Hard — DP-420 Practice Quiz

Question 1

You maintain a Cosmos DB container named 'orders' partitioned by /customerId. A reporting dashboard runs the query SELECT VALUE MAX(c.orderTotal) FROM c to display the single largest order value across all customers. The dashboard runs this query once every few seconds and RU consumption has become a concern. Which approach reduces the RU cost of repeatedly obtaining this value most effectively?

Accepted Answer

Maintain a separate 'stats' item updated via the change feed that stores the running maximum orderTotal, and issue a point read against it from the dashboard.

Answer

Add an ORDER BY c.orderTotal DESC clause with a TOP 1 to the existing query so the engine can use the index more efficiently.

Answer

Enable autoscale on the container so the query automatically gets more RU/s when the dashboard runs.

Answer

Switch the SDK from gateway mode to direct mode so the aggregate query bypasses cross-partition fan-out.

Question 2

A retail analytics team enables Azure Synapse Link on a Cosmos DB for NoSQL container. The transactional data must be purged after 30 days to control storage costs, but the data science team needs to query up to 2 years of historical data through the analytical store via Synapse serverless SQL pools. How should the team configure retention to meet both requirements?

Accepted Answer

Set the container's transactional TTL to 30 days and set a separate analytical store TTL (AnalyticalStoreTimeToLiveInSeconds) to approximately 2 years.

Answer

Set the transactional TTL to 2 years so the analytical store retains the same data, then filter to 30 days in transactional queries.

Answer

Set the analytical store TTL to 30 days and rely on change feed to archive older records into blob storage for the 2-year requirement.

Answer

Disable TTL entirely because enabling Synapse Link automatically retains all analytical data for 2 years by default.

Question 3

You are building an audit-and-replication pipeline on a Cosmos DB for NoSQL container. The downstream system must observe every intermediate change to each item — including updates that happened between reads — and must also be notified when items are deleted (so it can remove them from a search index). You will consume the change feed using the .NET SDK with the change feed processor. Which change feed configuration meets these requirements?

Accepted Answer

Use the change feed in all-versions-and-deletes mode, which surfaces intermediate updates and delete operations within the configured retention window.

Answer

Use the change feed in latest-version (default) mode, which delivers only the most recent version of each changed item and includes deletes.

Answer

Enable Time to Live (TTL) on the container so that deletes are emitted, then read the latest-version change feed for updates.

Answer

Query the container with SELECT * ORDER BY _ts on a timer, since the change feed cannot report deletes under any mode.

Question 4

You built a change feed consumer using the .NET SDK's change feed processor to project order events into a downstream reporting container. To maximize throughput, you configured the processor to checkpoint only after every 500 processed changes rather than after each batch. During a load test, the host process crashed unexpectedly. When the processor restarts, what behavior should you expect, and what is the primary trade-off of your checkpoint configuration?

Accepted Answer

The processor resumes from the last checkpoint, so up to 500 already-processed changes may be delivered again, requiring your delegate to handle duplicates idempotently.

Answer

The processor resumes exactly where it stopped with no reprocessing, because the SDK persists an in-memory offset independent of the lease container.

Answer

The processor skips all changes that occurred while the host was down, causing permanent data loss in the reporting container.

Answer

The processor automatically re-reads the entire change feed from the beginning because a crash invalidates all lease documents.

Question 5

You are building a .NET consumer using the change feed processor from the Cosmos DB SDK to process an order-events container. During peak hours, a single processor host cannot keep up, and the change feed estimator reports growing lag. Your monitored container has 6 physical partitions. You deploy 3 additional identical instances of the same application (4 total), all configured with the same processor name, the same lease container, and the same instance-specific 'instanceName'... except two instances were accidentally deployed with an identical 'instanceName'. What is the effect on lease (partition) distribution and processing?

Accepted Answer

The change feed processor uses instanceName to uniquely identify each host; two hosts sharing the same instanceName can cause leases to be reassigned back and forth between them, leading to duplicate processing and unstable distribution.

Answer

instanceName is ignored by the processor; leases are always split evenly by processor name alone, so all 4 instances process correctly regardless of duplicate names.

Answer

The duplicate instanceName causes both affected hosts to be rejected at startup, so only 2 of the 4 instances will acquire leases and process changes.

Answer

Because there are 6 partitions and 4 instances, the processor automatically renames duplicate hosts using a GUID suffix, so distribution is unaffected.

Question 6

A retail catalog container stores product documents, each with a nested 'reviews' array containing objects like { rating, verified, text }. A developer needs a query that returns each product's id and title along with only the count of reviews that have a rating of 5 AND verified = true. The team wants to minimize RU charge by avoiding returning the full reviews array to the client and by evaluating the filter server-side. Which SQL query approach best meets these requirements?

Accepted Answer

Use a correlated subquery in the SELECT: SELECT p.id, p.title, (SELECT VALUE COUNT(1) FROM r IN p.reviews WHERE r.rating = 5 AND r.verified = true) AS topReviews FROM p

Answer

SELECT p.id, p.title, p.reviews FROM p, then filter and count the reviews in application code after retrieval

Answer

SELECT p.id, p.title, COUNT(r) FROM p JOIN r IN p.reviews WHERE r.rating = 5 AND r.verified = true GROUP BY p.id, p.title, with no subquery

Answer

SELECT p.id, p.title, ARRAY_LENGTH(p.reviews) AS topReviews FROM p WHERE ARRAY_CONTAINS(p.reviews, { "rating": 5, "verified": true }, true)

Question 7

You maintain an Azure Cosmos DB for NoSQL container storing customer orders, partitioned by /customerId. Each order document contains an array named items, where each element has a productId and quantity. A reporting service must return, for a single specified customerId, the number of distinct products that customer has ever ordered across all their order documents. You want the most RU-efficient query that returns a single scalar value. Which query approach should you use?

