Question 450 A job succeeds when executed alone but fails on memory when many identical jobs overlap. What should be examined? Only the column names in the input schema Concurrency, shared compute capacity and per-job memory requirements Only notebook comment style Only dashboard formatting Explanation Correct Answer B:Overlapping executions can create resource contention that is absent in isolation. Incorrect Answers A:Column labels do not explain concurrency-dependent memory loss by themselves. C:Comment style does not prevent overlapping tasks. D:Presentation cannot allocate additional memory.
Question 451 A SQL query scan was efficient last month, but query predicates now target a newly added column not considered in table layout. What optimization decision is warranted? Change the source Git repository name Only increase notebook line spacing Assume the existing clustering must optimize all future columns Reevaluate clustering/layout against the changed filter workload and measure improvements Explanation Correct Answer D:Table layout should reflect actual predicates; old clustering may no longer support efficient skipping. Incorrect Answers A:A repository name cannot change physical file placement. B:UI display does not improve data pruning. C:Clustering effectiveness is workload dependent.
Question 452 An ETL job invokes `df.count()` several times on an expensive transformation and has no caching or reuse strategy. What should be considered? Only rename the DataFrame variable Increase the number of calls to count for more certainty Delete transaction logs Avoid unnecessary repeated actions or persist a reused intermediate when justified Explanation Correct Answer D:Each action may recompute the lineage; controlled persistence or restructuring can reduce repeated work. Incorrect Answers A:Variable renaming does not change Spark recomputation. B:More actions can increase computation without changing the result. C:Log deletion is unrelated and dangerous to data consistency.
Question 453 An operations analyst wants to identify a performance regression but compares a 2 GB test run to a 2 TB production run without normalization. What is flawed? The test run is automatically more expensive Attribute the timing difference to code changes without normalizing the 2 GB and 2 TB workloads Spark execution time never depends on input size Input sizes and execution conditions are not comparable for direct runtime attribution Explanation Correct Answer D:Raw elapsed-time comparison conflates workload scale with efficiency. Incorrect Answers A:Cost cannot be inferred solely from smaller input without compute billing evidence. B:An input-volume difference of three orders of magnitude prevents isolating a code or hardware regression from raw elapsed times. C:Data volume can strongly influence execution time.
Question 454 A table has costly repeated full scans and analysts frequently filter by one high-cardinality key. Which optimization topic should be evaluated? Unconditional global sorting inside every dashboard query Increase the shuffle partition count without considering repeated filter predicates Appropriate liquid clustering or other supported data layout for that filter key Only adding job retries Explanation Correct Answer C:File layout aligned to common predicates can help data skipping and reduce scanned bytes. Incorrect Answers A:Repeated query-level sorting does not provide durable table layout optimization. B:More shuffle partitions do not directly organize data files so queries can skip irrelevant high-cardinality key ranges. D:Retries cannot improve healthy query scan selectivity.
Question 455 A job previously started in two minutes but now spends twenty minutes waiting for compute. What class of issue is indicated? An expensive groupBy stage after notebook execution Inspect only the query output and ignore cluster provisioning and job start timestamps Change source JSON parsing options while the job has not yet acquired compute Compute provisioning/queueing rather than transformation execution time Explanation Correct Answer D:A pre-start delay should be diagnosed through cluster events, quotas and scheduling capacity. Incorrect Answers A:The job is not yet executing transformations during the wait. B:The delay occurs before notebook execution. Query output alone cannot distinguish compute capacity or provisioning queue issues. C:Parsing executes after compute starts; a pre-execution wait calls for provisioning, policy and capacity diagnostics instead.
Question 456 A large aggregation produces accurate output but its p95 latency is high because a few tasks spill heavily to disk. What correction should be evaluated? Change the query column aliases only Scale all workers equally without inspecting the few high-spill partitions Increase every task retry count Rebalance or size expensive partitions and reduce per-task intermediate data Explanation Correct Answer D:Skewed memory-heavy tasks can dominate tail latency despite correct aggregate results. Incorrect Answers A:Aliases do not necessarily change memory or shuffle workload. B:Uniform scaling can increase expense while leaving a small set of pathological partitions responsible for p95 latency. C:Retries do not make a healthy spilling task faster.
Question 457 A job fails only when run against one larger production dataset. The same logic passes with a small sample. What is the appropriate diagnosis? Rename the production workspace instead of investigating data-dependent execution behavior Only disable monitoring to save time Investigate data volume, skew, memory, and scale-sensitive execution behavior Assume sample success proves production correctness and capacity Explanation Correct Answer C:Small samples may not expose large partitions, high cardinality, or memory pressure that appears at scale. Incorrect Answers A:Changing a workspace label does not correct resource pressure, partition skew or other scale-sensitive causes of the failure. B:Losing metrics makes the scale-dependent fault harder to isolate. D:A small test does not guarantee scaled execution.
Question 458 A governance administrator has granted SELECT on a Unity Catalog table to an analyst group. The group still receives an insufficient privilege error when referencing `finance.curated.invoices`. What other access must be checked? Ownership of the curated schema CREATE TABLE on the finance catalog USE CATALOG on finance and USE SCHEMA on curated MANAGE on the invoices table Explanation Correct Answer C:A table SELECT grant does not remove the requirement to traverse its parent catalog and schema with the appropriate usage privileges. Incorrect Answers A:Schema ownership is not required for a read-only analyst to query an accessible table. B:Creating new tables is not needed to read an existing table. D:MANAGE administers privileges and does not replace required traversal privileges for ordinary reading.
Question 459 A reporting team must inspect rows in `sales.reporting.monthly_summary` but must not modify source data. Which table privilege meets that access boundary? MANAGE on monthly_summary MODIFY on monthly_summary SELECT on monthly_summary with parent usage privileges CREATE TABLE on reporting Explanation Correct Answer C:SELECT authorizes data reads without granting modifications, provided the principal also has requisite catalog and schema usage. Incorrect Answers A:Privilege administration is outside the read-only reporting requirement. B:MODIFY enables data-changing operations and exceeds read-only requirements. D:Schema object creation is unnecessary and broader than reading a particular table.
Question 460 An ETL service principal must append new records to one existing target table while analyst identities should remain read only. Which grant is pertinent to the ETL writer? CREATE CATALOG on the metastore MODIFY on the target table plus applicable traversal and read privileges Only USE SCHEMA on the parent MANAGE on every table in the catalog Explanation Correct Answer B:Data modification operations require table write authorization; the service account should receive it only for the intended target. Incorrect Answers A:Creating catalogs is not required to append to an existing table. C:Schema traversal does not independently permit inserting or updating rows. D:Catalog-wide administration is excessive for one pipeline writer.