What Is Azure Databricks?

Playback speed

Swipe, scroll or use ← →

Start with one platform, not three products

Before any cluster, catalog, or pipeline makes sense, you need to know what kind of thing Azure Databricks is. Azure Databricks is a unified, open analytics platform for building, deploying, sharing, and maintaining enterprise-grade data, analytics, and AI solutions at scale.

Read that definition as a promise about scope. It is not a storage service, not a reporting desktop application, and not a networking or identity product. It is the place where data work, analytical work, and AI work happen over the same data.

The platform integrates with storage and security in the customer's cloud account while managing and deploying cloud infrastructure. Your data and your security boundaries stay in your own cloud subscription; the platform takes over the job of standing up and running the infrastructure that processes them.

1 / 19

Start with one platform, not three products

"Unified" has a concrete meaning here: one foundation carries workloads that would otherwise live in separate products.

Workload	What people do with it
Data engineering	Ingest, clean, and model data so others can find and use it.
SQL analytics and BI	Run analytic queries and build reports over the same data.
Machine learning and AI	Train, track, and deploy models on that data.
Governance	Control who may see and use each data asset.
Streaming	Process new records continuously as they arrive.

Nothing in that list is a bolt-on product with its own copy of the data. That single point is the whole argument for the platform, and the next section explains what goes wrong without it.

2 / 19

Quick check

A new team asks what Azure Databricks actually is. Which description fits?
1. AAn archive service that retains raw enterprise files without analyzing them
  
  The platform processes and analyzes data at scale; storing untouched files is only a small part of what it does.
2. BA unified, open analytics platform for enterprise data, analytics, and AI
  
  Right. It is the open analytics platform where data, analytics, and AI solutions are built, deployed, shared, and maintained.
3. CA network security service for configuring identities and virtual networks
  
  It integrates with cloud storage and security, but its purpose is analytics rather than network and identity administration.
3 / 19
Why separate systems become the problem

Most organizations arrive with a warehouse for reporting and a lake for data science, and often a third copy for machine learning. Maintaining separate systems for engineering, reporting, and machine learning creates distributed data that must be built, maintained, and synchronized.

The cost is rarely the software licence. It is the synchronization work, the disagreement between two dashboards that read two copies, and the delay between a record arriving and every system agreeing on it.

4 / 19

Why separate systems become the problem

The architectural answer is the lakehouse. A data lakehouse combines the benefits of data lakes and data warehouses. It provides scalable storage and processing for organizations that want to avoid isolated systems for workloads such as machine learning and business intelligence.

System	Strength it brings	Weakness on its own
Data warehouse	Clean, structured data optimized for BI queries.	Often proprietary formats with limited machine learning support.
Data lake	Cheap, open storage for data of any nature and format.	Unvalidated data, so teams hesitate to report from it.
Lakehouse	Both: open formats with validated, query-ready data.	Requires the governance and table layers to be in place.

A lakehouse can establish a single source of truth, eliminate redundant costs, and help keep data fresh. Azure Databricks lets data engineers, data scientists, analysts, and production systems use the lakehouse as a shared source of consistent data.

5 / 19

Quick check

An organization keeps one data copy for BI and another for machine learning, spends real effort keeping them in step, and wants scalable processing over one consistent foundation. What direction fits?
1. AAdopt a lakehouse so BI and machine learning read the same governed foundation
  
  Right. The lakehouse exists to support different workloads over shared data, which removes the synchronization burden instead of managing it.
2. BKeep both copies and add more frequent synchronization jobs between them
  
  More synchronization preserves exactly the fragmented systems the lakehouse is meant to replace.
3. CMove the BI copy into local desktop files and leave machine learning on the lake
  
  A local file copy adds a third disconnected version rather than creating a single source of truth.
6 / 19
Data engineering on the shared foundation

Each workload sits on that same foundation, and the exam expects you to know which one owns which outcome. Start with the one this certification is about.

Data engineering makes data available, clean, and stored in models for efficient discovery and use. Notice that the definition is about the condition of the data, not about how large the cluster is or how often it runs. If analysts cannot find a table, or cannot trust it, the engineering job is not finished.

