DP-750 sample exam questions

A schema-level managed storage location in Unity Catalog is set with CREATE SCHEMA ... MANAGED LOCATION. Plain LOCATION is for external tables or other objects, CREATE CATALOG would make the wrong object, and DBPROPERTIES has no location property.

Libraries uploaded to the workspace and installed on the cluster are governed through the cluster's library installation and Unity Catalog compatible access. pip3 and custom scripts bypass governance, and installing then manually restarting does not by itself address access control.

Enabling autoscaling lets the cluster add and remove workers with load, and an auto-termination timeout stops paying for idle compute. A fixed worker count prevents scaling, disabling Photon does not drive scaling, and a policy for managing settings does not scale or cut costs.

Liquid clustering replaces partitioning and Z-ordering for new tables and adapts to growth on keys such as CustomerId and EventDate. Deletion vectors mark rows as removed without rewriting whole Parquet files, which reduces update and delete effort. VACUUM only cleans up old files.

Serverless compute for pipelines autoscales and needs no cluster configuration or management, so it minimizes both cost and administrative effort. Job clusters need configuring, all-purpose clusters cost more, single node cannot scale, and SQL warehouses are not pipeline compute.

A general purpose VM has no GPU, so GPU-accelerated training needs a GPU node type. A single node cluster has no workers, so work cannot be distributed across nodes. Databricks Runtime ML preinstalls common libraries such as scikit-learn and PyTorch for Python use.

Instructions added to the Genie space itself apply only to that space, so Space1 mapping the words to Sales1 leaves other spaces unaffected. Table, schema or catalog comments affect every space using them, and a notebook is not a Genie knowledge store.

Unity Catalog captures runtime lineage automatically for queries run on Unity Catalog-enabled compute, with no extra setup. Catalog Explorer shows the lineage graph on each table's Lineage tab. Log Analytics and Azure Data Explorer are not lineage viewers, and a DAG describes job task flow.

delta.deletedFileRetentionDuration set to 30 days controls how long removed data files are kept before VACUUM can delete them, and delta.logRetentionDuration keeps the history needed for time travel audits over the same period. Archive, deletion vectors and auto compaction do not govern retention.

Lakeflow Connect ingests data by copying it into Delta tables in Databricks-managed storage, which violates the no-copy requirement. Lakehouse Federation with a foreign catalog over the SQL Server connection queries it in place and shows it beside Unity Catalog objects.

Setting readChangeFeed to true streams row-level inserts, updates and deletes from the change data feed. Setting ignoreDeletes to false ensures delete events are not dropped, and false is also its default. ignoreChanges is deprecated in favour of skipChangeCommits, so avoid it in new code.

dropna with subset=["order_amount"] removes only rows where that column is null and keeps other columns' nulls untouched. This produces the required DataFrame.

A time series profile in Lakehouse Monitoring computes metrics, such as null counts, over time windows as new data arrives, letting you see trends. Snapshot profiles recompute the full table each time without history, and inference profiles are for model inputs and outputs.

The Spark UI shows runtime execution detail, while cluster event logs only record lifecycle events such as resizing and terminations. Its Executors tab reports memory use, garbage collection time and failures per executor, which is where out-of-memory problems show up most directly.

Configuring the notification on job failure fires only once the whole job has completed with a failure. Task-level failure alerts would also notify for non-critical tasks, and success events are not wanted. Non-critical tasks can be set to continue the run on failure.