Microsoft DP-700 Practice Test 47

Question 460

A Copy job has high sink capacity available but throughput plateaus exactly when the source API responds with 429 and caps simultaneous requests. Select TWO root-cause conclusions justified by these observations.

Source-side rate limiting, not destination compute capacity, is the immediate transfer throughput constraint.
Additional concurrent requests can worsen throttling if they exceed the API quota, so upstream limits must govern tuning.
The Copy sink has insufficient CPU despite the source returning explicit 429 responses.
The solution is to increase parallel REST requests until the source accepts the load.
The source is missing a warehouse statistics refresh for its rate-limit responses.

Explanation

Correct Answers

A rate-limited upstream API cannot achieve higher sustained transfer rates merely by increasing Fabric sink capacity.

More client threads can raise the 429 response rate while leaving actual accepted throughput unchanged.

Incorrect Answers

An upstream API concurrency quota, not spare sink capacity, is the observed bottleneck.

More concurrency can intensify HTTP rate limiting and reduce successful transfer throughput.

Warehouse statistics don't control a remote REST API's request budget and Retry-After policy.

Question 461

Engineers tune Copy concurrency after identifying upstream and sink limits. What is the best acceptance experiment?

Measure only sink CPU utilization when testing concurrency changes.
Run equivalent source data volumes before and after, comparing bytes/sec, throttled requests, sink commit throughput and total elapsed time.
Run one low-volume request and conclude the peak load is safe.
Ignore all 429 errors and compare only the largest observed Copy transfer speed.

Explanation

Correct Answer

A controlled end-to-end transfer benchmark establishes whether the specific bottleneck improved without increasing failures or missing records.

Incorrect Answers

The source 429 rate, retry timing and end-to-end completion must be included to assess the upstream limit.

A single request won't exercise the source API's real concurrency quota under sustained load.

Discarding throttled failures can make an incomplete load appear fast; reconcile transferred source records.

Question 462

A Fabric ingestion pipeline copies each source quickly, yet total duration is dominated by sequential dependency waits between independent activities. Which orchestration setting should be reviewed?

Enable Delta V-Order on every target and assume it removes all pipeline waits.
Make a cyclic dependency so every activity waits for every other one.
Review the activity dependency graph and appropriate permitted concurrency so independent copy activities can execute in parallel without violating source/sink limits.
Delete pipeline run history to remove the visible wait time.

Explanation

Correct Answer

Unnecessary serial dependencies extend critical-path time even when individual Copy activities transfer quickly.

Incorrect Answers

Parquet layout optimization is unrelated to the activity dependency graph. It does not implement the corrective control requested in this incident.

Circular dependencies prevent progress and make an orchestration deadlock worse. It does not implement the corrective control requested in this incident.

Hiding telemetry does not change scheduling or shorten the real workflow critical path. It does not implement the corrective control requested in this incident.

Question 463

A pipeline contains five independent one-minute Copy activities connected in an unconditional serial chain. The reported Copy throughput is healthy, but end-to-end duration is much longer than expected. What causes it?

Copy throughput per activity is low because the sources all share a single SQL table.
The job needs larger executors for each serial Copy step to eliminate sequential scheduling.
The pipeline's business domain must be split before two activities can run concurrently.
The pipeline waits on sequential dependencies instead of executing independent eligible activities concurrently.

Explanation

Correct Answer

The critical path includes all five waits; increasing a Copy transfer buffer does not eliminate scheduling-induced serialization.

Incorrect Answers

Independent sources and healthy per-activity speed indicate serialization overhead in the pipeline dependency graph.

Executor size is not the gate; unconditional serial dependencies prevent independent Copy tasks overlapping.

Domain assignment doesn't control activity dependency edges or scheduling eligibility.

Question 464

A Fabric pipeline is refactored to run truly independent Copy activities in parallel. Select TWO checks that validate the improvement safely.

