencies. Workers execute task attempts. A scheduled due time is not a promise that a worker starts at that exact moment.
Choose your next step
Run examples in the product
Use the complete definitions below each example to create workloads in the Workload Editor or through HTTP. Review the tasks and dependencies, validate the definition, save it, then start a run. Keep practice schedules Manual until you choose to enable recurring execution. Saved workloads and started runs remain in the engine after you leave the console.Last modified on November 4, 2026
