Question 710 A gold view is created over a sensitive silver table. A requester has visibility into the view definition but no usable permission path to underlying data. What should be reviewed? The applicable view access and underlying authorization semantics for that execution context Review only the view’s SQL definition without checking its ownership and applicable source authorization Assume the view definition grants all underlying source privileges automatically Only the number of task retries Explanation Correct Answer A:A saved view name alone does not guarantee every principal can query protected data. Incorrect Answers B:A visible view definition is not proof that an executing principal has the necessary permissions in the current view security context. C:View visibility alone cannot establish the effective authorization path needed in the applicable execution context. D:Retries cannot fix a predictable missing permission for the data source.
Question 711 A source feed provides one modified customer each hour, not a complete customer table snapshot. Why is overwriting the entire silver customer table with each feed unsafe? The update contains a customer ID, so it automatically includes all others The feed arrival frequency determines which records are valid The incremental feed omits unchanged customers, which a full-table overwrite would delete The silver table cannot accept any updates after creation Explanation Correct Answer C:Partial updates should be merged into existing state under stable keys, not mistaken for full snapshots. Incorrect Answers A:One changed record does not represent every unchanged customer in the table. B:Hourly delivery cadence does not make a partial update equal to a full snapshot. D:Delta tables can support updates; the problem is interpreting partial feed completeness.
Question 712 A DataFrame grouped by department contains 7 rows, 2 with NULL manager_id. What do count(*) and count(manager_id) return for that group? 5 and 5 5 and 7 7 and 7 7 and 5 Explanation Correct Answer D:COUNT(*) counts all seven rows, while COUNT(manager_id) counts only five non-null manager values. Incorrect Answers A:COUNT(*) does not discard rows simply because a particular field is NULL. B:This reverses the normal SQL counting semantics for star and nullable columns. C:A column-level count excludes the two NULL manager IDs.
Question 713 Two source systems both use integer account_id=9 for different organizations. A unified silver table requires globally unique account identity. What key design resolves this? Use a compound source-system identifier plus source account ID Rely only on insertion timestamp equality Assign both records to the first source-system ID Use the raw account_id alone as a global key Explanation Correct Answer A:A namespaced composite distinguishes equal numeric IDs originating from different independent systems. Incorrect Answers B:Timestamps are not a stable globally unique business identity across source systems. C:Losing source origin falsely merges two organizations. D:The collision demonstrates that the numeric value is not globally unique across sources.
Question 714 A small dimension table includes decades of obsolete entries, but a fact report needs only current active mappings. What change improves join efficiency without changing business semantics? Replace the join with a Cartesian product Broadcast the entire fact table regardless of size Drop all entries from the dimension Filter the dimension to active relevant rows before the join and inspect the plan Explanation Correct Answer D:Removing irrelevant dimension rows reduces potential join work while preserving the specified current mapping. Incorrect Answers A:Cross joining increases combinations and does not implement the current-key lookup. B:A large fact broadcast can exhaust memory and does not address the irrelevant dimension rows. C:Deleting the dimension removes the reference mappings required by the report.
Question 715 A notebook task receives a job parameter named processing_date. The notebook must use that run-specific value rather than a hard-coded date. Which implementation reads it? Read a Spark executor process ID Assume the value is always the current UTC date Use a notebook widget and retrieve processing_date with dbutils.widgets.get Read the notebook file modification timestamp Explanation Correct Answer C:Notebook tasks expose key-value parameters to notebook code through widgets, allowing the value for this run to control processing. Incorrect Answers A:An executor ID identifies infrastructure, not the user-provided processing date. B:The job may be backfilling an older partition, so current UTC date would change the requested behavior. D:The code file timestamp is not the processing_date supplied for a specific job run.
Question 716 A Lakeflow Jobs Python script task is configured with an ordered argument array containing a source path and an output path. How should the script receive these inputs? Infer the paths by listing every cloud bucket Read the command-line arguments passed to the Python script Only read a Databricks SQL dashboard parameter Only query task values from a different task Explanation Correct Answer B:Script tasks use positional arguments, which the program can parse from its command-line argument vector. Incorrect Answers A:Source discovery cannot replace explicit ordered arguments supplied in the task configuration. C:Dashboard settings are not the Python script invocation arguments. D:Task values are for sharing upstream results, not automatically the script task's own positional inputs.
Question 717 A job defines a key-value parameter region=west and its notebook task defines region=east. Both use the same key. Under Lakeflow Jobs parameter pushdown, what value does that notebook task receive? The task-level value, east, always overrides the job Both values concatenated as west,east The job-level region value, west The value is always unset because names collide Explanation Correct Answer C:For key-value tasks the job parameter is pushed down and takes precedence over a same-named task parameter. Incorrect Answers A:The documented pushdown precedence favors the job-level value for matching keys. B:Parameter pushdown does not automatically concatenate duplicate key values. D:A collision produces a warning but does not mean the job-level value is discarded.
Question 718 A scheduled workflow normally loads yesterday's partition, but an engineer needs a one-time replay of 2026-08-14 without editing the recurring schedule. What should they use? Delete previous job runs from history Alter the Delta table's retention policy Permanently modify the default schedule date and leave it changed Run now with different parameters and set the processing_date override Explanation Correct Answer D:A manual parameter override scopes the backfill value to that job run while the schedule keeps its normal default. Incorrect Answers A:Deleting run records cannot choose which partition the next execution processes. B:Table retention is unrelated to choosing a one-time backfill date. C:Changing the recurring default creates risk for future automated runs.
Question 719 A workflow writes an audit record and must tag the record with the specific Lakeflow Jobs run that generated it. Which available metadata is the right identifier? Use the stable job ID instead of the run ID for the audit event The dynamically resolved job run ID The job display name alone The table row count after processing Explanation Correct Answer B:The run ID distinguishes one execution of the job from later or earlier executions. Incorrect Answers A:The job ID identifies the definition and cannot distinguish two executions of that job. C:The same job name is reused for many independent runs and cannot uniquely identify an execution. D:The count can be equal across distinct runs and does not uniquely identify execution.
Question 720 An upstream task produces a 20 MB DataFrame that a downstream job task must analyze. What is the appropriate handoff rather than trying to put the dataset into a task value? Persist the dataset to an authorized table or location and pass a small reference Serialize the entire DataFrame into one task-value string Serialize the full 20 MB DataFrame into a task-value string to avoid writing a durable dataset Use an email attachment as the data exchange path between dependent job tasks Explanation Correct Answer A:Task values are for small orchestration metadata; durable storage is appropriate for a large dataset. Incorrect Answers B:Task-value size limits and orchestration design make a large DataFrame an unsuitable task value. C:Task values are designed for small JSON-serializable metadata, not large distributed datasets; a stored table plus reference is appropriate. D:Notifications are for operational communication; they do not provide a governed in-platform dataset reference for downstream processing.