Compare the monitored activity Gantt/critical path and total run duration before and after with equivalent data volumes.
Verify only that the overall job duration decreases while ignoring dependency correctness.
Compare throughput of one Copy task without measuring end-to-end critical-path runtime.
Reconcile each destination output and confirm upstream API and sink throttling remain within approved limits.
Skip row-count and key reconciliation because independent activities should never interfere.

Explanation

Correct Answers

Overlap in scheduled execution and a shorter end-to-end timeline establish that the prior dependency bottleneck was reduced.

Parallelism is useful only if the activities still write correct independent data and do not overload constrained services.

Incorrect Answers

An unsafe parallelization might launch activities before their prerequisites or alter shared outputs.

One activity's throughput was already healthy; the benefit comes from parallel wall-clock reduction.

Parallel runs may contend for shared destinations or change ordering; validate data integrity as well as timing.

Question 465

A Fabric Warehouse query suddenly selects an expensive join after a major change in fact-table distribution. Which evidence-guided optimization should be considered first?

Inspect the actual and estimated plan/cardinality, assess automatically maintained or stale column statistics, and refresh relevant statistics when evidence warrants it.
Enable a OneLake row-level security policy so the query optimizer automatically learns table cardinalities.
Increase only the reporting dashboard refresh interval instead of inspecting the join plan.
Apply an unrelated file-name change to force a distributed join plan.

Explanation

Correct Answer

Warehouse plans rely on row-count and distribution estimates; confirming the estimate mismatch before selectively refreshing statistics avoids arbitrary query changes.

Incorrect Answers

Access grants filter user-visible data but are not a general replacement for relevant optimizer statistics. It does not implement the corrective control requested in this incident.

A slower refresh cadence cannot resolve inaccurate query-plan estimates during execution. It does not implement the corrective control requested in this incident.

Renaming a file does not supply statistics or fix a bad join estimate. It does not implement the corrective control requested in this incident.

Question 466

A query plan expects a few hundred rows from a join input, but the actual input is millions of rows and a poor join algorithm dominates elapsed time. What is the relevant optimizer failure?

The slow join must be caused by a Fabric workspace domain lacking a description.
Changing the report display name forces Fabric to choose a different join strategy.
The cardinality estimates used to choose the physical plan are inaccurate for the current table distribution.
The query's SQL parser is missing support for a simple equijoin.

Explanation

Correct Answer

A misestimated join input can lead to nonoptimal join order, memory requirements and data movement even though the SQL query itself is valid.

Incorrect Answers

Governance metadata cannot explain an optimizer estimate of hundreds versus millions of rows.

Names don't update warehouse statistics or correct cardinality estimation.

The join executes; its excessive cost follows from a poor physical plan due to underestimated input.

Question 467

After refreshing justified Warehouse statistics for a badly estimated join, which test most directly measures the result?

Verify only that a statistics update command finishes, without inspecting the new estimated cardinality.
Compare query elapsed times on different filter predicates and claim the estimate is repaired.
Check table row counts but omit the physical plan and query result reconciliation.
Compare the updated plan’s estimated versus actual row counts, join choice, execution memory and elapsed time for identical parameters and results.

Explanation

Correct Answer

Plan comparison checks that the optimizer changed its estimated work and the real query improved without changing answer semantics.

Incorrect Answers

A completed update may not improve the specific join's plan or execution performance.

Changing selectivity prevents a fair assessment of whether the intended cardinality estimate improved.

Row counts alone don't prove join algorithm, estimation or output semantics after the change.

Question 468

A large Fabric Warehouse query displays “One or more non-scalable operation is detected” and most time is spent in a final global sort over millions of records. What should the engineer investigate?

Inspect the plan for the required single-node global ordering and reduce or restructure its intermediate/result data; consider a distributed alternative or documented plan hint only when semantics permit.
Ignore final result changes if the optimized query is faster.
Turn off statistics collection so the optimizer cannot choose a plan.
Create a separate unrelated Power BI dashboard and assume MPP serialization is fixed.

