DP-700

Free DP-700 practice questions

10 questions from our DP-700 practice tests, each with the correct answer and a written explanation. Read them with the answers shown, or switch to quiz mode and answer first.

Back to the DP-700 practice exams

Question 1 of 10, Implement and manage an analytics solution. Choose 2.

A Fabric workspace must run a nightly pipeline at 01:00 UTC, and a second pipeline must start whenever a new file lands in a specific Azure Blob Storage container. The team wants to avoid custom polling code. Which combination should the data engineer use? (Choose two.)

  1. AA storage event trigger on the second pipeline that reacts to Blob created events Correct answer
  2. BA schedule configured on the nightly pipeline with a daily recurrence Correct answer
  3. CA Spark Job Definition that runs continuously and lists the container contents
  4. DAn Until activity that loops every minute and checks the container with Get Metadata
  5. EA Copy job scheduled every minute that copies any new files found in the container

Answer and explanation: Fabric pipelines support schedules for time-based runs and event-based triggers (built on Real-Time hub and Activator) that fire on Azure Blob Storage events such as BlobCreated. An Until loop with Get Metadata and a continuously running Spark job are both custom polling that the team wants to avoid. A Copy job on a one-minute schedule is still time-based polling rather than reacting to the storage event.

Question 2 of 10, Implement and manage an analytics solution

A company has a Fabric warehouse with a SalesOrders table. Regional managers must see only the rows for their own region when they query the table with T-SQL. The solution must be enforced inside the warehouse for every query tool. What should the data engineer implement?

  1. ARow-level security roles defined in the Power BI semantic model that the reports build on top of the warehouse
  2. BA dynamic data masking rule on the Region column with the default() function
  3. CA view per region and a GRANT SELECT on each view to the matching regional security group
  4. DAn inline table-valued predicate function bound to the table through CREATE SECURITY POLICY Correct answer

Answer and explanation: Row-level security in a Fabric warehouse uses an inline table-valued function as a filter predicate that is applied by a security policy, so every T-SQL query against the table is filtered. Masking only obscures values and does not remove rows. Per-region views leave the base table exposed and do not scale. Semantic model RLS applies only to queries through the semantic model, not to direct T-SQL access.

Question 3 of 10, Implement and manage an analytics solution. Choose 2.

A data engineer configures dynamic data masking on several columns of a Fabric warehouse table. Which statements about the behaviour are correct? (Choose two.)

  1. AThe masking rules are also applied when Spark notebooks read the table's Delta files
  2. BMasked values are stored encrypted with a separate key in OneLake
  3. CA user without UNMASK can still infer masked values by filtering on the column in a WHERE clause Correct answer
  4. DUsers with the Admin, Member or Contributor workspace role see the unmasked data Correct answer
  5. EA user without UNMASK cannot reference a masked column in a WHERE clause

Answer and explanation: Dynamic data masking only changes query results for users without elevated permissions; workspace Admins, Members and Contributors see real values, and filtering on a masked column still works, so values can be inferred. The stored data is not changed or separately encrypted, masking is enforced by the SQL engine rather than in Delta files read by Spark, and masked columns can be used in WHERE clauses.

Question 4 of 10, Ingest and transform data

A team must choose a Fabric data store for clickstream telemetry that arrives continuously at high volume and is queried mostly with time-window aggregations over the last few hours. Which store is the BEST fit?

  1. AAn eventhouse with a KQL database Correct answer
  2. BA lakehouse loaded once a day by a Dataflow Gen2 refresh
  3. CA SQL database in Fabric with a clustered index on the event timestamp
  4. DA Fabric warehouse loaded every hour with COPY INTO from staging files

Answer and explanation: An eventhouse with a KQL database is optimized for high-volume streaming ingestion and fast time-series analytics with KQL. A warehouse loaded hourly and a lakehouse loaded daily add latency and are not optimized for continuous telemetry. A SQL database in Fabric is an OLTP store and is not designed for large-scale time-series analytics.

Question 5 of 10, Ingest and transform data

In a KQL database, the engineer must calculate the average temperature for each device in 5-minute intervals. Which query is correct?

  1. AReadings | extend Interval = 5m | summarize avg(Temperature) by Interval
  2. BReadings | summarize avg(Temperature) by DeviceId, bin(Timestamp, 5m) Correct answer
  3. CReadings | make-series avg(Temperature) on DeviceId step 5m
  4. DReadings | where Timestamp > ago(5m) | summarize avg(Temperature)

Answer and explanation: summarize with bin(Timestamp, 5m) groups rows into 5-minute buckets, and adding DeviceId produces one average per device per bucket. Filtering on ago(5m) returns only the last five minutes as a single value. Extending a constant column groups every row into the same bucket. make-series must be ordered on a datetime axis, not on DeviceId.

