Question 580
During a model refresh rollout at Frostline Distribution, the information governance lead identifies a Databricks issue: A production workflow retries a non-idempotent charge-enrichment operation and duplicates the side effect on every attempt. The team reports that the correction is complete. Which result BEST verifies that the specific issue was resolved?
Explanation
Correct Answer
Two executions of the same input yield one accepted business side effect and a traceable retry record. The observation provides evidence that the proposed fix (Introduce idempotency keys or checkpointed write semantics before enabling automatic retries.) addressed the reported issue (A production workflow retries a non-idempotent charge-enrichment operation and duplicates the side effect on every attempt.) for Frostline Distribution and its inventory systems and supplier access. It verifies an observable outcome instead of relying only on a completed task ticket.
Incorrect Answers
An intentionally failed destination write marks the job unsuccessful and retains its error details. This is suitable evidence for another condition (A notebook catches every exception and returns success even when a destination write has failed.) after a different response (Propagate unrecoverable exceptions to the orchestrator while preserving useful diagnostic context.). For Frostline Distribution, it does not prove the required result: Two executions of the same input yield one accepted business side effect and a traceable retry record.
A simulated transient failure is retried within policy and either completes or generates an actionable terminal failure. This is suitable evidence for another condition (An ETL task fails intermittently on recoverable network timeouts but the job has no retry or timeout handling.) after a different response (Add bounded retries, timeout limits, and safe failure reporting appropriate to the operation.). For Frostline Distribution, it does not prove the required result: Two executions of the same input yield one accepted business side effect and a traceable retry record.
A staging execution writes only to staging objects and a production write requires the appropriate deployment context. This is suitable evidence for another condition (A data pipeline hard-codes a production catalog name in every notebook and developers accidentally write tests into production.) after a different response (Parameterize environment-specific object names and validate the resolved target at deployment time.). For Frostline Distribution, it does not prove the required result: Two executions of the same input yield one accepted business side effect and a traceable retry record.