Explanation

Correct Answer

MPP warehouses can serialize global ordering or final merges; reducing the sorted set or finding a semantically valid distributed execution reduces the bottleneck.

Incorrect Answers

A faster query that changes required results or ordering fails correctness despite improved elapsed time. It does not implement the corrective control requested in this incident.

Removing plan inputs commonly worsens physical decisions and does not solve the need for global order. It does not implement the corrective control requested in this incident.

A consumer artifact does not rewrite the Warehouse distributed execution plan. It does not implement the corrective control requested in this incident.

Question 469

A query reads distributed Warehouse partitions efficiently but then spends most of its time on one node producing a complete global ORDER BY result. What explains the slow tail?

Every distributed ORDER BY must finish in parallel with no final merge or coordination.
The final global ordering requires a non-scalable data consolidation/sort stage that limits parallel execution.
The query is bottlenecked by the input partitions even though they have completed promptly.
An Eventstream event-time watermark automatically controls SQL Warehouse global sorts.

Explanation

Correct Answer

A globally ordered single result can require concentration of distributed rows after the scan; adding scan partitions alone does not eliminate the merge.

Incorrect Answers

A global ordering can require concentrated sort/merge work; the slow final node can dominate latency.

Observed distribution work is fast; the remaining global result-ordering stage causes the tail.

Streaming watermarks don't eliminate a warehouse query's required global ordering work.

Question 470

A data engineer rewrites a non-scalable Warehouse query. Select THREE checks that distinguish legitimate improvement from a changed query result.

Review the query plan to confirm the expensive single-node/global-sort stage was reduced or safely eliminated.
Remove ORDER BY entirely and count reduced runtime as an equivalent query optimization.
Compare only the first ten rows without checking full result and required ordering.
Reconcile the final result keys, aggregates and any required sort order with the original query.
Compare end-to-end elapsed time and resource use on representative data under comparable capacity and cache conditions.
Inspect reduced CPU consumption but ignore final output row counts and stable tie-break rules.

Explanation

Correct Answers

Plan evidence targets the actual MPP bottleneck rather than relying solely on observed duration.

An optimization is not valid if it silently changes business totals or drops ordering that is contractually required.

A controlled benchmark verifies whether reducing non-scalable work produces a reproducible performance benefit.

Incorrect Answers

Dropping required global order changes query semantics and is not a valid performance comparison.

A partial result can look correct while the global order or other rows differ after rewriting.

Resource savings cannot justify missing or misordered rows where the business requires complete ordered output.

Implementing Data Engineering Solutions Using Microsoft Fabric (DP-700) Practice Tests

Microsoft DP-700 Practice Test 45 - 10 Questions (Last Reviewed 20-Sep-2026)
Microsoft DP-700 Practice Test 46 - 10 Questions (Last Reviewed 19-Sep-2026)
Microsoft DP-700 Practice Test 47 - 10 Questions (Last Reviewed 18-Sep-2026)
Microsoft DP-700 Practice Test 48 - 10 Questions (Last Reviewed 17-Sep-2026)
Microsoft DP-700 Practice Test 49 - 10 Questions (Last Reviewed 16-Sep-2026)

Pass Your Microsoft DP-700 Exam Easy!

Question 460

Explanation

Correct Answers

Incorrect Answers

Question 461

Explanation

Correct Answer

Incorrect Answers

Question 462

Explanation

Correct Answer

Incorrect Answers

Question 463

Explanation

Correct Answer

Incorrect Answers

Question 464

Explanation

Correct Answers

Incorrect Answers

Question 465

Explanation

Correct Answer

Incorrect Answers

Question 466

Explanation

Correct Answer

Incorrect Answers

Question 467

Explanation

Correct Answer

Incorrect Answers

Question 468

Explanation

Correct Answer

Incorrect Answers

Question 469

Explanation

Correct Answer

Incorrect Answers

Question 470

Explanation

Correct Answers

Incorrect Answers

Implementing Data Engineering Solutions Using Microsoft Fabric (DP-700) Practice Tests