Question 6 of 10, Ingest and transform data

A company wants an operational Azure SQL Database replicated into OneLake in near real time as Delta tables, without building or scheduling any ETL pipelines. Which Fabric feature should be used?

  1. AAn eventstream with the Azure SQL Database CDC source writing to an eventhouse
  2. BA OneLake shortcut that points to the Azure SQL Database
  3. CA Copy job configured for a full load once per day
  4. DDatabase mirroring Correct answer

Answer and explanation: Mirroring continuously replicates supported databases such as Azure SQL Database into OneLake as Delta tables with no pipeline to build. Shortcuts point to storage locations such as ADLS Gen2, S3 or other OneLake paths, not to relational databases. A CDC eventstream into an eventhouse lands KQL tables and requires building a stream. A daily Copy job is batch, not near real time.

Question 7 of 10, Ingest and transform data

A Spark structured streaming query reads orders from a source and computes counts per 10-minute window. Events can arrive up to 15 minutes late, and state must not grow without limit. Which configuration should be used?

  1. AIncrease spark.sql.shuffle.partitions so that late events fit in memory
  2. BCall withWatermark('eventTime', '15 minutes') before the window aggregation Correct answer
  3. CUse outputMode('complete') so the full aggregation table is rewritten on each trigger
  4. DSet trigger(once=True) so that late events are processed in the next run

Answer and explanation: withWatermark tells Spark how late events can arrive and allows it to drop old window state once the watermark passes, bounding state size while accepting events up to 15 minutes late. Complete output mode keeps all aggregation state forever. A one-time trigger changes scheduling, not lateness handling. More shuffle partitions changes parallelism but does not limit state growth.

Question 8 of 10, Monitor and optimize an analytics solution

A notebook fails with an out-of-memory error on the driver after a developer added df.toPandas() on a DataFrame with hundreds of millions of rows. Which change resolves the error while keeping the processing distributed?

  1. AEnable high concurrency mode for notebooks in the workspace
  2. BSwitch the notebook language to Spark SQL and keep the toPandas() call in a later cell
  3. CKeep the logic in Spark DataFrame operations and write the result to a Delta table instead of collecting it Correct answer
  4. DIncrease the number of executors in the Spark pool while keeping toPandas()

Answer and explanation: toPandas() collects all rows to the driver, which causes the driver out-of-memory error; keeping the work in distributed DataFrame operations and writing to Delta avoids collecting the data. More executors do not increase driver memory. High concurrency mode shares sessions but does not change driver memory usage. Keeping toPandas() in any language still collects all data to the driver.

Question 9 of 10, Monitor and optimize an analytics solution

Queries on a large lakehouse Delta table have slowed down because frequent small appends created thousands of small Parquet files. Which action improves read performance MOST directly?

  1. AConvert the table to CSV files in the Files section
  2. BRun OPTIMIZE on the table to compact small files, applying V-Order Correct answer
  3. CAdd a partition column with a unique value for every row to spread the data across many folders
  4. DRun VACUUM with a retention of 0 hours

Answer and explanation: OPTIMIZE performs bin compaction of small files into larger ones, and V-Order further improves read performance. VACUUM only removes files that are no longer referenced by the table and does not compact active files; a 0-hour retention can also break time travel and concurrent readers. CSV is slower to scan than Parquet. Partitioning on a unique value creates even more small files.

Question 10 of 10, Monitor and optimize an analytics solution

A lakehouse table is read mostly by Direct Lake semantic models in Power BI. Which table property MOST improves Direct Lake read performance on that table?

  1. AEnabling V-Order on the Parquet files of the table Correct answer
  2. BSetting the Delta deleted-file retention to one hour
  3. CDisabling the Delta transaction log checkpoint interval
  4. DStoring the table as JSON in the Files section

Answer and explanation: V-Order is a write-time optimization of Parquet that applies sorting, encoding and compression favored by the VertiPaq engine, which significantly speeds up Direct Lake and SQL reads. Disabling checkpoints slows log replay. JSON files in Files are not Delta tables and cannot be used by Direct Lake. Deleted-file retention controls VACUUM and time travel, not read speed.