Azure Databricks combines Apache Spark, Delta, and platform tools so engineers can compose ETL logic in SQL, Python, or Scala and orchestrate scheduled jobs. Those three languages are the documented set for ETL work; configuration formats such as YAML and markup languages such as HTML are not part of it.

7 / 19
Quick check

Which pairing matches the stated purpose of data engineering with the languages used to compose ETL logic?
1. APermanently running compute sized for peak demand, using SQL, YAML, and markdown
  
  Data engineering is defined by the condition of the data rather than by compute sizing, and YAML and markdown are not the ETL language set.
2. BLocally stored reports disconnected from source data, using Python, HTML, and CSS
  
  Disconnected local reporting contradicts the shared-data purpose of the lakehouse, and HTML and CSS are presentation languages.
3. CData that is available, clean, and modeled for discovery, using SQL, Python, and Scala
  
  Right. The outcome is available, clean, modeled data, and the platform names SQL, Python, and Scala for composing ETL logic.
8 / 19
Keep your progress in the app

That’s 3 of 8 quick checks. In the app they stay answered, and every lesson remembers where you left off.

Get the app Keep reading here
Analytics, machine learning, streaming, and governance

The same foundation serves three more outcomes.

SQL warehouses provide compute for analytic queries, and users can query lakehouse data from the SQL editor or notebooks. Administrators configure scalable compute clusters as SQL warehouses so that business users can run queries without managing the complexities of working in the cloud. A SQL warehouse is the engine that runs the query; it is not where the data lives.

Azure Databricks machine learning includes tools such as MLflow and Databricks Runtime for Machine Learning. MLflow tracks and manages the model lifecycle, and the machine learning runtime prepares an environment with the libraries that model work needs.

9 / 19
Quick check

Business users need to run lakehouse queries through a SQL interface, and administrators want them insulated from cloud compute complexity. What should execute the queries?
1. AA Unity Catalog access control list scoped to the analytic tables
  
  An access control list decides who is allowed to query; it does not execute anything.
2. BA Git folder synchronized with the team's query repository
  
  A Git folder synchronizes project files with a git provider and supplies no query compute.
3. CA scalable SQL warehouse configured for analytic queries
  
  Right. Administrators configure scalable compute as SQL warehouses precisely so end users can query without managing cloud complexity.
10 / 19

Analytics, machine learning, streaming, and governance

Unity Catalog supplies a unified governance model for the data lakehouse. Privileges are managed through access control lists, using either the interface or SQL syntax, so securing a table does not require rebuilding cloud-native identity and networking rules.

Apache Spark Structured Streaming processes streaming data and incremental changes and underpins Lakeflow pipelines and Auto Loader together with Delta Lake. Auto Loader is the tool that incrementally and idempotently loads files from cloud object storage into the lakehouse.

Responsibility	Component
Continuous and incremental processing	Structured Streaming
Incremental file loading from object storage	Auto Loader
Unified governance of data assets	Unity Catalog
Model tracking and machine learning environments	MLflow and the machine learning runtime

11 / 19

Quick check

A team must process incremental records continuously and enforce one governance model over the resulting lakehouse data. Which pairing is right?
1. AStructured Streaming for the incremental processing and Unity Catalog for the governance
  
  Right. Structured Streaming handles streaming data and incremental changes, and Unity Catalog supplies the unified governance model.
2. BUnity Catalog for the incremental record processing and SQL warehouses for the governance model
  
  Unity Catalog governs rather than processes, and a SQL warehouse is query compute rather than a governance layer.
3. CMLflow for the incremental processing and Auto Loader for the governance
  
  MLflow serves machine learning work, and Auto Loader loads files incrementally instead of governing them.
12 / 19

Place the remaining product names

A handful of names appear constantly in Azure Databricks material and are easy to mistake for data-processing components. Fix them now so they stop being plausible answers later.

Name	What it actually does
Git folders	Sync Azure Databricks projects with supported git providers.
Declarative Automation Bundles	Define, deploy, and run resources such as jobs and pipelines programmatically.
Access control lists	Manage privileges over data assets through the interface or SQL.
Jobs	Schedule notebooks, SQL queries, and other code.
Notebooks	Author code in Python, R, Scala, or SQL with visualizations and commentary.

None of these five is the streaming engine, and none of them is the compute that answers an analytic query. That is why they make convincing wrong answers: they are real, useful, and belong to a different job.

