Free Databricks Data Analyst Exam Flashcards

Metastore -> Catalog -> Schema -> Table (or View/Volume/Function/Model). Every governed object is addressed with a three-level catalog.schema.object namespace.

Browse catalogs, schemas, tables, and volumes, and inspect permissions, tags, comments, and lineage without leaving the workspace UI.

An open marketplace for discovering and subscribing to shared data, notebooks, and AI assets from providers outside your own workspace.

It traces upstream sources and downstream consumers of a table or dashboard, letting an analyst confirm where data comes from and what would break if it changed.

Built-in, queryable tables that expose operational metadata such as audit logs, billing, and lineage, so admins and analysts can monitor platform usage with SQL.

A single lakehouse platform that unifies data engineering, SQL analytics, BI, and AI/ML, governed end-to-end by Unity Catalog.

A steward-reviewed label marking a dataset as trusted and production-ready; it signals review, not an automatic guarantee of data quality.

A custom tag is any searchable key-value metadata label an owner adds for discovery; it carries no built-in stewardship guarantee like a certified tag does.

COALESCE(a, b) returns the first non-null value; NULLIF(a, b) returns NULL if a equals b, otherwise returns a. Both are used to clean up data in SQL.

Use REGEXP_REPLACE to pattern-match and rewrite inconsistent substrings, often paired with a lineage check to confirm which upstream source introduced the issue.

Auto Loader incrementally and automatically ingests new cloud files as they arrive; COPY INTO is a simpler, idempotent SQL command better suited to scheduled or one-time batch loads.

Delta Sharing is a direct, point-to-point secure share from a specific provider that works cross-platform; Marketplace is a public or private catalog of listings you browse and subscribe to.

Serverless starts in seconds, Pro takes roughly four minutes to provision, and Classic is the slowest, entry-level tier.

Only Serverless warehouses use AI-based IWM to scale compute to bursty, unpredictable analyst workloads; Pro and Classic do not.

Dropping a managed table deletes the underlying data files because Unity Catalog owns them. Dropping an external table removes only the metadata reference; the files in your own storage location remain.

A view recomputes its query logic on every read, so it is always fresh but can be slow. A materialized view precomputes and stores the result, refreshing on a batch schedule, so reads are fast but can be stale.

Use a streaming table for continuous, append-only ingestion where each row is processed exactly once, typically at the Bronze layer; use a materialized view for batch-refreshed, BI-ready aggregations at Silver or Gold.

It rotates unique values from one column into multiple output columns, turning long, row-based data into a wide cross-tab summary.

Both operate element-by-element on array columns: FILTER keeps only elements matching a lambda condition, and TRANSFORM applies a lambda to map each element to a new value.

Use time travel syntax such as SELECT * FROM table VERSION AS OF n (or TIMESTAMP AS OF), which reads the table's historical state without restoring or copying data.

UNION ALL keeps every row from both queries, including duplicates. Plain UNION removes duplicate rows, which costs extra computation to de-duplicate.

Federated querying (Lakehouse Federation) lets Databricks SQL query external databases directly through Unity Catalog without first copying the data in.

Query History lists past query runs with duration and status, a list view. Query Profile drills into one execution's stage-level timings and bottlenecks, a deep-dive view.

The result cache reuses the output of an identical prior query with no recomputation. The disk cache stores remote file reads locally on the warehouse, speeding up repeated scans of the same files even if the query text differs.

It provides flexible, incrementally-maintained data layout on a table instead of fixed partitioning, so queries filtering on frequently changing columns skip more files and run faster.

Run DESCRIBE HISTORY on the table to see every version, operation, timestamp, and user that modified it.

It means an operation such as a join or sort exceeded available memory and had to write intermediate data to disk, a strong sign of a slow, under-resourced query.

Photon is Databricks' vectorized, native query execution engine that accelerates SQL and DataFrame workloads across all SQL warehouse tiers; it is an engine, not a separate warehouse type.

It uses learned statistics to intelligently skip irrelevant files and rows before scanning, reducing the I/O a query needs to perform.

Start in Query History to confirm the run and its duration, then open its Query Profile to see stage-level timings and identify which join or shuffle stage is the bottleneck.

A reusable, user-facing input, such as a date range or region picker, that filters every widget on a dashboard at once without editing each underlying query.

Scheduled refresh automatically re-runs the underlying queries at set intervals so viewers always see recently computed data, instead of requiring a manual reload.

An alert fires a notification automatically when a query result crosses a defined threshold condition, without a person needing to be watching the dashboard.

A controlled-access link that lets you distribute a dashboard to a defined external or internal audience without granting them full workspace access.

It surfaces the live AI/BI Dashboard inside another application or portal, so stakeholders view it without opening the Databricks workspace directly.

It organizes related visuals into separate pages within one dashboard, keeping a large set of metrics navigable instead of cramming everything onto one screen.

A notebook chart is a one-off visualization living inside a single cell for exploration. An AI/BI Dashboard is the shareable, scheduled, alertable artifact meant for recurring stakeholder use.

Dashboard permissions, set independently from the underlying tables' Unity Catalog grants, control which users or groups can view, edit, or manage the dashboard itself.

They teach Genie the business-specific terms, definitions, and context it needs to correctly interpret natural-language questions against your data.

They show users example prompts they can ask, guiding them toward the kinds of questions the space is curated to answer well.

Verified queries or views that Genie is instructed to prioritize when answering, improving accuracy over letting it generate SQL from scratch every time.

Benchmarks track answer quality against a known set of questions and expected results, letting an owner measure and improve accuracy as the space evolves.

Use the built-in feedback mechanism to capture the flag, then adjust instructions, trusted assets, or curated datasets to correct the underlying cause.

Limiting Genie to curated datasets keeps its answers grounded in vetted, relevant tables and reduces the chance it queries the wrong or ungoverned data.

A star schema denormalizes dimension tables for fewer joins and faster BI reads. A snowflake schema normalizes dimensions further, reducing redundancy at the cost of more joins.

Bronze holds raw ingested data, Silver holds cleaned and joined data, and Gold holds curated, BI-ready models such as star schemas; modeling patterns typically get applied at the Gold layer.

Every object is addressed as catalog.schema.table (or view/volume/function), giving a consistent governed path instead of a bare table name.

RBAC (GRANT/REVOKE) is all-or-nothing object-level access control. Row filters and column masks go further, restricting which specific rows are visible or obscuring specific column values for users who already have object access.

The object's owner, or a user with the MANAGE privilege, controls granting; ownership is the root of delegated access control in Unity Catalog.

Apply a column mask so the raw PII value is obscured (or a row filter to hide rows entirely) for users without clearance, while the table itself remains queryable.