Microsoft DP-420 Practice Test

DP-420 practice test: what's covered

• Azure Cosmos DB Developer Specialty

Free DP-420 sample questions

1. A retail team models an order that contains up to 5 line items always read and written together with the order header. Each line item is small and never queried independently. Which modeling approach minimizes request units for the common read?

   • A. Embed the line items as an array inside the order document so a single point read returns everything
   • B. Store each line item as a separate document referencing the orderId in a different container
   • C. Create one container per customer and write line items as standalone documents
   • D. Split the order header and line items across two databases joined at query time
   Show answer

   Correct: A. Embedding fits 1:few data that is read and written together, so one point read retrieves the order and its items cheaply. Referencing into a separate container is incorrect because it forces extra cross-document lookups. Per-customer containers and cross-database splits add complexity and cost, not savings, because they fragment data that belongs in one item.

2. A product catalog item embeds a large reviews array that grows unbounded as customers comment, and reviews change far more often than the product. The item is approaching the per-item size ceiling. How should the model be refactored?

   • A. Reference reviews as separate documents in their own container keyed by productId, keeping the product item small
   • B. Embed every review but compress the product document with gzip before writing it
   • C. Keep embedding reviews and raise the item size limit in the container settings
   • D. Move the reviews array into the partition key of the product document
   Show answer

   Correct: A. Unbounded, independently changing 1:many data should be referenced, not embedded, so the product item stays small and well under the 2 MB ceiling. Compression and a configurable size limit are incorrect because the 2 MB item cap is fixed and cannot be raised. Putting reviews in the partition key is nonsensical because keys must be a scalar property, not an array.

3. A multi-tenant SaaS app writes telemetry where one tenant generates 90% of traffic, creating a hot logical partition when tenantId is the key. The team must keep tenant-scoped queries efficient while spreading writes. What is the recommended technique?

   • A. Build a synthetic partition key like tenantId plus a bucket suffix to fan large tenants across many logical partitions
   • B. Switch the container to use the document id alone as the partition key
   • C. Disable indexing on the telemetry container to reduce write pressure
   • D. Lower the container throughput so the hot tenant is rate limited evenly
   Show answer

   Correct: A. A synthetic key combining tenantId with a random or hashed suffix raises cardinality so a heavy tenant spreads across partitions while queries can still scan its buckets. Using id alone loses tenant locality and is incorrect. Disabling indexing does not fix skew, and lowering throughput just throttles everyone rather than removing the hot partition.

4. An IoT solution stores readings and frequently filters by tenantId, then by deviceId, then by date. The team wants partition pruning on prefix queries plus the ability to scale each tenant independently beyond a single logical partition. Which feature fits?

   • A. Define hierarchical partition keys with up to three levels such as tenantId, deviceId, and date
   • B. Use a single partition key on date and rely on cross-partition fan-out
   • C. Create a composite index on tenantId and deviceId and keep a single-level key
   • D. Enable the analytical store so queries read from Synapse instead
   Show answer

   Correct: A. Hierarchical (subpartition) keys support up to three levels and let prefix queries like filtering on tenantId get partition pruning while each prefix scales past the single-partition limit. A date-only key concentrates writes and is incorrect. A composite index speeds ordering but does not change partitioning, and the analytical store serves analytics, not transactional prefix routing.

5. A team is choosing a partition key for a high-volume container and wants to avoid hot partitions and storage limits. Which characteristics describe a good partition key choice? (Choose THREE)

   • A. High cardinality with many distinct values so data and requests spread evenly
   • B. A value that appears in the filter of the most frequent queries for in-partition reads
   • C. A distribution that keeps any single logical partition under 20 GB of storage
   • D. A monotonically increasing timestamp so all writes land in the newest partition
   • E. A boolean status flag because only two values are easy to reason about
   Show answer

   Correct: A and B and C. Good keys have high cardinality, match common query filters for cheap in-partition reads, and keep each logical partition under the 20 GB ceiling. A monotonic timestamp is wrong because every write hits one hot partition. A boolean flag is wrong because two values give terrible distribution, concentrating data and throughput instead of spreading them.

Last updated: 2026-10-19

DP-420 practice test FAQ

Related practice tests

• Microsoft DP-900 practice test
• Microsoft DP-300 practice test
• Azure AZ-305 practice test
• Azure AZ-104 practice test
• Microsoft DP-700 practice test
• Azure AZ-204 practice test