Question 390 A Spark stage shows high task scheduling overhead because a tiny dataset was split into tens of thousands of shuffle partitions. What can help? Drop all query predicates Change table owner privileges Reassess shuffle partition count and adaptive coalescing behavior Increase partition count without limit Explanation Correct Answer C:Excessively small tasks can cost more to schedule than to process; suitable partition sizing reduces overhead. Incorrect Answers A:Processing more rows will not solve too many tiny tasks. B:Permissions do not set physical shuffle task granularity. D:More tiny tasks can amplify scheduler overhead.
Question 391 A wide aggregation writes huge intermediate spill files, while available executor memory is constrained. Which two properties should be examined to choose a fix? Increase the executor count without examining how large each aggregation task’s shuffled partitions are Per-task memory footprint and the number/distribution of shuffle partitions Inspect only total input row counts and assume that each reduce task receives comparable data Increase the driver heap without investigating executor spill metrics Explanation Correct Answer B:Skew and insufficient task memory can jointly cause disk spills and slow aggregation. Incorrect Answers A:More workers cannot guarantee that one oversized aggregation partition fits into the per-task memory budget. C:A total row count does not show whether skew or wide rows make particular aggregation partitions spill to disk. D:A distributed shuffle spilling on executors is not necessarily fixed by larger driver memory.
Question 392 A large Delta table suffers slow queries after many tiny files accumulate. What maintenance topic should the engineer evaluate? File compaction/optimization through supported Delta maintenance Increase the number of scheduled notebook runs without compacting the small Delta files Grant broader table MODIFY privileges without changing the physical file layout Deleting arbitrary transaction-log entries Explanation Correct Answer A:Consolidating small files can reduce planning and I/O overhead in compatible workloads. Incorrect Answers B:Additional runs can add files or load but do not address the high file-count overhead slowing Delta scans. C:Access privileges do not compact fragmented Delta files or improve the many-small-files query plan. D:Uncoordinated log deletion risks corrupting table consistency.
Question 393 A table has high-cardinality filters and frequent inserts, and manual partitioning is becoming difficult. Which Databricks feature is relevant for layout management? Create fixed high-cardinality directory partitions for every possible filter value Adding dozens of static partition columns without evaluation Liquid clustering where supported Keeping the full table in notebook variables Explanation Correct Answer C:Liquid clustering can simplify adaptive data layout for frequently queried Delta tables. Incorrect Answers A:Numerous fine-grained static partitions can be hard to maintain under frequent inserts and may increase small-file overhead. B:Overpartitioning can create many tiny partitions and maintenance overhead. D:Notebook memory is not a durable scalable data layout strategy.
Question 394 An engineer assumes Liquid Clustering means every query will be fast without examining its predicates. What should be clarified? Liquid clustering removes need for data quality Clustering effectiveness depends on query patterns, data and maintenance state Liquid clustering guarantees zero compute cost Liquid clustering overrides all SELECT permissions Explanation Correct Answer B:Layout choices still need representative measurement and compatible workload filters. Incorrect Answers A:File organization does not validate record semantics. C:Clustering does not eliminate query execution or maintenance charges. D:Physical layout does not bypass Unity Catalog data access.
Question 395 A Databricks managed table is eligible for predictive optimization, and the team wants maintenance recommendations applied without managing every operation manually. Which capability fits? A manually repeated unrelated Git merge Schedule a manually maintained OPTIMIZE command for every table without using eligibility Applying arbitrary vacuum retention of zero without safeguards Predictive optimization for supported Unity Catalog managed tables Explanation Correct Answer D:Predictive optimization can automate suitable maintenance operations based on observed table needs. Incorrect Answers A:Git merges cannot maintain Delta table file layout automatically. B:Manual maintenance can be useful but does not deliver the supported automated predictive optimization requested for eligible tables. C:Unsafe retention choices can break time travel and do not implement intelligent maintenance.
Question 396 A team wants to monitor whether predictive optimization actually improves a heavy query rather than assume its benefits. What should be measured? Compare only the table size before and after optimization Representative query runtimes and cost before and after maintenance Only the daily number of Git commits Compare maintenance job duration alone without measuring representative query runtime Explanation Correct Answer B:Workload measurements establish whether optimized layout helps the target analytic queries. Incorrect Answers A:Table bytes alone cannot show the impact on representative query latency or the cost of maintenance operations. C:Commit frequency does not directly measure data-file query efficiency. D:Optimization duration is a cost, but it cannot establish whether important user queries became faster or cheaper.
Question 397 A SQL workload suddenly experiences high latency after a data volume spike. What diagnosis distinguishes compute pressure from a table-layout issue? Increase compute before checking whether filters stopped pruning files Only delete the table's history Inspect query plans, warehouse load, and data-scan/file metrics together Always rewrite every query as Python Explanation Correct Answer C:Both execution resources and table I/O can drive latency, so evidence from each is necessary. Incorrect Answers A:Scaling a warehouse might mask some delay, but without the plan and scan metrics it cannot distinguish a layout regression from capacity pressure. B:History deletion would not distinguish compute pressure from scanning cost. D:Language changes alone are not a reliable performance diagnosis.
Question 398 A Databricks job had a failed run, and the operator wants to identify which task first failed before downstream tasks were skipped. What should be inspected? Check only the top-level job result state and not individual task histories Only the number of catalog users The task-level DAG statuses and logs for the first failed dependency Inspect only downstream tasks marked skipped without examining their failed prerequisites Explanation Correct Answer C:The graph shows which failure blocked later tasks and where the original error occurred. Incorrect Answers A:The overall failure state cannot isolate the first failed node from tasks skipped as consequences. B:Principal count does not reveal which task generated the failure. D:Skipped downstream nodes reveal an effect of the failure, but cannot independently locate the first task error or its underlying log.
Question 399 A job begins failing during cluster startup after a new organizational policy restricts allowed instance types. Which root cause should be checked? The compute policy and selected node types against allowed settings Increase the instance size without examining policy restrictions Only the row counts in silver Only the SQL aggregation statement Explanation Correct Answer A:A newly restrictive policy can reject the configured cluster before any notebook logic executes. Incorrect Answers B:A larger node type might also violate the policy; startup failures should be checked against allowed types and cluster settings. C:Data volume does not explain a policy rejection before compute exists. D:The notebook never started, so SQL syntax is unlikely to be the pre-start blocker.
Question 400 A job has many stages with executor CPU near idle and high network shuffle time. Which tuning lever is more plausible than adding CPU cores blindly? Reduce unnecessary exchanges or improve join and aggregation partitioning Increase executor CPU indefinitely regardless of plan Increase executor heap without inspecting remote shuffle transfer and skew metrics Revoke dataset SELECT grants Explanation Correct Answer A:When network movement dominates, more CPU may not address the bottleneck. Incorrect Answers B:Additional cores cannot remove excessive data exchange by themselves. C:More heap might not address a network-bound shuffle stage; the diagnosis should start from the observed fetch waits and partition distribution. D:Reducing permissions would not make the authorized query execute faster.