Question 610 A partner supplies records as a Parquet dataset partitioned by `year` and `month`, and an analyst notices the partition columns do not appear as ordinary fields in each file. How should the reader recover them? Read the partitioned root using supported partition discovery and validate directory naming Assume partition columns must be manually inserted into every Parquet row Turn off schema reading entirely Delete every directory containing an equals sign Explanation Correct Answer A:Spark can infer partition columns from appropriate directory structure and path context, even when those values are not repeated inside each file. Incorrect Answers B:Well-formed partitioned datasets do not require duplicating partition keys in each data file. C:Parquet schema and partition discovery are needed to interpret the dataset reliably. D:Hive-style partition folders use key=value names as metadata rather than invalid files.
Question 611 A daily supplier feed is partitioned by date, but the downstream consumer needs corrections from the prior week. An engineer reads only today's partition. What is missing? The ingest window or change feed must include late corrections to older partitions Only a larger SQL dashboard query limit Assuming older source dates are immutable without supplier agreement Increase the lookback in the reporting query while leaving previously dated source partitions unprocessed Explanation Correct Answer A:A date-only arrival filter can exclude amended historical business records that arrive after their nominal partition. Incorrect Answers B:Dashboard row limits cannot retrieve files never ingested. C:Correction policies must follow the source contract rather than an unsupported assumption. D:A report cannot recover corrections that ingestion never loaded; the source window or change capture must include historical amendments.
Question 612 An API delivers customer records using offset pagination while new records are inserted during the fetch. What consistency hazard arises? Offsets can shift between requests, causing missed or repeated records without snapshot stability Only Delta transaction metadata can change Offset pagination guarantees an immutable source snapshot The file format will automatically switch from JSON to Parquet Explanation Correct Answer A:Concurrent source changes can alter page membership during offset pagination. Incorrect Answers B:The pagination hazard occurs at the external source before Delta writes. C:Page offsets do not by themselves freeze a changing upstream dataset. D:Source concurrency does not determine output serialization format.
Question 613 A partner API provides a stable snapshot token for a large export. Why is using it across all pages preferable to mixing live queries? It guarantees every field is free from source errors Use a new live snapshot for each successive page and rely on an ingestion checkpoint Store only the last source page ID without retaining the stable export snapshot token It keeps pages in one consistent source snapshot boundary Explanation Correct Answer D:Stable snapshot pagination reduces omitted or duplicated records caused by concurrent source mutation. Incorrect Answers A:Snapshot stability is not proof of row-level data quality. B:A checkpoint records progress, but changing source snapshots between pages can introduce gaps or duplicates in the exported dataset. C:Page position without the same source snapshot cannot ensure that later pages belong to the consistent dataset initially extracted.
Question 614 An ingestion pipeline receives a file containing a valid header but zero records. Business owners define this as a valid no-data delivery. What is appropriate? Create a fabricated record to avoid an empty table Mark the delivery missing without examining the header and contract Record successful empty delivery separately from missing or failed delivery Delete all prior target rows immediately Explanation Correct Answer C:A zero-row file may satisfy the source contract and should not be confused with file absence. Incorrect Answers A:Inventing business rows would violate source truth. B:Empty valid and missing delivery are distinct states. D:No-data delivery does not necessarily imply full source deletion.
Question 615 A supplier delivers two batches with the same filename to different dated directories. What must an ingestion idempotency key consider? Only the last three filename characters The full source object identity and relevant delivery version rather than the basename alone Only the first invoice ID in each batch Only job start time Explanation Correct Answer B:Identical basenames in different paths may represent distinct legitimate deliveries. Incorrect Answers A:A suffix does not uniquely identify a delivery. C:Different deliveries may contain overlapping business IDs without being duplicate files. D:Job starts do not correspond uniquely to supplier file objects.
Question 616 An ingest job receives rows whose primary key is null despite the source contract requiring it. What is the correct handling at a trusted promotion boundary? Drop the primary-key column from the entire trusted table Quarantine or reject invalid-key rows with traceable validation outcomes Make each null key equal to the job run ID Coerce every missing key to the same fixed value Explanation Correct Answer B:The trusted target should not silently accept records that violate its identity contract. Incorrect Answers A:Removing the key defeats downstream entity joins. C:A run identifier does not recover the original business entity identity. D:A common fabricated key creates false collisions and loses identity.
Question 617 An ingestion source includes records with the same customer ID but a source `record_version` field. The target represents current state. What should determine the winner? Only whichever record arrives at the warehouse fastest The highest valid source version subject to the documented update and delete rules The lexically shortest customer name The first row encountered in a nondeterministic Spark partition Explanation Correct Answer B:A source-defined version allows the target to identify the authoritative latest state. Incorrect Answers A:Network timing can reorder older and newer versions. C:Name length does not specify version ordering. D:Read order is not a reliable current-state rule.
Question 618 A compressed JSON feed uses a character encoding different from UTF-8, and Spark reads accented names incorrectly. What should the source contract specify? Only the ZIP file size The actual character encoding and appropriate decoding behavior Only the number of executor cores Only the target table sort order Explanation Correct Answer B:Byte decoding must match the source's encoding to preserve names and other text fields. Incorrect Answers A:Compressed size does not specify text encoding. C:Parallelism does not repair wrong text decoding. D:Ordering cannot recover characters corrupted during decoding.
Question 619 A JDBC extraction reads a large source table while the source applies many updates. The extract must represent one consistent cut. What is the concern? The source transaction isolation or snapshot consistency of the extraction Only cloud object storage compression Increase JDBC fetch size without validating the source snapshot or transaction isolation Raise Spark partition count without controlling the consistency of the JDBC source snapshot Explanation Correct Answer A:Without a consistent source read boundary, a multi-part extract may combine different source states. Incorrect Answers B:Compression cannot align source rows to one transactional view. C:Fetch size changes data transfer granularity, not consistency of a long-running extraction in the presence of concurrent source updates. D:Spark parallelism does not guarantee one consistent source database view across concurrent writes during extraction.
Question 620 A source database has an update timestamp but no guarantee that every row change updates that timestamp. Is a watermark-only incremental extract safe? Yes; target Delta ACID fills in changes the source never emitted Yes; any timestamp column provides a complete CDC feed Yes; increasing the hourly job frequency eliminates missing updates No; the source change indicator must be complete or another capture mechanism is required Explanation Correct Answer D:A watermark that misses some updates cannot guarantee the target sees all changed records. Incorrect Answers A:Target transactions cannot invent missing upstream events. B:An unreliable timestamp does not necessarily change on each source mutation. C:More runs cannot detect source changes not reflected in the extraction predicate.