Skip to main content
How-to guideApplies to: Unified engine deployments
Repeat selected tasks from a completed run while reusing eligible successful work. A selective rerun creates a new linked run; the source run stays unchanged.

Choose the right action

Use a new normal run if the purpose is to execute corrected current code. Selective reruns preserve the earlier recorded context, parameters, and window.

1. Complete the example graph

Run Build a multi-task workflow successfully. Record its workload ID and run ID. The graph has orders, by_category, by_region, and combine. To repeat only the final combination, preview a selection containing combine. In the completed run, select Rerun tasks. Clear Include failed, skipped, and cancelled tasks and Include downstream tasks, then select combine. Choose Preview and review Execute, Reuse, and any blocked reasons. The equivalent HTTP preview is:

2. Read the preview

The plan separates tasks into execute, reuse, and blocked. For a successful source run with retained required results, expect combine to execute and the three upstream tasks to be reused. blocked should be empty. A successful upstream task is not enough when its rows are required but no longer available. Include that task in the execution selection, and include further upstream tasks if their own required results are unavailable. To repeat the category summary and its consumers, select by_category with include_downstream: true. This adds combine. The region summary can be reused if its result is still available. failed_only: true adds failed, skipped, and cancelled tasks to the selection. It does not mean that every task outside that selection can always be reused. Check blocked prerequisites.

3. Start the linked run

Select Start rerun when the plan is ready. Through HTTP, POST the same reviewed selection to /api/v1/workloads/{workload_id}/runs/{run_id}/rerun. Reuse the submission ID and identical request if the response is lost. Use a new submission ID for another deliberate rerun. In the dialog, Check submission recovers an uncertain start with the same request. Do not change the selection and create another run just because an acknowledgement was lost. Open the new run. Confirm a different run ID, its source-run link, the selected tasks’ new attempts, and reused-task origins. The final summary should remain Books 60, Games 45, Garden 35, North 85, and South 55.

Understand the boundaries

The source run must be terminal with confirmed outcomes and completed cleanup. Pending, active, uncertain, or still-cleaning runs cannot be used. Runs without saved invocation settings cannot supply the required historical context. Selective reruns are for saved batch workloads without streaming or incremental cursors. They do not reset progress or implement streaming repair. Reusing a successful write means preserving that confirmed outcome, not replaying the write. It does not freeze the current contents of the destination table. Re-executing a selected append can still add duplicate data. Use Results and table commits when deciding whether a publication task should be reused or executed again.
Last modified on November 4, 2026