DP-900 Course Notes

Relational data is organized into tables, rows, and columns. A row represents one instance of an entity, such as one customer or one order, while each column stores a named attribute with a defined data type. DP-900 questions usually signal relational storage with structured records, repeated transactions, joins, and a need for integrity between related tables.

• Confusing a primary key with a foreign key; the primary key identifies the row, while the foreign key points to a row in another table.

• If the question says each row needs a unique identifier, choose primary key.

NoSQL and non-relational questions start with data shape. Semi-structured data such as JSON and XML has tags or fields but can vary by record, while unstructured data includes files, images, audio, video, and documents with no fixed table schema. The service choice should match whether the data is a document, key-value item, graph, file, or analytical lake asset.

• Choosing Cosmos DB for every non-relational clue even when the scenario is really unstructured file storage for Blob Storage.

• If the data is images, video, PDFs, or backups, look for Blob Storage before database answers.

OLTP and OLAP are different workload patterns. OLTP handles current transactions with fast, reliable writes; OLAP handles analytical queries that summarize large volumes of historical data. If the question mentions aggregations, trends, reporting, or slicing by dimensions, it is pointing away from transactional design and toward analytical processing.

• Confusing OLTP and OLAP; OLTP is transactional, while OLAP is analytical and aggregation-heavy.

• If the question mentions measures such as sales amount or quantity, think fact table.

Azure SQL Database is a managed relational database service with deployment choices for different workload shapes. A single database fits an isolated application database, while an elastic pool shares compute across multiple databases with uneven or unpredictable usage. Serverless adds auto-scaling and auto-pause for intermittent workloads where paying for idle compute should be minimized.

• Choosing Azure SQL Managed Instance for every SQL workload instead of reserving it for migration scenarios that need instance-level compatibility.

• If multiple databases have uneven usage, think elastic pool.

Cosmos DB API selection is mostly about compatibility and data model. The API for NoSQL, often called Core in older material, fits native Cosmos DB document workloads. MongoDB, Cassandra, Gremlin, and Table APIs are chosen when the application model or existing client compatibility points to those ecosystems.

• Choosing the API for NoSQL when the scenario is really about compatibility with an existing MongoDB, Cassandra, Gremlin, or Table app.

• If the question says graph traversal or relationships between connected entities, choose Gremlin API.

OneLake is the unified logical storage layer in Microsoft Fabric. It is not the Fabric Warehouse; the warehouse is one Fabric workload that stores and queries relational warehouse data. Shortcuts are logical references to data, so Fabric items can use data in another location without physically copying it into a new storage area.

• Calling Fabric Warehouse the central data lake; OneLake is the unified storage layer, while Warehouse is a T-SQL warehouse workload.

• If the clue says unified logical storage across Fabric, choose OneLake.