Fabric Data Engineer Flashcards & Memory Notes

A collaborative container for Fabric items such as lakehouses, warehouses, notebooks, pipelines, and semantic models, with its own capacity assignment and access roles.

The unified data lake for Microsoft Fabric: one logical lake per tenant that stores lakehouse and warehouse data in open formats under a hierarchical namespace.

Admin, Member, and Contributor already have broad workspace data access. A narrow OneLake role does not subtract that grant. For consumers, inspect Viewer or item-level access and all applicable OneLake roles, then test the real identity.

Supported item definitions across environments. Rule support is item-specific; notebooks can use a default Lakehouse rule. Validate target connections, permissions, and dependencies. Promotion is not a backup of every table’s data.

No. Git restores supported item definitions. Inspect and repair the committed data separately, then replay safely using the corrected code and recorded load boundaries.

Rules that filter which rows a user can see in a semantic model or warehouse based on identity or role, without duplicating tables per audience.

No. It obscures values in SQL results without changing stored data. Users with query access may infer values. Use suitable row and column permissions when the underlying data must be restricted.

It records classification and interacts with configured policies. Actual protection depends on the policy, supported item, and access or export path. A label name alone does not prove that access is blocked.

Pipelines orchestrate activities and schedules. Dataflow Gen2 provides low-code Power Query transforms. Notebooks run Spark or other code for custom logic.

When you need DAG-based orchestration with Airflow operators and dependencies beyond a single Fabric pipeline canvas.

A dedicated pool of compute resources that powers Fabric workloads; workspaces are assigned to a capacity that sets scale and feature availability.

No. It provides read-only T-SQL access to Lakehouse table data. Use a supported write path such as Spark. A Fabric Warehouse has a different T-SQL write surface; a stored procedure cannot bypass the Lakehouse endpoint’s boundary.

Connector support for the selected identity type, the required source and destination permissions, and network access. A workspace role alone does not authorize an external database connection.

Target connection bindings, parameters, permissions, dependencies, and a small test run. Do not assume notebook or semantic-model deployment rules remap every data pipeline connection automatically.

After the bounded batch has been applied successfully and validated. Make destination writes idempotent so a crash after the write but before marker persistence can be retried safely. Handle deletes and marker resets explicitly.

A pointer that makes data in another location—such as ADLS Gen2, Amazon S3, or another OneLake folder—appear inside a lakehouse without physically copying all files.

It continuously replicates supported operational databases into OneLake as mirrored tables so analytics can read near-current data with low engineering overhead.

For tabular prep with Power Query mashups, connectors, and low-code transforms when custom distributed Spark code is unnecessary.

Use a window partitioned by the business key, ordered by authoritative version and a deterministic tie-breaker, then keep row_number equal to 1. dropDuplicates on the key alone does not express which version wins.

A real-time ingestion and processing item that captures event sources, applies stream operations, and routes results to destinations such as Eventhouse or lakehouse.

High-volume time-series and telemetry analytics with Kusto Query Language (KQL), often fed by Eventstream or other real-time paths.

No. Window duration defines grouping. Event time, watermark or late-arrival policy, checkpoint/state behavior, and engine semantics determine how late events are handled. Hopping windows can overlap; tumbling windows do not.

Bronze holds raw or near-raw landings. Silver cleans and conforms. Gold serves business-ready aggregates and models for consumers.

Copy activity moves data between stores at scale with mapping. A dataflow activity runs Power Query Gen2 transforms as a pipeline step.

Warehouse supports its documented T-SQL table-write and query operations. The Lakehouse SQL analytics endpoint reads Delta table data; use Spark for Lakehouse row modifications. Check the supported SQL surface for the actual item.

Deduplicate by a stable event key before aggregation, for example with arg_max over the chosen version or arrival marker. Then summarize with bin(EventTime, 5m). Define tie rules and distinguish the query from an upstream streaming lateness policy.

No. A standard shortcut references existing data, while Eventhouse query acceleration for supported external Delta tables adds cache and indexing work. Evaluate hot-data coverage, synchronization, storage, and compute cost.

Batch Spark processes bounded datasets on a schedule or trigger. Eventstream handles continuous events with low-latency stream operators.

In the Monitoring hub and each item’s run history, which show pipeline, notebook, dataflow, and other job statuses, durations, and errors.

How Fabric capacity CU consumption is used across workspaces and operations so you can spot throttling risk and heavy workloads.

Open the failed activity’s output and error details, confirm connection and parameter values, then re-run after fixing the root cause rather than only retrying blindly.

It compacts small files into larger ones to improve read performance by reducing file-listing and open overhead.

It removes obsolete data files that are no longer referenced by the table after the retention period, reclaiming storage.

Readers or time-travel queries that still need older versions can fail if files are deleted before the required retention window.

Too many notebooks or Spark jobs competing for the same capacity, which can queue, slow, or fail workloads until you stagger schedules or scale capacity.

Inspect query plans, statistics, data types, predicates, projected columns, and concurrent workloads. Use supported Fabric Warehouse features rather than assuming SQL Server indexes or Synapse distribution recipes apply unchanged.

Source throughput, operator complexity, destination write speed, and capacity limits; scale or simplify the stream path before data falls further behind.

Pipeline failure notifications, Monitoring hub reviews, and optionally Azure Monitor or other alerting integrations for operational response.

It colocates related values in files so queries with predicates on those columns can skip irrelevant data and read less.

The report may use an Import model whose refresh failed or ran before the load, or it may point to a different source. Identify the model’s mode and source binding, then inspect the relevant refresh or query path. Pipeline success alone does not prove freshness.