Question 580 A supplier quotes fields with embedded line breaks in CSV descriptions. The initial ingest splits a single invoice into multiple records. What should be investigated? CSV multiline and quoting settings for the actual supplier format Only renaming the description column Only increasing cluster idle timeout Only lowering Delta retention Explanation Correct Answer A:Embedded newlines inside quoted fields require reader options compatible with the source CSV dialect. Incorrect Answers B:Column aliases cannot recover rows split at incorrect quoted line breaks. C:Idle timeout does not fix row boundary parsing. D:Retention does not make malformed row boundaries valid.
Question 581 A JSON ingestion job now receives records where a field varies between an object and a string. The business cannot discard the malformed cases. What is the safer handling choice? Capture unexpected or incompatible source content in a recoverable record path for review Force every new object into an unrelated numeric field Overwrite the entire target with an empty table Drop every record that differs from yesterday's schema without any evidence Explanation Correct Answer A:Preserving the conflicting payload allows schema governance and remediation instead of silent data loss. Incorrect Answers B:Incorrect coercion loses the incoming structure and meaning. C:An incompatible field does not justify destroying previously accepted data. D:Silent drops prevent investigation of a genuine upstream contract change.
Question 582 A source system occasionally sends duplicate change events with the same immutable event ID. The bronze layer must support replay without multiplying logical changes. What metadata is useful? Use the Databricks job run ID as a substitute for source event identity Discard all events that arrive in the same minute Use current ingestion time as an always-unique logical event key Persist the source event ID and deduplicate only under the correct event identity semantics Explanation Correct Answer D:A durable event identifier lets ingestion identify replayed deliveries without assuming file names are transaction IDs. Incorrect Answers A:A new job run can reprocess the same source event; run IDs do not identify business events. B:Time-window deletion can remove legitimate distinct source changes. C:Repeated source events can arrive at new times, so arrival timestamps do not establish identity.
Question 583 A file ingestion process begins writing its checkpoint under an ephemeral cluster-local directory. What happens when a new job cluster is started? The cluster-local path is automatically persisted across all job clusters Delta transaction logs are automatically deleted The checkpoint may not survive, causing loss of reliable source progress continuity Every old file becomes permanently inaccessible in cloud storage Explanation Correct Answer C:Checkpoint state should be durable across job-cluster replacement for resumable streaming ingestion. Incorrect Answers A:Ephemeral local paths are not a reliable shared checkpoint location. B:Checkpoint location is separate from the target Delta log and does not imply its deletion. D:Cloud object files do not disappear solely because a local checkpoint is missing.
Question 584 Two independent Auto Loader jobs point to the same input prefix and write to the same output table using different checkpoints. What ingestion risk should be considered? Checkpoints from different jobs automatically merge atomically Targeting Delta automatically enforces unique business keys Each job may process the same files independently and create overlapping output records The jobs cannot read the same object storage path Explanation Correct Answer C:Progress is scoped by checkpoint; two independent consumers are not automatically coordinated as one logical writer. Incorrect Answers A:Independent progress stores do not inherently share a deduplication contract. B:Delta transactions do not by themselves impose row-level business-key uniqueness. D:Multiple readers can often access the same files; correctness of their combined writes is the issue.
Question 585 A streaming file source is replayed from the beginning into an existing target. The engineer expects no duplicate business records solely because the target is Delta. What must be added? An additional job schedule label An OPTIMIZE command before replay An explicit replay-safe keying, deduplication, or merge strategy for the business records Reuse the same streaming checkpoint but overwrite the old output table with raw appended records Explanation Correct Answer C:Transactional writes alone do not guarantee uniqueness across deliberate source replay. Incorrect Answers A:Changing scheduling metadata does not enforce row uniqueness. B:Compaction can reorganize files but does not remove logical duplicate rows. D:A checkpoint tracks source consumption, not logical business-key uniqueness when an existing target is replayed.
Question 586 An upstream publisher writes a file in chunks to an object path, and the ingest job sees the object before the last chunk is finalized. Which contract should be established? Only lower the target Delta log retention Skip all files that are larger than one megabyte Publish immutable completed files atomically or signal readiness with a validated manifest Begin reading all file names as soon as they appear regardless of completion Explanation Correct Answer C:Ingestion should discover only finished objects to avoid loading incomplete source deliveries. Incorrect Answers A:Target retention does not establish source file completion. B:File size alone does not reliably distinguish complete and incomplete deliveries. D:Premature discovery can capture partial data and cause inconsistent records.
Question 587 A partner republishes a corrected file under the exact same cloud key after an Auto Loader job has already processed the original. What is the central risk? File mutation may not behave like a new immutable delivery and needs a supported correction policy Every overwrite is guaranteed to be interpreted as an independent new object Only schema column ordering can be affected Delta automatically rewrites all dependent gold tables Explanation Correct Answer A:File-discovery state is designed around source object identity and progress; overwrites require explicit source contract handling. Incorrect Answers B:Reusing source object identity is not equivalent to delivering a new immutable file. C:A corrected file can change substantive records, not merely field ordering. D:Changes in source object bytes do not ensure downstream transformations replay correctly.
Question 588 An invoice CSV has `invoice_id` values with leading zeros, but the ingest schema casts them to integers. What should the ingest design do? Always cast invoice IDs to DOUBLE Preserve the identifier as a string when leading zeros carry business meaning Discard invoices whose IDs start with zero Rely on an ORDER BY to restore the zeros later Explanation Correct Answer B:Numeric coercion can erase significant formatting and prevent reliable joins to upstream identifiers. Incorrect Answers A:Floating-point representation can further distort identifiers. C:Leading zeros do not inherently make an invoice invalid. D:Row ordering does not recover characters removed during casting.
Question 589 A feed supplies monetary amounts such as `12.39`, and the downstream ledger requires exact cents. Which raw-to-standardized conversion is appropriate? Use FLOAT as the only persistent financial representation Convert every amount to a Boolean Convert to a fixed-precision decimal with an agreed scale and validate parsing Only sort the text lexically Explanation Correct Answer C:Decimal types avoid representing currency as an inexact binary floating-point value for ledger calculations. Incorrect Answers A:Binary floating point can introduce rounding artifacts for exact monetary reconciliation. B:A Boolean cannot represent the needed currency values. D:String ordering does not produce reliable monetary values or arithmetic.
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.