Question 590 A CSV source delivers dates as `31/12/2026`, and a generic date parser treats the field as null. What should ingestion apply? An explicit source-aligned date format parse with invalid-value monitoring Only increase Spark executor count Discard the entire source without logging the error Only replace nulls with the current date Explanation Correct Answer A:The reader must interpret the documented day/month/year format rather than assuming a different default. Incorrect Answers B:More executors cannot fix an incompatible date format. C:Silent data loss is not a controlled data quality response. D:Fabricating dates masks parse failures and corrupts event chronology.
Question 591 A near-real-time payment event feed occasionally produces out-of-order updates for the same transaction. What must the target-state logic use? Only file modification timestamps An arbitrary alphabetical sort of transaction IDs An authoritative source sequence or event ordering field with tie handling Whichever event is ingested last regardless of source sequence Explanation Correct Answer C:Arrival order alone is unreliable when change events are delayed or replayed. Incorrect Answers A:Cloud file timestamps may not match logical source update order. B:Transaction identifier sorting does not define event chronology. D:A delayed older event can overwrite newer truth if arrival order is used blindly.
Question 592 A full-load snapshot contains a complete current customer population each night, including removed customers. Which target reconciliation must the design consider? Ignore missing keys forever Delete all target records not seen in the first ten rows Detect keys absent from the latest snapshot and apply the agreed delete or deactivate policy Only append every nightly snapshot into the serving table Explanation Correct Answer C:An authoritative full snapshot can imply removals through missing keys, not only through explicit change rows. Incorrect Answers A:Missing keys may represent genuine source deletions or deactivations. B:A partial sample is insufficient to infer absence from the complete source snapshot. D:Appending snapshots alone leaves old/deleted records appearing active without version or status handling.
Question 593 An event stream can deliver two different updates with identical event time but different source sequence numbers. Which tie-breaker is most defensible? Pick the update with the lexically longer customer name Use the source-defined monotonic sequence to order tied updates Use whichever record reached the driver first Keep both as independent current-state records for one entity Explanation Correct Answer B:A documented source sequence distinguishes events that timestamps cannot uniquely order. Incorrect Answers A:Name length is not a reliable event chronology key. C:Transport arrival timing does not guarantee source ordering. D:A latest-state table requires one resolved state per business key.
Question 594 A scheduled ingestion runs every hour, but one batch contained no new source files. What is the correct behavior for the target record count? Reload all history unconditionally to prove the job executed Delete existing records because the current batch is empty No new logical rows should be added solely because the schedule ran Append one placeholder row for every empty run Explanation Correct Answer C:An empty source batch should not manufacture data or replay previously accepted rows. Incorrect Answers A:Unconditional reload can duplicate business records and waste resources. B:No new input does not imply that previously accepted history should disappear. D:Placeholder records misrepresent actual source events.
Question 595 A regulated ingestion pipeline must retain corrupt source records for review while making curated silver unavailable until correction. Which layering choice fits? Discard corrupt input without recording file and row provenance Overwrite the bad raw bytes with fabricated default values Preserve recoverable raw payloads in a restricted bronze/quarantine path and gate promotion Expose invalid records directly in the gold business dashboard Explanation Correct Answer C:A recoverable source record and separate clean promotion boundary support auditability without contaminating standardized data. Incorrect Answers A:Silent deletion weakens reconciliation and incident handling. B:Invented replacements erase the evidence needed to correct the source. D:Unvalidated corrupted records should not be presented as trusted metrics.
Question 596 An API paginates invoices with `next_cursor`, and the cursor may be absent only after the final page. What condition should end ingestion? Exactly ten pages every time The first page with fewer than 100 rows regardless of API contract The first transient HTTP 429 The documented terminal cursor state after all returned pages are processed Explanation Correct Answer D:Completion is determined by the API's pagination contract, not by a guessed page count. Incorrect Answers A:A fixed page assumption can truncate large invoice results. B:Page size alone may not signal completeness when the API supplies an explicit cursor. C:Rate limiting is a retry condition, not a successful end-of-feed indicator.
Question 597 A REST source includes a continuation token that expires after several minutes. A batch job resumes the next day with the old token and fails. What should ingestion do? Repeatedly submit the same expired token forever Restart or refresh pagination according to the API's checkpoint and token validity contract Delete accepted Delta data to make the token work Use the Databricks notebook URL as the API token Explanation Correct Answer B:An expired token requires a documented resumption/re-scan strategy rather than blindly reusing stale pagination state. Incorrect Answers A:An invalid token is not made valid by retries without renewal. C:Data deletion cannot fix an API token expiry and risks loss. D:Notebook URLs do not replace source pagination credentials.
Question 598 A SaaS ingestion connector reports success, yet the source count and bronze count diverge. What is the most appropriate verification? Only increase the job cluster idle timeout Reconcile documented source scope, filters, cursor windows and accepted/failed row counts Assume connector success proves every source row arrived Check only the connector run status without reconciling the source extraction window and accepted rows Explanation Correct Answer B:Success status alone does not guarantee complete business coverage when source filters and error handling vary. Incorrect Answers A:Idle timeout does not explain accepted versus missing source rows. C:A successful call may still apply filters or exclude records according to configuration. D:A successful run can still omit filtered or delayed records. The discrepancy needs source-scope, cursor and rejection evidence.
Question 599 An organization has a collection of many small JSON files and asks whether to run one large COPY INTO daily or continuously ingest with Auto Loader. What criterion should drive the choice? The file names' alphabetical order only The required latency, discovery cadence and operational checkpoint behavior Choose based only on whether the target table has many columns Whether one developer prefers SQL syntax Explanation Correct Answer B:Batch versus incremental stream selection should reflect source arrival and freshness requirements, not tool branding alone. Incorrect Answers A:Lexicographic ordering is not a reliable basis for ingestion architecture. C:Schema width does not decide the ingestion cadence and restart state required by COPY INTO or Auto Loader. D:Personal preference alone is not enough to choose operational semantics.
Question 600 An ingestion job appends every day to a bronze table but must prevent an internal reader from mistaking incomplete daily input for finalized reporting data. What is the safer publish contract? Allow gold reports to assume every append event means all supplier files arrived Write a completion/control record and promote only reconciled daily loads to trusted consumers Have downstream reports read the bronze daily partition as soon as ingestion begins Assume a scheduled daily batch is complete as soon as the append job starts Explanation Correct Answer B:Explicit batch completeness evidence separates partial ingestion from accepted reporting snapshots. Incorrect Answers A:A write event cannot prove all expected deliveries are complete. C:A live bronze partition can be incomplete; the consumer needs a tested completion and reconciliation boundary before treating it as final. D:A started append is not a reconciled publication boundary; downstream readers can see partial or unapproved daily data.