Resume from the point of failure
A retry is useful when a temporary issue affected one step and the earlier work is still valid. The workflow begins again at the selected step, or at the failed step by default, while retaining preceding step outputs. This avoids repeating every earlier operation when the workflow's dependencies are unchanged.
A run cannot be retried while it is still running, and a step cannot be retried if a preceding step did not succeed. Those conditions make the restart point explicit: earlier outputs must be available before the selected step can run again. If the workflow's input or earlier data has changed, consider starting a new run instead of reusing stale outputs.
Inspect attempts and side effects
The run data keeps previous attempts for a retried step. Review the errors and outputs to see whether an earlier attempt partially completed before it failed. A retry may repeat external actions, such as sending a message or creating a record, so check whether the operation is safe to repeat or whether it needs a duplicate check.
If a failed step depends on a changed configuration, update the workflow and then retry from the correct step or start a fresh run. Do not assume a retry rolls back earlier changes: it preserves successful prior outputs and re-executes the selected portion. Confirm the final run status and any external result before telling users the workflow completed.