Microsoft DP-700 Practice Test 32

Question 310

An Eventhouse is expected to serve very frequent low-latency telemetry searches with native indexing and retention. Which storage architecture fits?

Use an external OneLake shortcut for all hot telemetry reads without assessing latency or acceleration.
Reimport every event into a SQL Warehouse at a nightly cadence.
Ingest the event feed into a native Eventhouse KQL table using a supported connector and appropriate retention/ingestion mapping.
Set Eventhouse retention to zero while attempting to preserve a history of alarm records.

Explanation

Correct Answer

Native KQL tables provide Eventhouse indexing and query behavior designed for high-QPS event workloads.

Incorrect Answers

External shortcut scans may be slower than native KQL ingestion/indexing under frequent search workloads.

A scheduled warehouse import cannot serve the required frequent low-latency native KQL search pattern.

Aggressive removal conflicts with the stated retention and recent-search use case.

Question 311

A high-QPS telemetry workload relies solely on a slow external shortcut and misses latency goals. No native ingestion or acceleration is configured. What limits it?

Native Eventhouse table ingestion requires a separate lakehouse Files shortcut to achieve indexing.
The slow query proves KQL has no support for low-latency retention policies.
The source telemetry timestamps are automatically deleted at the end of every query.
The external shortcut query path has different indexing/cache behavior from a native Eventhouse table and may repeatedly scan source Delta files.

Explanation

Correct Answer

A shortcut is not equivalent to native ingest for performance-sensitive event-search workloads.

Incorrect Answers

Native KQL ingestion/indexing does not depend on a lakehouse shortcut, and a shortcut could add remote-scan latency.

Native Eventhouse ingestion and retention are supported; the chosen external-only architecture misses native query characteristics.

A query does not remove source events or explain slow repeated external storage scans.

Question 312

A native Eventhouse ingestion design replaces a slow shortcut. What test confirms the migration improved the intended service?

Measure source-to-query freshness and representative KQL query latency under expected concurrency while verifying event count and schema parity.
Compare source and destination display names without measuring native KQL query latency.
Validate only the total storage bytes, ignoring native ingestion and query output.
Test a KQL table unrelated to the migrated telemetry stream.

Explanation

Correct Answer

Native ingestion success must preserve data completeness and improve the query service level, not just create a table item.

Incorrect Answers

Names cannot show whether the migrated workload meets query-throughput and retention targets.

Storage size alone cannot show native indexing, data freshness or correct filters.

Unrelated query timing cannot prove the intended workload benefits from native ingestion.

Question 313

A streaming window has a configured watermark, but some legitimate events arrive near the allowed-lateness boundary. What policy should engineers set?

Set the watermark delay to zero regardless of the two-hour late-arrival requirement.
Choose an explicit event-time lateness/watermark policy supported by the engine, with a documented handling path for events arriving too late.
Use processing-time buckets in place of event time without reconciling delayed source timestamps.
Set an unlimited state-retention policy without measuring backlog or correcting late-data rules.

Explanation

Correct Answer

Window state cannot be retained indefinitely at no cost; the lateness bound must match business tolerance and available memory.

Incorrect Answers

Zero tolerance can discard legitimate late events that should still update the intended event-time window.

Late records are then attributed to arrival windows rather than their original event occurrence.

Unbounded state can exhaust memory; lateness needs a documented bounded threshold and policy.

Question 314

A late event arrives after the streaming aggregation has discarded the state for its time window. Why is it not added to the total?

The configured watermark always accepts any event whose timestamp is earlier than current wall-clock time.
The engine is missing an extra Eventhouse table to hold late events automatically.
The watermark or window expiry policy has advanced beyond the event’s timestamp, so the engine treats it as too late for normal stateful inclusion.
An output chart's time zone setting determines the Spark state eviction moment.

Explanation

Correct Answer

Event-time windows enforce a bounded state-retention policy that can drop events arriving after the configured horizon.

Incorrect Answers

Watermark bounds lateness relative to observed event time; events after state eviction may be omitted.

A new table alone cannot restore evicted state or modify the stream's allowed-lateness policy.

Display time zones do not directly set Structured Streaming watermark progress.

Question 315

A revised event-time late-data policy is deployed. What validation distinguishes in-bound from out-of-bound arrivals?

Use only in-order events to test the revised late-data policy.
Confirm the job's batch-status page is green without checking window totals.
Compare generated ingestion timestamps but ignore original event timestamps.
Inject events just inside and outside the allowed-lateness boundary, then inspect aggregate changes and the documented late-event rejection/repair path.

Explanation

Correct Answer

Boundary tests demonstrate whether the configured watermark is honored, rather than checking only average arrival rates.

Incorrect Answers

In-order events cannot distinguish permitted lateness from events beyond the watermark.

A successful streaming batch can still drop excessively late events according to its watermark rule.

The configured watermark is based on event time; ingestion time alone can't test lateness semantics.

Question 316

A telemetry producer retries events after intermittent network failures. The sink should count each logical `event_id` only once. How should processing work?

Apply deduplication or idempotent sink updates keyed by the stable event identifier, with a suitable state/retention policy.
Assign a new random event ID each time the producer retries the same telemetry.
Deduplicate only on delivery timestamp because retries arrive at different times.
Suppress every event from a producer after its first message.

