Task 2.2: Design data storage solutions for semi-structured and unstructured data

• A/region

  Incorrect

  A handful of regions gives few distinct values, so a few partitions would take most of the writes.

• B/readingDate

  Incorrect

  Every write of a day lands in the same partition, creating a hot partition for the current date.

• C/deviceModel

  Incorrect

  Only a few dozen models exist, so the key has low cardinality and popular models become hot partitions.

• D/deviceId

  Correct

  Device ID has very high cardinality, spreads writes evenly, and lets the common read target a single logical partition.

A good partition key has a wide range of values, spreads request units and storage evenly, and matches the most frequent query filter. Device ID meets all three; dates and low-cardinality attributes concentrate writes.

• ACold

  Correct

  Cold is an online tier for data accessed rarely and kept at least 90 days, with lower storage cost than cool.

• BArchive

  Incorrect

  Archive is offline; a blob must be rehydrated, which can take hours, before it can be read.

• CCool

  Incorrect

  Cool is online too, but costs more to store than cold for data kept this long and read this rarely.

• DHot

  Incorrect

  Hot has the highest storage cost and suits frequently accessed data.

Hot, cool and cold are online tiers; archive is offline and needs rehydration. Cold suits data that is rarely accessed but must be available immediately, and keeps a 90-day minimum retention.

• ABlob versioning

  Correct

  Versioning keeps the previous version each time a blob is overwritten, so it can be restored.

• BA legal hold on each container

  Incorrect

  A legal hold blocks overwrites and deletes entirely, which would stop the app from writing normally.

• CContainer soft delete

  Correct

  Container soft delete keeps a deleted container and its blobs for the retention period so it can be restored.

• DGeo-zone-redundant storage for the account

  Incorrect

  Replication copies overwrites and deletes to the other copies, so it doesn't bring back lost data.

• EA lifecycle rule that moves blobs to the cool tier after 30 days

  Incorrect

  A tier change keeps only the current blob; it adds no recovery copies of overwritten blobs or deleted containers.

Microsoft's data protection guidance for blobs combines container soft delete, blob soft delete and blob versioning. Versioning restores overwritten blobs, and container soft delete recovers a deleted container within its retention period.

• AA standard general-purpose v2 account that keeps the default flat blob namespace

  Incorrect

  In a flat namespace, a directory rename means copying and deleting every blob, and there are no directory ACLs.

• BAn Azure Files share in the transaction optimized tier

  Incorrect

  Azure Files serves SMB and NFS shares, not the Data Lake APIs that Spark analytics jobs use.

• CA storage account with the hierarchical namespace enabled (Data Lake Storage)

  Correct

  The hierarchical namespace provides real directories with ACLs and atomic directory rename and delete for analytics engines.

• DAn Azure Cosmos DB for NoSQL container with analytical store

  Incorrect

  Cosmos DB stores JSON items, not 300 TB of files with directory ACLs.

Data Lake Storage is Blob Storage with a hierarchical namespace: files sit in directories, directory operations are atomic, and access control lists follow the POSIX model, which Hadoop and Spark workloads expect.

• AStrong

  Incorrect

  Strong gives every reader the latest write, at the cost of higher latency and quorum reads, more than this app needs.

• BEventual

  Incorrect

  Eventual doesn't guarantee that users see their own writes right away.

• CSession

  Correct

  Session guarantees read-your-writes within a client session while keeping single-replica reads.

• DConsistent prefix

  Incorrect

  Consistent prefix preserves order but doesn't guarantee read-your-writes.

Azure Cosmos DB offers five consistency levels. Session, the default, guarantees read-your-writes and write-follows-reads within a single client session, with read latency and throughput close to the weaker levels.

• AA timer-triggered function that deletes old carts each night

  Incorrect

  That's the scheduled delete job the requirement avoids.

• BA lifecycle management rule on the account

  Incorrect

  That feature doesn't apply to Cosmos DB items.

• CTime to live set on the container to 30 days

  Correct

  With TTL, Cosmos DB deletes items automatically after the period since their last modification, using leftover request units.

• DThe analytical store with a 30-day retention

  Incorrect

  Analytical store keeps a columnar copy for analytics; it doesn't remove carts from the transactional store.

Time to live deletes items automatically after a set number of seconds since they were last modified. It can be set per container and overridden per item, and deletes run in the background with leftover request units.

• AAn HDD (standard) Azure file share over SMB

  Incorrect

  HDD file shares don't deliver the latency or throughput of a high-performance NFS cluster.

• BBlob Storage mounted with BlobFuse

  Incorrect

  BlobFuse presents object storage as files; it isn't a low-latency NFS file system.

• CAzure NetApp Files

  Correct

  Azure NetApp Files offers NFS and SMB volumes with sub-millisecond latency, snapshots and service levels that can change dynamically.

• DManaged disks attached to one VM that exports them over NFS

  Incorrect

  A self-built NFS server is a single point of failure and has to be managed by the team.

Azure NetApp Files is an enterprise file service for performance-heavy workloads such as HPC and SAP, with NFS and SMB, sub-millisecond latency, snapshots and service levels that can be changed without data movement.

• AStandard (manual) provisioned throughput of 40,000 RU/s

  Incorrect

  Manual throughput bills 40,000 RU/s around the clock, including the quiet hours.

• BStandard (manual) throughput of 2,000 RU/s, raised by a script before sales

  Incorrect

  Sales start at unpredictable times, so a script would raise throughput too late.

• CA serverless account

  Incorrect

  Serverless accounts run in a single region and cap throughput per partition, so they don't fit this two-region spike.

• DAutoscale provisioned throughput with a maximum of 40,000 RU/s

  Correct

  Autoscale scales instantly between 10% and 100% of the maximum and suits variable or unpredictable traffic.

Autoscale throughput scales the RU/s between 10% of the configured maximum and the maximum based on usage, which fits variable or unpredictable workloads. Manual throughput stays fixed, and serverless is limited to one region.