Question 60 A source database publishes frequent changes that must reach Databricks incrementally without full-table copies every hour. What ingestion principle best fits? Reimport the entire database and append it unconditionally every hour Re-export the entire database table on every hourly schedule Use a supported change-aware incremental ingestion/CDC mechanism Use a recurring full-table export but de-duplicate it after loading every complete snapshot Explanation Correct Answer C:Change-aware loads capture source inserts/updates/deletes and avoid unnecessary full snapshots. Incorrect Answers A:Blind full snapshots introduce repeated rows and unnecessary work. B:A repeated full load can work in some settings but does not meet the stated efficient change-aware incremental ingestion goal. D:A reconciliation process can address duplicates but still repeats the full source copy and does not capture only changed rows as requested.
Question 61 A team uses a JDBC connection to import a large source table nightly. It wants subsequent runs to fetch only records with newer update timestamps. What additional logic is needed? Increase JDBC fetch size while still selecting all source rows on each run Maintain a reliable watermark and select records beyond the committed checkpoint Read every JDBC source row and filter changed timestamps only after extraction Use a new random filter on each run Explanation Correct Answer B:A tracked source watermark supports bounded incremental queries instead of full-table pulls. Incorrect Answers A:Fetch size can affect throughput but does not apply the update-timestamp predicate needed for incremental extraction. C:Filtering after transmission still pulls the full source table rather than using a source-side incremental timestamp predicate. D:Random filters are not stable evidence of source changes and can lose records.
Question 62 A REST API returns one page of 100 records with a next-page token. The pipeline must ingest the complete extraction. Which logic is required? Delete the token and restart the first request repeatedly Follow pagination tokens until the API indicates no further pages Assume one HTTP 200 response contains the complete dataset Stop after reaching a fixed estimated row count without checking whether the API returns a continuation token Explanation Correct Answer B:Without pagination, records beyond the first page would be silently omitted. Incorrect Answers A:Repeated first-page calls would not retrieve later pages. C:A successful first page does not imply there are no subsequent pages. D:An estimate is not evidence that every API page was fetched; source-provided tokens identify additional pages.
Question 63 A source API occasionally returns HTTP 429 with retry guidance during scheduled ingestion. Which behavior best avoids accidental permanent data gaps? Increase the Unity Catalog table owner privileges Honor rate-limit/backoff guidance and resume extraction from the last committed cursor Delete the checkpoint before every retry Treat all HTTP 429 responses as successful empty batches Explanation Correct Answer B:Controlled retries and a persisted cursor protect completeness during transient throttling. Incorrect Answers A:Table authorization is unrelated to a source service rejecting excessive requests. C:Resetting progress can cause duplicates and does not solve API rate limits. D:This can mark missing source records as already processed.
Question 64 A Python notebook uses JDBC to read source rows and then writes to a UC Delta table. Which distinction matters for permissions? Successful JDBC SELECT automatically grants all Databricks table privileges The JDBC source credentials and Unity Catalog target permissions are separate requirements Configure a single SQL warehouse service identity and assume it grants JDBC access Only target table privileges are needed even if the source rejects logins Explanation Correct Answer B:Reading the external source does not imply authorization to write the governed destination table. Incorrect Answers A:Source database access never grants unrelated Unity Catalog MODIFY rights. C:Warehouse execution identity and source database authentication are independent; permissions for the external JDBC source remain necessary. D:The extraction also needs valid authorization to the JDBC source.
Question 65 A batch ingestion job has multiple input files, one malformed and the others valid. The owner wants malformed records investigated without pretending that all records landed correctly. Which approach is defensible? Select permissive CSV parsing but discard every malformed record without retaining failure evidence Treat the job as successful whenever at least one input file parses, without reconciling rejected rows Skip malformed source files while reporting the full expected row count as loaded Capture parse-error or rescued-data evidence and enforce a documented quality disposition Explanation Correct Answer D:Error preservation and explicit quality handling prevent silent loss of malformed source content. Incorrect Answers A:Permissive parsing can assist ingestion, but silently dropping malformed records cannot satisfy the stated need to investigate and account for them. B:A partially successful load cannot be called complete without recording and addressing malformed records or rejected files. C:Skipping bad files can be an explicit disposition in some pipelines, but reporting all rows landed when source records were skipped is misleading.
