Picking a Cosmos DB API that does not match the workload
The wrong answers choose a Cosmos DB API without matching it to the workload. An existing Azure Table Storage application is offered the NoSQL API or Cassandra, a MongoDB application is offered the NoSQL API, and a Cassandra application is offered MongoDB or Gremlin. Each compatible API keeps the existing programming model, drivers or query language such as CQL, so a migration takes the matching API: Table, MongoDB or Cassandra. The NoSQL API, which stores JSON documents queried with a SQL-like syntax, is the native choice for new applications.
See it in a sample questionConfusing Blob, Files, Table and Queue storage
The wrong answers hand a workload to the service next door. Images, video and archives are offered to Azure Files or Table storage, an SMB network drive is offered to Blob containers, and messaging between components is offered to Tables or Blobs. Blob storage holds unstructured objects read through its API and suits streaming media. Azure Files is a mountable SMB share. Table storage holds key/value rows. Queue storage holds messages for asynchronous processing between components.
See it in a sample questionGiving non-relational stores relational features
The wrong answers credit Table storage or Cosmos DB with relational behaviour: foreign keys enforced between tables, an entity normalised across joined tables, or one fixed column layout defined upfront. Table storage has no foreign keys, relationships, stored procedures or views, and it denormalises data so one row holds an entire entity and rows can differ in their fields. A Cosmos DB container is schema-agnostic, so its items can have entirely different properties and adding a property needs no schema migration.
See it in a sample questionMixing up block, page and append blobs
The wrong answers attach one blob type's behaviour to another. Block blobs are offered for virtual machine disks, and append blobs are offered as the type behind a managed disk. Page blobs are optimised for random read and write, which is why Azure uses them for virtual disk storage and stores managed disks as page blobs. Block blobs suit discrete, large binary objects that change infrequently, and append blobs only add blocks to the end, with no update of existing blocks.
See it in a sample questionReplacing global distribution with throughput, caching or copying
The wrong answers offer other routes to low latency across continents: more throughput in one region, a content delivery network, a browser cache, a manual export and re-import, a scheduled nightly copy, or one primary region that every user reads from. None gives each user a nearby read and write replica. In Cosmos DB, global distribution means adding Azure regions to the account, after which the service replicates data automatically and users read from and write to the nearest regional replica.
See it in a sample question