Databricks Data Analyst Associate

The exam

45 scored multiple-choice questions · 90 min · pass mark not published · SQL · October 2025 guide. Section weights are the ones Databricks publishes on the certification page.

Weights

New names

Genie spaces → Genie Agents · Databricks Assistant → Genie Code · Delta Sharing → OpenSharing · DLT → Lakeflow pipelines · legacy dashboards retired: use AI/BI dashboards.

Components

Catalog Explorer

Browse catalog.schema.table, views, volumes, functions · managed vs external tables · permissions tab · Lineage tab (automatic, column level) · certified / deprecated via system.certification_status.

Marketplace

Third-party datasets, models and notebooks, delivered over OpenSharing: no copy, read-only, listed as a shared catalog.

Find and trust data

Catalog Explorer search and tags find certified datasets; query them with the three-level name. One owner per object: make it a group.

Tags and lineage

Tags on catalogs, schemas, tables, columns (UI or SET TAGS). Governed tags: allowed values, need ASSIGN. Lineage: upstream and downstream tables, columns, notebooks, jobs, dashboards.

Clean data in SQL

NULLs: COUNT(*) - COUNT(col), coalesce, WHERE col IS NOT NULL · Duplicates: GROUP BY … HAVING COUNT(*) > 1, SELECT DISTINCT · Fix values: trim, lower, try_cast (NULL instead of an error) · Range checks: BETWEEN. MERGE: WHEN MATCHED UPDATE / DELETE, WHEN NOT MATCHED INSERT.

SQL warehouses

Genie Code: /explain, /fix (diff view), /optimize; Diagnose error runs /fix.

Tables and views

Managed: UC storage, DROP deletes (UNDROP 7 days). External: LOCATION, DROP keeps files. CREATE TABLE … AS SELECT from a join or read_files(...) builds a unified dataset; CREATE OR REPLACE TABLE keeps history and grants. Dynamic view: session_user(), is_account_group_member().

Aggregates

• WHERE before grouping, HAVING after. Non-aggregated SELECT columns go in GROUP BY.
• COUNT(*) counts rows; COUNT(col), SUM, AVG, MIN, MAX ignore NULLs.
• APPROX_COUNT_DISTINCT: HyperLogLog++, within 5% by default.
• COUNT_IF(cond) → 0 when nothing matches; SUM(x) FILTER (WHERE …) → NULL.
• ROLLUP(a, b) → (a, b), (a), () · CUBE → every combination · GROUPING SETS → only those listed. grouping(a) = 1 on subtotal rows.
• Summary stats: notebook + › Data Profile; percentile_approx(x, 0.5) median; stddev consistency.

Joins and sets

Composite keys: ON a.k1 = b.k1 AND a.k2 = b.k2. Set operations match columns by position.

Window functions

• ROW_NUMBER unique · RANK 1, 1, 3 · DENSE_RANK 1, 1, 2 · NTILE(4) quartiles.
• LAG/LEAD: previous/next row. Growth: (x - LAG(x)) / LAG(x).
• Running total: ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW; 7-day average: 6 PRECEDING.
• x / SUM(x) OVER (): percent of total. QUALIFY rn <= 3: top N per group, no subquery.

Functions and time travel

date_trunc('MONTH', d) → TIMESTAMP at midnight on the 1st · datediff(end, start) · coalesce first non-NULL · nullif(a, b) NULL when equal · collect_list keeps duplicates, collect_set drops them · concat_ws('-', …) · DECIMAL for money, TIMESTAMP for date + time, ARRAY for lists.

SELECT … VERSION AS OF 12 · TIMESTAMP AS OF '…' · RESTORE TABLE … TO VERSION AS OF 12. VACUUM keeps 7 days by default; older versions can't be read afterwards.

Federation

Lakehouse Federation: CREATE CONNECTION, then CREATE FOREIGN CATALOG; query in place, read-only, pushed down over JDBC; pro or serverless warehouse.

AI/BI dashboards

• Datasets (table, view or SQL) on the Data tab; up to 15 pages; widgets: visualizations, filters, text (Markdown, links, images).
• Edit the draft; viewers see the published version until you republish.
• Publish with shared data permissions (default: publisher's) or individual (viewer's own).
• CAN VIEW · CAN RUN · CAN EDIT · CAN MANAGE. Account users without workspace access: published only, at most CAN RUN.
• Schedule refresh (cron allowed); up to 100 subscribers get a PDF by email, or PNG + PDF in Slack or Teams.
• Embed: iframe; basic needs a Databricks sign-in, embedding for external users uses a service principal.

Parameters vs filters

Parameters change the SQL: WHERE region = :region; date ranges use :p.min / :p.max. No {{ }} in dashboards. Filters slice dataset results; global, page or widget level.

Pick the chart

Notebook or SQL editor: + above the results › Visualization.

Alerts

Scheduled query + condition (first value or SUM/AVG vs threshold) → users or destinations: email, Slack, Teams, PagerDuty, webhook. States OK, TRIGGERED, ERROR. No parameters.