Question 410 A long-running job is failing during dependency installation because package versions conflict. Which logs and state matter? Retry the same job without examining the declared dependency versions Inspect only the job run result and omit the library installation errors Library installation logs and the declared dependency graph Upgrade every installed package blindly to its latest available version Explanation Correct Answer C:Version constraints and installation errors reveal incompatible package combinations. Incorrect Answers A:A retry may encounter the identical package-resolution conflict; library installation evidence is needed to identify incompatible version constraints. B:A failed overall run status can show when installation failed but not which dependency constraints or installed artifacts caused the conflict. D:Updating all packages without checking compatibility constraints can introduce new conflicts instead of resolving the reported dependency error.
Question 411 An engineer responds to a Spark shuffle regression by raising `spark.sql.shuffle.partitions` from 200 to 20,000. What risk must be measured? Excess tiny task overhead versus reduced per-task data volume All extra partitions grant additional data privileges It deletes Delta transaction logs The setting changes the table business keys Explanation Correct Answer A:More partitions can reduce task sizes but may introduce substantial scheduling and shuffle overhead. Incorrect Answers B:Physical partition count does not affect Unity Catalog authorization. C:The configuration is not a storage cleanup command. D:Shuffle partitions do not rewrite source entity identities.
Question 412 A query plan chooses a sort-merge join where a small dimension could be broadcast. What should be validated before forcing a hint? Force a broadcast join without checking the actual lookup size or executor memory Increase shuffle partitions while refusing to inspect the small side’s statistics and broadcast eligibility That all dimensions must be broadcast regardless of size Estimated small-side size, memory safety and join compatibility Explanation Correct Answer D:Broadcast hints help only when the relation fits and the join semantics support the intended strategy. Incorrect Answers A:A broadcast hint can be useful when the build side is small enough, but blindly forcing it can exhaust memory. B:Partition count tuning does not validate whether the dimension can be safely broadcast as requested. C:Large dimensions may cause memory pressure when broadcast.
Question 413 A production compute cluster cannot start because an instance type is unavailable in its cloud region. What mitigation is appropriate? Use supported available instance capacity or adjust compute configuration under policy Only increase notebook cell retries Grant broader Delta table read rights Rewrite downstream aggregation SQL before compute exists Explanation Correct Answer A:Startup needs compatible cloud capacity; a supported alternative can restore provisioning. Incorrect Answers B:The source code is not the immediate provisioning failure. C:Data grants do not create unavailable cloud instances. D:Execution code cannot run until a cluster is provisioned.
Question 414 An engineer wants to know if a failed Lakeflow Job task never started or ran and threw an exception. Which evidence distinguishes the two? Only the number of source schemas Per-task state together with start time and task logs Read only the overall job failure message without task start or state details Only the job display name Explanation Correct Answer B:A skipped or pending task differs from an executed task that raised a runtime exception. Incorrect Answers A:Catalog structure does not show task attempt state. C:Top-level failure can include tasks that were skipped; start timestamps and task logs distinguish skipped from executed failures. D:Names cannot distinguish execution from skip state.
Question 415 A Spark SQL query has high executor CPU and very little shuffle or I/O. Which bottleneck category should be examined before changing storage layout? Only cross-region object-store bandwidth CPU-bound expression and computation cost Only Delta file compaction Increase network bandwidth without examining the CPU-intensive expressions in the execution plan Explanation Correct Answer B:When computation dominates without significant I/O, optimizing expressions or CPU resources is more relevant. Incorrect Answers A:Little observed I/O makes bandwidth a weaker primary explanation. C:Compaction may not solve a query dominated by heavy CPU expressions. D:When CPU is saturated and I/O/shuffle is low, network bandwidth is unlikely to be the principal bottleneck.
Question 416 A Spark job consistently starts quickly but fails after hours when the data volume reaches a high watermark. Which diagnosis is most helpful? Inspect stage metrics, memory and spill behavior at the failing scale Inspect only cluster startup logs before the input reaches the failing scale Only investigate cluster startup events Only inspect dashboard permissions Explanation Correct Answer A:A scale-dependent runtime failure is more likely in task resources or data distribution than initial provisioning. Incorrect Answers B:Successful startup is not evidence about later high-volume shuffle, memory, or spill conditions. C:The cluster started successfully and processed data for hours. D:Read permissions are unlikely to cause a late out-of-memory stage.
Question 417 A query performs poorly after adding an unnecessary `orderBy` before a large join. What should the engineer evaluate? Revoke every source table grant Remove or defer sorts not required by the query semantics Add additional arbitrary global sorts Add a second global sort after the join and assume this cancels the earlier sort Explanation Correct Answer B:An unneeded global sort can cause expensive shuffle work without improving join correctness. Incorrect Answers A:Authorization changes are unrelated to the extra sort cost. C:More sorting generally increases exchange and comparison work. D:Extra ordering generally adds work; it does not eliminate unnecessary pre-join sorting.
Question 418 A cluster frequently terminates for inactivity while a long-running scheduled job still needs a stable execution environment. What configuration should be reviewed? Keep the current idle auto-termination value and only increase notebook retries Keep interactive compute awake using manual notebook pings instead of reviewing the job compute lifecycle The compute lifecycle and whether the job uses appropriately configured job compute Disable all job retries without checking whether the compute lifecycle is terminating active work Explanation Correct Answer C:Job compute lifecycle must cover the task run rather than rely on inappropriate interactive idle settings. Incorrect Answers A:A task retry setting does not fix compute lifecycle policies incompatible with the intended execution. B:Ad hoc keep-alive traffic is not a reliable production execution policy and can waste resources. D:Retry policy cannot substitute for correctly configured job compute and lifecycle controls.
Question 419 An operator uses Spark UI and sees substantial disk spill but not elevated data input volume compared with yesterday. What change should be investigated? Assume the same input row count guarantees the same physical execution plan Check only the aggregate cluster size and ignore changes in task-level partition sizes Execution plan, partition skew and available task memory compared with the previous run Increase only the driver heap without inspecting spilled partitions and executor memory Explanation Correct Answer C:Spill changes can follow a plan regression or reduced memory even with stable source volume. Incorrect Answers A:Input volume may be stable while join strategy, skew, or memory configuration changes enough to cause new spills. B:Nominal cluster capacity does not reveal a skewed partition or changed plan causing new disk spills. D:Remote task spill usually concerns shuffle partitioning, skew and executor working sets; driver heap is not the observed bottleneck.
Question 420 A gold query has fast compute but spends excessive time opening many Delta data files. Which issue should be evaluated? Change access grants for the query user while ignoring physical file fragmentation Rewrite the notebook comments while leaving the many small Delta files unchanged Small-file fragmentation and compaction state Increase SQL warehouse size without inspecting file fragmentation Explanation Correct Answer C:Opening a very large number of tiny files can dominate I/O and metadata overhead. Incorrect Answers A:Authorization changes do not correct file-open overhead once queries already execute successfully. B:Comment updates do not change the number of files the engine must open and schedule during the scan. D:Compute scaling alone does not remove per-file open and metadata overhead from many tiny files.