Explanation

Correct Answer

Delivery retries can repeat an event; stable-key reconciliation prevents replay from inflating metrics.

Incorrect Answers

Randomized IDs prevent matching repeat deliveries and inflate logical event counts.

A changing timestamp is unsuitable as a stable logical event identity key.

Dropping future unique events is not equivalent to retaining one copy per `event_id`.

Question 317

An Eventstream receives the same `event_id` twice, but both messages are counted as distinct observations. Select TWO likely causes.

The duplicate occurs because the producer didn't choose a new random ID for the retry.
The processing path uses append-only semantics without a stable event-id deduplication step.
The Eventstream has too few destinations to preserve per-ID uniqueness.
A source KQL query must be limited to one row to prevent all duplicates.
The deduplication key includes a changing delivery timestamp, so retries no longer match.

Explanation

Correct Answers

If every delivery is treated as new, a repeated message increments the count again.

A different delivery time creates a new composite identity even when the underlying event is unchanged.

Incorrect Answers

Retries should retain the same stable logical ID; randomizing it defeats reliable deduplication.

The number of sink branches does not make a single output deduplicate redeliveries.

A `take 1` query hides records rather than enforcing stable event-ID deduplication in the processing path.

Question 318

A streaming deduplication rule uses a stable event_id within its documented retry horizon. A source retries an event after an event-time minute boundary, but the retry is still within that horizon; a genuinely new event also arrives. Which controlled sequence verifies that delivery timestamps do not inflate the stream count?

Send two events with the same event_id in separate minute bins and accept both as distinct observations solely because their delivery timestamps differ.
Test only one initial event and assume retry behavior works across minute boundaries.
Send an original event, a retry with the same ID just across the minute boundary within the deduplication horizon, and a different event ID; confirm only two logical observations are counted.
Inspect the upstream connector label rather than the downstream distinct-event result.

Explanation

Correct Answer

The cross-minute retry exposes the relevant streaming boundary while preserving one logical event per stable ID; the distinct ID proves legitimate new events are not dropped.

Incorrect Answers

A delivery-time change does not create a new logical event when the deduplication contract uses stable IDs within the configured horizon.

A one-event fixture does not test whether retry-time or window boundaries accidentally create a second count.

A connector display label provides no evidence that cross-window retries are correctly suppressed in the stream aggregate.

Question 319

Two streaming sources encode event times differently: one ISO UTC, one local time with an offset. Before a shared window calculation, what should the pipeline do?

Sort mixed offset-bearing local timestamp strings lexically before a shared window.
Replace every source event time with current ingestion timestamp.
Drop numeric UTC offsets and treat all local wall-clock values as UTC.
Parse each supported timestamp representation and normalize to one documented event-time basis before binning or windowing.

Explanation

Correct Answer

Window membership depends on timestamp instants. Inconsistent type/zone parsing can place simultaneous events into different buckets.

Incorrect Answers

Lexical order and wall-clock text do not normalize physically equal instants across time zones.

Network and producer delays can move events to the wrong window, violating event-time reporting.

This shifts nonzero-offset events, particularly those crossing the business-day boundary.

Question 320

A joined Eventstream places events from different producers in inconsistent windows even though their physical times match. What is the likely cause?

The producers encode all events as plain local-time strings that can be directly compared lexically.
Source timestamps were parsed with inconsistent data types or time-zone assumptions before aggregation.
The event-time inconsistency arises because the database name is not UTC.
A later ingestion timestamp should always replace a supplied event timestamp for window attribution.

Explanation

Correct Answer

If one value is treated as local without its offset, the shared window sees a different instant than the other source.

Incorrect Answers

Equal physical instants with different offsets need normalization; lexical text order can differ from time order.

KQL database names have no authority over timestamp offset parsing or conversion semantics.

Ingestion time can differ from physical event time and cannot preserve the intended source-time windows.

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

Microsoft DP-700 Practice Test 30 - 10 Questions (Last Reviewed 05-Oct-2026)
Microsoft DP-700 Practice Test 31 - 10 Questions (Last Reviewed 04-Oct-2026)
Microsoft DP-700 Practice Test 32 - 10 Questions (Last Reviewed 03-Oct-2026)
Microsoft DP-700 Practice Test 33 - 10 Questions (Last Reviewed 02-Oct-2026)
Microsoft DP-700 Practice Test 34 - 10 Questions (Last Reviewed 01-Oct-2026)

Pass Your Microsoft DP-700 Exam Easy!

Question 310

Explanation

Correct Answer

Incorrect Answers

Question 311

Explanation

Correct Answer

Incorrect Answers

Question 312

Explanation

Correct Answer

Incorrect Answers

Question 313

Explanation

Correct Answer

Incorrect Answers

Question 314

Explanation

Correct Answer

Incorrect Answers

Question 315

Explanation

Correct Answer

Incorrect Answers

Question 316

Explanation

Correct Answer

Incorrect Answers

Question 317

Explanation

Correct Answers

Incorrect Answers

Question 318

Explanation

Correct Answer

Incorrect Answers

Question 319

Explanation

Correct Answer

Incorrect Answers

Question 320

Explanation

Correct Answer

Incorrect Answers

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