Before you start
You need a running engine, workload access, and available compute. All input rows are included in the query. This example creates no tables and uses a Manual schedule.1. Open the complete workload
Create a workload with a task namedsummarize using the definition below. Keep the workload name or choose your own.
2. Validate and save
Review the task and keep the schedule Manual. Validate the definition, then create the workload. Validation checks the definition and SQL syntax. It does not execute the task or prove that worker allocation will succeed. Saving a Manual workload does not start a run.3. Start a run
Select Start run. Open the created run and record its run ID and workload revision. Followsummarize until it reaches Succeeded.
The expected query result is:
Run and task pages show execution details. To check rows interactively, run the task’s SQL in Workbench. The same SQL includes all its input rows. See Results and table commits for retained result access through the API.
4. Compare a second run
Start another run without editing the workload. Expect a different run ID, the same workload revision, and the same result. Both runs should show a successfulsummarize task.
Change the SQL to count only region = 'North', save the revision, then start a new run. Expect four orders and sales of 85. The earlier runs keep their recorded invocation.