Question 100 A team wants to ingest a proprietary source via Lakeflow Connect but the required connector is not supported. What is the practical next assessment? Evaluate an approved partner connector or authenticated custom extraction path Configure a connector for an unrelated supported source and assume it can read the proprietary source Assume any managed connector can connect to every database Assign the extraction principal broader Databricks workspace permissions without implementing source authentication Explanation Correct Answer A:Connector selection must respect actual supported sources; partner or custom integrations can address gaps. Incorrect Answers B:Connector support is specific to each source system; choosing a different vendor connector cannot create compatibility with an unsupported source. C:Managed connectors support defined source systems rather than all arbitrary endpoints. D:Workspace role changes cannot create support for an unavailable source connector or replace a source-specific authenticated extraction method.
Question 101 A source emits daily data partition files with occasional re-deliveries. An operator wants to know which files were actually imported after a failed job. Where should evidence come from? The number of source bucket folders Use a raw list of object names without the target load history Compare only source object modification times but ignore the destination ingestion state Load operation history and file-level ingestion state for the affected target Explanation Correct Answer D:A recorded ingested-file state distinguishes committed files from those awaiting retry. Incorrect Answers A:Folder count is not proof of successful record loading. B:Source inventory alone does not prove which objects were actually committed to the target in the failed/retried run. C:Cloud file timestamps alone cannot establish which files were previously processed or skipped by the target loader.
Question 102 A stream must ingest new cloud objects but retain source event timestamps instead of overwriting them with processing time. What is the right design? Replace the source event timestamp with the cloud-object ingestion timestamp Order the source files by their cloud modification time and infer that as every event timestamp Replace all event timestamps with the current run timestamp Parse and preserve event-time fields separately from ingestion-time metadata Explanation Correct Answer D:Source event time and processing time answer different lineage and watermarking questions. Incorrect Answers A:Arrival time may lag the actual business event; overwriting source event time prevents correct event-time analysis. B:Object timestamps describe file handling, not necessarily when each contained business event occurred; preserve the real event-time column. C:Doing so destroys original event chronology needed for late-arrival analysis.
Question 103 A JSON feed contains a price value as a string. The data must land in bronze even when the string is malformed, but silver requires a numeric price. Which boundary is correct? Convert all malformed prices silently to zero without a quality record Discard all source files containing any string prices Land source value, then apply validated numeric conversion and quarantine invalid rows in silver Grant users broader table rights to fix the type Explanation Correct Answer C:Bronze source fidelity and silver type-quality enforcement serve different stages of processing. Incorrect Answers A:A zero can misrepresent a missing or corrupt price as valid business data. B:String-encoded numbers may be valid after controlled parsing; wholesale deletion loses data. D:Privileges do not correct malformed source values.
Question 104 An organization receives very small CSV batches once per month and already has a scheduled SQL run. It wants the fewest moving parts for idempotent file ingestion. What should be considered? A scheduled COPY INTO operation Weekly recreations of the UC metastore A custom always-on REST polling service A continuous Spark streaming job with no checkpoint Explanation Correct Answer A:COPY INTO is a simple built-in SQL file-loading pattern for repeatedly loading only new immutable files. Incorrect Answers B:Recreating the catalog is unrelated and destructive to governance. C:A perpetual custom poller would be unnecessary for a monthly object-file batch. D:A perpetual stream without progress state is not a low-maintenance safe alternative.
Question 105 An Auto Loader source produces an added CSV column. An operator wants to know why the stream stopped under default schema evolution rather than treating it as a fatal design error. What explanation applies? That a new column always requires manually dropping the destination and recreating the stream Auto Loader can update the tracked schema and require a restart to continue with the new column All new CSV columns are always ignored permanently Schema evolution automatically deletes earlier Delta rows Explanation Correct Answer B:The default addNewColumns flow records the schema addition, raises an exception, and resumes after restart. Incorrect Answers A:In addNewColumns mode, tracked schema metadata is updated and a restart can resume; the existing table and checkpoint need not be discarded. C:Default schema inference and evolution can add columns rather than silently discarding them. D:Introducing an additional source column does not require deleting prior target data.
Question 106 A team uses multiple independent ingestion jobs to append records into the same Delta table. What should it confirm to avoid confusing file-level idempotency with business-level uniqueness? That Delta transaction atomicity automatically deduplicates equal records That warehouse autoscaling can choose the latest entity version That every distinct source filename implies distinct business events Whether different jobs share reliable record IDs and idempotent upsert/dedup logic Explanation Correct Answer D:Tracking files within one ingestion mechanism does not by itself ensure global unique business records across jobs. Incorrect Answers A:ACID commits do not mean logically duplicate rows are removed automatically. B:Compute scaling is not entity-level reconciliation. C:The same event can appear in several files, so filenames alone are not business keys.
Question 107 A source bucket has a manifest with 2,000 expected files, but the Delta target load reports 1,998 successful new files and no skips. Which action best supports accurate closure? Add two empty Delta rows to match the count Reconcile the two missing manifest entries before declaring completeness Replace the manifest count with the observed count Declare success because most objects were loaded Explanation Correct Answer B:A manifest-to-load discrepancy is evidence of potentially missing data and requires follow-up. Incorrect Answers A:Artificial rows are not replacements for missing source files or their records. C:Changing expected evidence does not explain the unaccounted source objects. D:A stated completeness requirement cannot be waived by a high percentage alone.
Question 108 An engineer reruns an ingestion notebook after an API timeout and finds duplicate transactions with the same invoice ID. Which repair targets the root cause? Disable Delta commit history Use invoice-key-based deduplication or merge and a durable extraction cursor Add more source-file copies with different names Increase compute size without adding an idempotent target key or extraction cursor Explanation Correct Answer B:Business keys and source progress together make retries safe for logical invoices. Incorrect Answers A:Removing history does not solve duplicated logical transactions. C:More copies can increase duplicate invoice records. D:Additional compute capacity cannot prevent duplicate business events caused by repeated API pages after a timeout.
Question 109 An organization has a high-volume cloud source in which files arrive every few seconds. It wants to avoid repeated full folder scans where file notifications are supported. Which Auto Loader discovery mode is a candidate? One daily manually entered list of new names Poll the same full cloud directory repeatedly even when file notifications are available File-notification discovery for the supported cloud storage Unconditional COPY INTO over all files on each second Explanation Correct Answer C:Notifications can support scalable detection of new objects without repeated exhaustive listings. Incorrect Answers A:Manual discovery is impractical for high-frequency sources. B:Periodic directory listing can find arrivals but adds unnecessary scans at high volume compared with supported file-event discovery. D:Repeated polling by SQL commands is unnecessarily intensive for a frequent event-driven feed.
Question 110 An upstream API can produce either snapshot or CDC records. The target is a historical append-only audit ledger rather than the latest current state. Which ingestion model fits? Append immutable events with a stable source event ID and timestamp Overwrite the historical target for every new source event Truncate the target before every batch MERGE away every historical event for the same entity Explanation Correct Answer A:An audit ledger retains individual change events rather than overwriting them into a current-state projection. Incorrect Answers B:A latest-state overwrite discards prior change events and contradicts the immutable audit-ledger requirement. C:Deleting past events contradicts append-only audit retention. D:Keeping only one current entity row would erase the requested audit chronology.