13 / 19

Quick check

Which statement places these supporting names correctly?
1. AGit folders sync projects with git providers, and Automation Bundles deploy jobs and pipelines
  
  Right. Git folders handle source synchronization and Automation Bundles define, deploy, and run resources such as jobs and pipelines.
2. BGit folders provide the query compute, and Automation Bundles process streaming and incremental data
  
  Query compute comes from SQL warehouses and incremental processing from Structured Streaming, not from either of these.
3. CGit folders govern the table privileges, and Automation Bundles supply the machine learning runtime
  
  Privileges are governed through Unity Catalog access control lists, and the machine learning environment comes from the machine learning runtime.
14 / 19
A mental model that survives the whole course

Every later topic in this certification slots into a small frame. Learn the frame once, and new product names become easy to file.

Storage holds the data, while compute processes it. That single split explains most of the confusion beginners have: a table is storage, a warehouse or cluster is compute, and the two scale independently.

Engineering tools ingest and transform data, Unity Catalog governs it, and jobs and pipelines make work repeatable. SQL and machine learning tools use the shared lakehouse foundation for different outcomes.

15 / 19

A mental model that survives the whole course

Two distinctions are worth memorizing verbatim, because both are tested by swapping the two halves.

A SQL warehouse is query compute rather than stored data. Unity Catalog is the governance layer rather than the execution engine.

If you are asked about	Answer with
Where the data sits	Lakehouse storage, as Delta tables
What runs the analytic query	A SQL warehouse
Who is allowed to read it	Unity Catalog
What keeps the work repeatable	Jobs and pipelines

16 / 19

Quick check

Which pair of statements is correct?
1. AA SQL warehouse stores the governed lakehouse tables, and Unity Catalog executes every query against them
  
  This reverses both roles: warehouses do not hold the data and Unity Catalog does not run queries.
2. BA SQL warehouse is query compute, and Unity Catalog is the governance layer rather than the engine
  
  Right. Compute processes the data and governance decides who may use it; the two responsibilities stay separate.
3. CA SQL warehouse governs table privileges, and Unity Catalog stores the lakehouse data files
  
  Governance belongs to Unity Catalog rather than to a warehouse, and neither component is the storage layer.
17 / 19
Key takeaways
- Azure Databricks unifies data, analytics, and AI workloads on an open analytics platform.
- A lakehouse combines data-lake and data-warehouse benefits to reduce isolated systems.
- The platform supports data engineering, SQL analytics, machine learning, governance, and streaming over shared data.
- Storage and compute are separate concerns, and every product name belongs to one responsibility only.
Keep the frame handy: the platform is the setting, the lakehouse is the shared data foundation, and each named component earns its place by the single job it does.
18 / 19
Quick check

Which summary keeps the platform, the pattern, and the components in their right roles?
1. AA lakehouse is a reporting tool, Structured Streaming governs privileges, and MLflow stores lakehouse tables
  
  A lakehouse is a data management pattern rather than a reporting tool, streaming does not govern, and MLflow does not store tables.
2. BAzure Databricks is a storage archive, a SQL warehouse is a governance layer, and Unity Catalog runs ETL
  
  The platform analyzes rather than archives, a warehouse is compute rather than governance, and Unity Catalog does not execute ETL.
3. CAzure Databricks unifies workloads, a lakehouse joins lake and warehouse benefits, and each component owns one job
  
  Right. That is the platform, the architectural pattern it is built on, and the discipline of one responsibility per component.
19 / 19
8 quick checks · then the test

In the app, finishing the quick checks opens this lesson’s 10-question test, and the ones you miss come back exactly when you’re about to forget them.

Take this lesson’s test → Next lesson: Azure Databricks Platform Objects Get the app

← Previous Keep your progress · Get the app Next card →

The whole course, on your phone

Lessons you can read, audio you can listen to on the way to work, and practice that remembers what you got wrong.

Get the app

What Is Azure Databricks?

Start with one platform, not three products

Why separate systems become the problem

Data engineering on the shared foundation

Keep your progress in the app

Analytics, machine learning, streaming, and governance

Place the remaining product names

A mental model that survives the whole course

Key takeaways

8 quick checks · then the test

The whole course, on your phone