Data-Engineer-Associate - Data Ingestion and Loading (21% of the exam) - Section 2.7

Ingest semi-structured and unstructured data, such as JSON and nested data, via Lakeflow Connect and other managed connectors into Unity Catalog governed Delta tables.

Ingest JSON and nested records into Delta tables, choosing between inferring a struct schema, storing raw strings, or using the VARIANT type. Land unstructured files in Unity Catalog volumes and reference them from tables.

JSON ingestionnested structsVARIANT typeUnity Catalog volumesDelta tables

Practice question for this objective

Free sampleData Ingestion and Loadingmedium

A data engineer calls a paginated vendor REST API from a notebook and must land each page of raw JSON in cloud storage before any parsing, so that a later task can reprocess the untouched payloads. The landing area has to be governed and audited by Unity Catalog alongside the team's tables, and it must remain readable by a Lakeflow Job that runs the next morning on different compute. Where should the notebook write the raw files?

  • ATo a path inside a Unity Catalog volume, for example /Volumes/main/raw/api_landing, so the files are governed by the metastore. Correct
  • BTo the driver node's local file system under /tmp, and then register that directory as an external location on the same path.
  • CTo the DBFS root of the workspace, which every cluster in that workspace can reach without any further configuration.
  • DTo the cluster log delivery destination configured on the job compute, which keeps its contents after the compute terminates.
Unity Catalog volumes are the governed location for raw files landed by REST or other non-tabular ingestion in notebooks. A volume is a first class Unity Catalog securable that maps a catalog and schema path onto cloud object storage, so files written under /Volumes inherit the same privilege model and audit trail as tables in that schema. Any compute that can reach the metastore addresses the same path, which is what lets a later job read yesterday's payloads without workspace specific mounts.

Why A is correct: A volume is the Unity Catalog object designed to hold non-tabular files, so the payloads inherit metastore permissions and audit logging and any later compute can reach the same path by name.

Why B is wrong: Writing with ordinary Python file calls does land bytes in /tmp, which makes this look workable, but that storage belongs to the driver instance and disappears when the compute terminates, and an external location must point at cloud object storage.

Why C is wrong: The DBFS root is reachable and needs no setup, which is why it is tempting, but it sits outside Unity Catalog, so the files carry no metastore permissions and no governed audit record.

Why D is wrong: Log delivery does persist beyond the life of the compute, so the durability requirement appears satisfied, but it is a diagnostics destination for driver and executor logs rather than a governed data landing area.

See more Data-Engineer-Associate practice questions, answers explained.

Exam traps in Data Ingestion and Loading

Answers that look right on this material and are not. Each one is a distractor from a different question in the Data-Engineer-Associate bank for this domain.

  • A managed Delta table holding the file bytes in a BINARY column, loaded from the storage folder by a scheduled ingestion job.

    Why it is wrong: Tempting because it does place the content under Unity Catalog governance, but it copies the bytes into a table and removes the ability to open the original files by path, which the notebooks and tasks depend on.

  • Register the folder of PDF files as an external Delta table so that each document becomes one governed row inside that table.

    Why it is wrong: Tempting because external tables do govern data left in cloud storage, but a Delta table needs a tabular file format, and a folder of PDF documents cannot be read as one.

  • Both approaches copy the source tables into Unity Catalog tables, and the difference is that federation refreshes those copies far more frequently than an ingestion pipeline is able to.

    Why it is wrong: Tempting because both are governed through Unity Catalog and both start from a connection, but federation makes no copy at all. Framing the choice as a difference in refresh frequency misses what each one does with the data.

Examworthy is not affiliated with or endorsed by Databricks. Original, blueprint-aligned practice material only.