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.
Question 71 A streaming JSON ingestion uses Auto Loader with addNewColumns. A previously observed column changes from integer to an incompatible string format. What should the team avoid assuming? That a Spark checkpoint can store ingestion state That schema inference will always coerce incompatible incoming strings to the existing integer type That default new-column evolution automatically fixes arbitrary existing-column type changes That every JSON record is always typed without interpretation Explanation Correct Answer C:addNewColumns handles new fields; incompatible existing field types require deliberate handling or rescue. Incorrect Answers A:Checkpoints do store state, so removing them would make replay risk worse rather than fix type mismatches. B:An incompatible type change cannot be assumed to cast safely under addNewColumns; rescue or deliberate type handling may be required. D:JSON sources can contain inconsistent field types that demand parsing and quality decisions.
Question 72 An ingestion pipeline reads new files from a Unity Catalog volume. The compute identity has table write privileges but receives an authorization error when accessing the volume. What must be checked? Relevant storage/volume access privileges for the source path Add SELECT on the destination table without granting access to the protected source volume Change Delta file layout but leave the denied source volume privileges unchanged Table SELECT privileges on the target without read authorization on the source volume Explanation Correct Answer A:Target table writes do not automatically confer read privileges on a separately governed source volume. Incorrect Answers B:Destination table access cannot satisfy a missing governed source-path privilege used to read incoming files. C:Optimization controls data layout and cannot grant a principal permission to read protected input objects. D:A principal can write a table while lacking the separate volume read privileges needed to retrieve the input files.
Question 73 A source is a partner application available only through an HTTPS API. The target is a governed table, and extraction must happen nightly. What orchestration pattern fits? Use Auto Loader directly against a URL returning paginated JSON responses Use a file-arrival trigger pointed at the API URL without writing a paginated extraction client Call the API with authenticated paginated extraction and schedule the loading task in Lakeflow Jobs Ask the dashboard to import HTTP responses Explanation Correct Answer C:A scheduled API extraction followed by governed target writes handles the source and cadence requirements. Incorrect Answers A:Auto Loader discovers cloud files rather than serving as a general paginated API client. B:Lakeflow file-arrival triggers monitor supported storage locations, not arbitrary paginated HTTPS APIs. D:Dashboard query refresh is not the ingest mechanism for arbitrary external APIs.
Question 74 A source offers a supported Lakeflow Connect connector and an ad hoc custom JDBC script. The company needs managed source-specific extraction with fewer custom maintenance tasks. Which choice is stronger? Prefer the supported Lakeflow Connect connector after validating its capabilities Build a connector-like notebook using generic full-snapshot JDBC extracts every hour Only create a Unity Catalog user group Write a JDBC script without change tracking or retries by default Explanation Correct Answer A:A supported managed connector can reduce custom code and handle source-specific ingestion patterns. Incorrect Answers B:Repeated full extracts can move data but do not inherently provide the managed source-aware incremental behavior or lower maintenance burden requested. C:An identity group governs access and does not ingest external application changes. D:Unmanaged extraction increases ownership burden and may not meet reliable incremental needs.
Question 75 A one-time local CSV file needs to be imported into a governed table for a small pilot. What is the relevant ingestion path? Upload the source to an approved governed location and load it into a Delta table Treat the local desktop file as if all jobs can read it forever Revoke target table CREATE/WRITE rights to enable import Create a job trigger without making the file available Explanation Correct Answer A:A small local file can be staged under governed storage and imported using an appropriate file-loading path. Incorrect Answers B:A personal laptop path is not a durable shared source for scheduled cloud processing. C:Removing necessary permissions would obstruct rather than enable loading. D:A trigger cannot read a file that the compute cannot access.
Question 76 An ETL owner wants to know whether an ingestion job loaded 15 source objects this run and whether some were skipped as previously processed. Which evidence is most relevant? Ingestion operation metrics and source-file tracking Inspect the latest destination table row count without the ingestion command results Inspect the source bucket inventory but not the job load or commit metrics Compare only current and prior job start timestamps Explanation Correct Answer A:The load result and ingestion state distinguish newly processed files from ones already recorded. Incorrect Answers B:A destination count cannot distinguish source files loaded this run from previously processed files and skipped input objects. C:A list of objects establishes availability but cannot show which files the current COPY INTO or Auto Loader run actually committed. D:Schedule timestamps do not reveal the processed/skipped source file identities or committed row metrics.
Question 77 A company receives records in a CSV file where commas appear inside quoted string fields. What should the reader configuration address? CSV parsing options that honor quoted fields and delimiters Split each line at every comma without applying quote-aware parsing Drop any row containing a comma Change every source record into an image Explanation Correct Answer A:Correct CSV quote handling prevents embedded delimiters from being mistaken for column boundaries. Incorrect Answers B:Naive delimiter splitting wrongly treats commas inside quoted fields as separators and corrupts valid records. C:That rule can discard legitimate quoted content and is not a valid CSV parser. D:Images destroy the structured fields the pipeline needs.
Question 78 A Parquet source contains typed columns, but ingestion unexpectedly treats a numeric column as text after a custom cast. What should the engineer inspect? Assume the original Parquet physical type always remains unchanged after any SQL transform Only the number of retries configured The explicit schema and cast transformations applied during loading The notebook repository name Explanation Correct Answer C:Parquet carries types, but explicit ingestion transformations can override the resulting column type. Incorrect Answers A:Parquet types are recorded, but an explicit downstream cast can change the resulting type and should be inspected. B:Retry policy does not convert numeric values to strings. D:Git folder names do not determine the parsed Parquet schema.
Question 79 An upstream billing API can return duplicate pages after a timeout, and the ingest job may retry a request. What protects the Delta target from duplicate business records? Restart extraction from the first page and append the replay without comparing stable event identities Combine resumable extraction with deterministic event IDs and idempotent target writes Append every returned page blindly after each retry Persist the API cursor but always append every returned page without a record key Explanation Correct Answer B:An extraction cursor plus key-based deduplication or merge handles retries without double counting. Incorrect Answers A:Replaying pages without key-aware target writes can duplicate data when the source API retries or returns already committed events. C:A retry can repeat a page and create duplicate business transactions. D:A saved cursor supports extraction progress but does not prevent duplicate business records if a page is retried after the target commit.
Question 80 A source system publishes a full snapshot rather than changes. The downstream goal is to retain only the latest state keyed by customer ID. Which operation is needed after ingestion? VACUUM with zero retention after every run Unconditional append of every full snapshot Append each complete snapshot with an ingestion-time timestamp but do not reconcile customer IDs Key-aware replacement or MERGE of the latest snapshot into the curated table Explanation Correct Answer D:A latest-state target must reconcile existing keys rather than blindly append all snapshot rows. Incorrect Answers A:VACUUM is storage cleanup, not business-key state reconciliation. B:This accumulates outdated versions and does not by itself represent current customer state. C:Snapshot history can be stored separately, but a current-state target must select or reconcile one latest row per business key.