Free Databricks Certified Data Analyst Associate practice test: 30 questions with answers

Query profile shows the operators in a query, the time each takes, and data read, so the analyst can see the slow step. Audit logs record who did what, lineage shows upstream and downstream dependencies, and the cluster size setting changes capacity rather than explaining the slow step.

Liquid clustering organizes data by the chosen columns and adapts as data grows, without fixed partition layout decisions. Manual daily folders with unions add maintenance and do not adapt, a larger warehouse does not reorganize the data, and CSV is a slower format with no clustering.

Query history and audit logs record the user, statement and time for queries, and Unity Catalog audit events show table access. A dashboard schedule shows refresh timing, the comment field is free text that is not updated automatically, and the Delta transaction log records table commits rather than who queried the table.

Filters on the clustering or partition column let the engine skip files, so data read falls when the query uses that column in its predicate. Warehouse size changes speed rather than the amount read, dashboard titles have no effect on scans, and ALL is not a keyword that controls file sorting.

A view stores the query definition, so colleagues can select from it and the filter stays in one place. A temporary table disappears when the session ends and needs manual refresh, a CSV export is a static copy, and a dashboard filter must be reapplied by each user.

A left outer join keeps every row from the left table, here orders, and fills region columns with nulls where no match exists. An inner join drops unmatched orders, a right join keeps unmatched regions rather than orders, and a cross join multiplies every order by every region.

HAVING filters groups after aggregation, so the condition on the summed units is applied per category. A WHERE clause cannot reference an aggregate, ORDER BY with LIMIT returns a fixed number of rows rather than a threshold, and a filtered window function does not remove groups in the same way.

ORDER BY month DESC sorts newest first, and LIMIT 12 returns the first twelve rows, which is the Databricks SQL syntax. ASC would put the oldest month first, GROUP BY alone does not sort or limit, and TOP is not the Databricks SQL keyword for this.

A shared filter widget linked to a parameter on each dataset keeps every chart in step with one control. Copying dashboards multiplies maintenance, asking executives to edit SQL is not a usable design, and static images lose the interactivity the requirement asks for.

A line chart with time on the x-axis and one series per product line shows the trend for each line across the period. A pie chart shows a single total split and has no time dimension, a counter shows one value, and a scatter plot without a time axis hides the trend.

A scheduled refresh timed after the overnight load keeps the dashboard current each morning without manual action. Asking each person to refresh pushes work onto users, a weekly PDF is out of date between exports, and a zero cache time makes every view expensive without making the data fresher.

Units, time periods, a last-updated time and consistent colours help readers interpret the figures correctly. A 3D pie with many slices distorts and hides values, and untitled charts leave readers guessing what each figure means.

A row filter attached to the table applies automatically to every query, so users see only the rows they are allowed to see. Per-region tables require analysts to choose the right one, dashboard filters are optional for users, and a view that people must remember to use does not enforce the rule.

Granting SELECT only on the needed tables, adding USE privileges on parents, and granting to groups keep access narrow and manageable. Admin roles and ALL PRIVILEGES on production give far more than analysts need.

In Unity Catalog, a user needs USE CATALOG and USE SCHEMA on the parent objects to reach a table, along with SELECT on the table itself. MODIFY allows changes to rows rather than reads, CREATE TABLE is for making new tables, and metastore-wide privileges go well beyond what this analyst needs.

An AI/BI Genie space lets business users ask natural-language questions over a chosen set of tables, with instructions that define terms. A pipeline per question is impractical, a notebook still expects code from users, and Databricks Connect is for developers running code locally.

A clear instruction in the Genie space gives the model the business definition it needs, which fixes this class of error at its source. Renaming the table breaks other consumers, sharing more widely spreads wrong answers, and turning off the tool removes the self-service benefit without fixing the definition.

Can run lets users ask questions and see answers without editing the space's instructions or tables. Can manage and can edit both let users change the space, and making every user a workspace admin grants far more access than the requirement allows.

A column mask in Unity Catalog applies a function to the column at query time, so the returned value depends on the querying user's group. A separate copy duplicates data and needs upkeep, a comment does not enforce anything, and notebook output settings do not protect data in SQL results.

A cleaned table or view created by SQL gives analysts a governed, repeatable result, and access can be granted on it. Repeating cleanup in every query invites inconsistency, a hand-edited spreadsheet is not reproducible, and permanently dropping rows loses records that may be needed for audit.

Descriptions explain meaning and lineage to readers, and tags or naming conventions show which datasets are certified. A misc schema hides meaning, personal copies drift from certified data, and deleting descriptions removes information analysts rely on.

Unity Catalog is the governance layer that stores metadata for catalogs, schemas, tables, views and volumes across workspaces and applies one set of grants. Delta Live Tables builds pipelines rather than governing data, Databricks Connect runs local code against a cluster, and the workspace file browser only stores notebooks and files.

Serverless SQL warehouses are managed by Databricks, start quickly and need no cluster sizing by the team. An always-on all-purpose cluster is costly and managed by the team, a job cluster per query is slow to start, and a laptop is not Databricks-managed compute.

The SQL editor and SQL warehouses are the core of day-to-day SQL analysis in Databricks SQL. Init scripts, Terraform state and MLflow experiment tracking support infrastructure or machine learning work rather than everyday analysis.

Auto Loader with the cloudFiles source discovers and ingests new files incrementally, which avoids relisting the whole folder on each run. Manual uploads are not hourly automation, daily overwrites reprocess everything, and Delta Sharing is for reading data shared by another organization rather than loading local files.

Delta Sharing lets a recipient create a catalog from a share and query the shared tables live, without copying files. Laptop downloads and emailed exports are stale copies, and sharing storage credentials exposes secrets and bypasses the sharing controls.

The table creation UI can upload a small CSV and create a managed table from it that analysts can query with SQL. A text editor is not a query tool, converting to an image makes the data unusable, and loading into a notebook list is manual and not shared.

In a star schema, descriptive attributes such as name, category and brand belong in a dimension table keyed on product_id, and the fact table holds measures and foreign keys. Repeating names on every line adds storage and update risk, a second fact table mixes grains, and a dropped staging table is not a reporting model.

A star-style model keeps descriptive attributes in a dimension table, and a join on customer_id brings the region in for reporting. Copying full customer records into every line inflates storage and creates update problems, arrays of addresses make filters awkward, and removing the key makes customer analysis impossible.

An order table and an order lines table joined by a foreign key model the one-to-many relationship cleanly, and aggregation works directly. A text string of lines cannot be summed without parsing, joins on dates are unreliable, and one table per order cannot be queried as a whole.