Question 320 A team needs a job that repeatedly invokes a notebook for a known array of file partitions and wants built-in looping behavior. Which feature aligns? No dependencies and a single notebook invocation A For each task controlling a parameterized notebook step A table update trigger on an unrelated source One massive cross join of partition metadata Explanation Correct Answer B:For each provides iteration over array items without separately authoring a task for every partition. Incorrect Answers A:One unparameterized run does not perform the requested per-partition repetition. C:A trigger starts the workflow but does not iterate the partition list. D:Cross joins perform data operations, not orchestration loops.
Question 321 A job uses three parallel processing branches and a common final audit. The audit must see all three completion states even if one failed. Which dependency arrangement is correct? Depend on all three branches using All done Depend only on whichever branch finishes first Run the audit independently at job start Use All succeeded and expect failed branches to be reported Explanation Correct Answer A:This waits for each branch to finish and allows the audit to inspect failures and successes. Incorrect Answers B:Early audit would lack the remaining branch results. C:It cannot summarize branches that have not yet run. D:All succeeded suppresses audit execution after any upstream failure.
Question 322 A downstream report task is skipped although its source task succeeded; another required upstream task failed. Which explanation fits an All succeeded dependency? The report always runs after any one task finishes A skipped task proves SQL permissions are wrong Only one upstream success is needed for All succeeded Every declared upstream task must succeed, so the failed one blocks the report Explanation Correct Answer D:The All succeeded condition is conjunctive across required prerequisites. Incorrect Answers A:The declared multi-parent dependency requires all prerequisites. B:The skip may be caused solely by orchestration state, before SQL is executed. C:That is not the meaning of the all-success dependency.
Question 323 A periodic job is started by a cron schedule but also receives a manual Run Now from an operator. Which risk should be controlled? An automatic change to source table ownership Overlapping runs and repeated side effects when two initiations occur close together Assume a manual Run Now inherits an automatic lock that prevents any overlapping starts Every job asset becoming public Explanation Correct Answer B:Concurrency configuration and idempotent processing prevent duplicate updates across the scheduled and manual runs. Incorrect Answers A:Manual Run Now does not inherently reassign catalog owners. C:Manual initiation can overlap a scheduled run depending on job concurrency configuration; side effects need protection. D:Starting a job does not automatically change file visibility.
Question 324 A team is choosing between a notebook task and a SQL query task for a saved SELECT transformation. What criterion matters? Where the executable code is defined and which compute service should run it Select notebook task solely because the SQL transformation references a Unity Catalog table The length of the query author's username Choose the task solely from the size of the SQL result rather than where the code is defined Explanation Correct Answer A:A saved SQL query fits a SQL task; a notebook with Python/Spark logic fits a notebook task. Incorrect Answers B:Both task types can access governed tables when authorized; choose based on the stored code artifact and compute execution style. C:User-name length does not define task type. D:Row count may influence sizing, but the executable artifact and compute type determine whether to use a notebook or SQL task.
Question 325 A file-arrival-triggered ingestion workflow runs even when the file source contains only duplicates of already loaded objects. What must the final report distinguish? Gold must be overwritten on every notification All trigger events represent new business records Every successful ingestion run must append a nonzero row count Run initiated from event versus actual newly committed input records Explanation Correct Answer D:Trigger events prove orchestration started, not that the run inserted novel data. Incorrect Answers A:Unconditional overwriting can be unnecessary or harmful for empty batches. B:Duplicate notifications or repeated file contents can produce zero new logical records. C:A valid idempotent run may commit no new rows when nothing new exists.
Question 326 A workflow has A, B, C tasks with B and C depending only on A. B takes ten minutes and C takes one minute. When can C finish? It must finish after B finishes It must run before A It may finish shortly after A succeeds, independently of B Wait for B to succeed even though C has no explicit dependency on B Explanation Correct Answer C:B and C share A as prerequisite but have no dependency on each other, so C need not await B. Incorrect Answers A:No B → C edge was specified, so serialization is not required. B:C requires A success, so it cannot begin before A. D:A DAG allows B and C to proceed independently after A; waiting for B introduces a dependency absent from the configured graph.
Question 327 A job needs to use one upstream calculated statistic to pick a separate success path or exception path. Which feature set is most direct? Pass the statistic through an ordinary job parameter fixed when the run starts A file arrival trigger without conditional tasks A task value produced upstream and an If/else conditional task A cron schedule and unrelated notebook folder Explanation Correct Answer C:The task output supplies the branch condition, letting the DAG route processing by a real metric. Incorrect Answers A:A run-start parameter does not automatically capture a value computed later by a task for dynamic branching. B:The trigger starts a run but does not choose a branch from a computed statistic. D:Time scheduling cannot evaluate the calculated statistic.
Question 328 A developer needs to fix ETL code without changing the main production branch. Which Git workflow is appropriate in a Databricks Git folder? Edit production main directly without review Create a feature branch and commit the fix there Create a new Unity Catalog metastore Edit the production Git branch directly instead of isolating the fix Explanation Correct Answer B:A feature branch isolates the change until review and merge. Incorrect Answers A:Direct main edits eliminate the requested isolation. C:A metastore is for governance, not source-control branching. D:Direct changes to the primary branch bypass isolated review and increase the risk of accidental production code changes.
Question 329 An engineer finishes a notebook change on a feature branch and wants collaborators to review it before merging into main. What should the engineer do? Overwrite the main branch without review Grant SELECT on a data table Merge the feature branch without requesting review or recording its diff Push the branch and open a pull request Explanation Correct Answer D:A pull request provides a review and merge boundary for the proposed code change. Incorrect Answers A:This bypasses the required approval step. B:A data privilege cannot initiate source-code review. C:A direct merge skips the requested collaborative review, whereas a pull request makes the changes and checks visible.
Question 330 A feature branch has new commits locally in the Databricks Git folder, but the remote repository does not show them. Which action is missing? Increase Spark executor count Push the branch commits to the remote repository Run VACUUM on a Delta table Rename the Git folder without pushing Explanation Correct Answer B:Local commits are not automatically visible in the remote until pushed. Incorrect Answers A:Compute scaling cannot transfer source-control commits. C:Delta cleanup does not synchronize Git commits. D:A folder rename does not upload the missing commits.