Question 66 A team chooses between COPY INTO and Auto Loader for a constantly arriving, large file stream. What factor most directly drives the decision? Whether the source contains parquet files and therefore always requires a continuous stream Ingestion frequency, discovery scale, and orchestration/streaming needs Whether the job uses Python or SQL exclusively regardless of event cadence Choose ingestion based only on whether the file names begin with the same prefix Explanation Correct Answer B:Auto Loader is oriented to scalable incremental file discovery; COPY INTO is a convenient retriable SQL file load. Incorrect Answers A:File format alone does not determine streaming versus batch architecture; simple SQL COPY INTO can load Parquet files when the cadence permits. C:Supported APIs and languages matter, but the decision also depends on arrival scale, checkpointing, and required freshness. D:A naming convention does not establish whether discovery scale, event frequency or checkpointed streaming requires Auto Loader.
Question 67 An Auto Loader source has millions of objects and frequent arrivals. The platform supports managed file-event notifications for that location. What should the team evaluate? Frequent recursive directory listing without using available cloud file-event notifications A fixed daily manifest that includes only objects known when the job was written Run a daily rename of source objects to make each file appear newly modified File-notification-based discovery for scalable incremental ingestion Explanation Correct Answer D:Event-based discovery can avoid repeatedly listing vast object namespaces when appropriately supported. Incorrect Answers A:Full listings can find new files but add inventory overhead at the stated scale instead of using the supported notification-based path. B:A static file manifest cannot discover frequently arriving objects without manual refreshes and is a poor fit for continuous additions. C:Renaming every file creates unnecessary ingestion work and can produce duplicate discovery rather than using supported file events for new arrivals.
Question 68 A source publishes one new Parquet partition each evening. The pipeline may run just after delivery and should not reload past partitions. What is a sound simple implementation? Use idempotent COPY INTO with the stable source directory and Delta target Append all files from the entire directory blindly each evening Truncate the table before each run regardless of downstream history Reread every historical Parquet partition and rely on downstream analysts to deduplicate Explanation Correct Answer A:COPY INTO tracks files loaded in previous executions, making scheduled incremental file loads straightforward. Incorrect Answers B:Unconditional append will duplicate old partitions on every run. C:A full reset is unnecessary and may remove retained data needed by analysts. D:Repeated full reprocessing adds unnecessary work and can duplicate target records instead of using file-level load tracking.
Question 69 An ingestion team is deciding whether to use a scheduled batch load or a continuously updated stream. Which explicit business requirement matters most? Whether the storage bucket name contains a date Only the source file format, ignoring allowed end-to-end freshness latency Allowed data freshness latency and source arrival pattern The preferred chart palette of the analysts Explanation Correct Answer C:Freshness needs and arrival cadence determine whether scheduled batches are sufficient or streaming is useful. Incorrect Answers A:Naming convention alone does not establish arrival timing or freshness SLA. B:Batch and streaming both support common formats; the SLA and arrival cadence determine whether continuous processing is justified. D:Visualization colors do not decide ingestion timing.
Question 70 A raw file has two new optional JSON fields that the analytics team is not ready to model. They still want to keep the unknown fields in a recoverable column. Which Auto Loader approach matches? Drop the raw source objects after parsing Use none without a rescued column Use failOnNewColumns and permanently stop ingestion Select rescue schema evolution with a rescued-data column Explanation Correct Answer D:Rescue preserves unexpected fields rather than silently dropping them or immediately adding them as modeled columns. Incorrect Answers A:Deleting source evidence contradicts the recoverability requirement. B:Without configured rescue behavior, new fields may be ignored. C:A hard stop is not necessary when the requirement is to preserve fields for later interpretation.