Accepted Answer

SELECT VALUE COUNT(1) FROM (SELECT DISTINCT i.productId FROM c JOIN i IN c.items WHERE c.customerId = 'C123') AS d

Answer

SELECT c.items FROM c WHERE c.customerId = 'C123', then count distinct productIds in application code after loading all documents

Answer

SELECT DISTINCT i.productId FROM c JOIN i IN c.items WHERE c.customerId = 'C123' and read the returned row count from the SDK

Answer

SELECT COUNT(DISTINCT i.productId) FROM c JOIN i IN c.items WHERE c.customerId = 'C123'

Question 8

A gaming company stores telemetry in a Cosmos DB container partitioned by /gameId. One extremely popular game generates over 40 GB of telemetry for its single gameId value, and writes for that game are now failing. The team wants to keep gameId as the primary way to group data while resolving the storage ceiling. Which redesign best addresses the problem?

Accepted Answer

Recreate the container with a hierarchical partition key of /gameId then /sessionId, distributing each game's data across many logical partitions.

Answer

Increase the provisioned throughput (RU/s) on the container so the large partition can accept more writes.

Answer

Enable autoscale on the container to automatically expand the storage limit of the logical partition.

Answer

Switch the account to serverless mode, which removes the 20 GB per-logical-partition storage limit.

Question 9

You are designing a Cosmos DB container for a logistics platform that stores shipment events. Each shipment belongs to a specific carrier, and there are only 8 carriers. Within a carrier, events are queried almost exclusively by shipmentId. The largest carrier will generate roughly 400 GB of event data over the container's lifetime. Most read queries filter by both carrierId and shipmentId, while a nightly report scans all events for a single carrier. Which partitioning strategy best balances query efficiency and avoids exceeding logical partition limits?

Accepted Answer

Use a hierarchical partition key of carrierId then shipmentId

Answer

Use carrierId alone as the partition key

Answer

Use shipmentId alone as the partition key

Answer

Use a synthetic partition key concatenating a random 1–100 suffix with carrierId

Question 10

A retail analytics team enabled the integrated cache via a dedicated gateway to reduce RU charges on a set of frequently repeated queries. Most queries now return RU charge of 0, but one high-volume query keeps showing full RU cost on every execution even though it is identical each time. Investigation shows the SDK client for that query sets the request consistency level to Strong, while the account default is Session. What is the most likely cause of the cache never being used for that query?

Accepted Answer

The integrated cache only serves requests made with Session or Eventual consistency; requests using Strong or Bounded Staleness bypass the cache entirely.

Answer

The MaxIntegratedCacheStaleness value must be explicitly set to zero for Strong consistency requests to be cached.

Answer

The integrated cache caches only point reads, so any query with a filter always incurs full RU cost regardless of consistency.

Answer

Strong consistency requires provisioned throughput above 100,000 RU/s before the integrated cache can be enabled.

Question 11

Your team enabled the Azure Cosmos DB integrated cache by provisioning a dedicated gateway and routing the .NET SDK client through gateway mode with the dedicated gateway endpoint. A frequently repeated query still shows full RU charges on most executions rather than the expected near-zero cost. Investigation reveals the application sets the request consistency level to Strong on each query. What is the most likely reason the integrated cache is not serving these queries?

Accepted Answer

The integrated cache only serves requests using Session or Eventual consistency; requests with Strong or Bounded Staleness bypass the cache.

Answer

The integrated cache only works with point reads, never with queries, so query results are never cached regardless of configuration.

Answer

The integrated cache requires the SDK to use Direct mode, so gateway mode disables caching entirely.

Answer

The integrated cache must be explicitly enabled per query with a MaxIntegratedCacheStaleness of 0 to force cache hits.

Question 12

A logistics company runs a globally distributed Cosmos DB account with multi-region writes enabled across four regions. Their shipment-tracking documents are frequently updated concurrently in different regions, and business rules require that when two updates conflict, the conflicting values must be merged into a single document by keeping the highest recorded package weight and concatenating the two status notes. Simple last-writer-wins behavior is not acceptable. Which conflict resolution configuration should they implement?

Accepted Answer

Use a Custom conflict resolution policy with a registered merge stored procedure that reads the conflicting versions and produces the merged document.

Answer

Use a Last Writer Wins policy configured on a numeric _ts-based path so the most recent write always prevails.

Answer

Use a Custom conflict resolution policy without registering a stored procedure so conflicts appear in the conflicts feed for the application to resolve, and Cosmos DB automatically merges them.

Answer

Switch the account to Strong consistency so that write conflicts can never occur across regions.

Question 13

Your team runs a production Azure Cosmos DB for NoSQL container provisioned with autoscale at a maximum of 10,000 RU/s. Users report intermittent HTTP 429 responses during peak hours, yet the total RU/s consumed across the account stays well below the 10,000 RU/s ceiling in Azure Monitor. Which metric should you examine first to confirm the most likely root cause of the throttling?

Accepted Answer

Normalized RU Consumption, to detect whether a single physical partition is hitting its per-partition throughput limit

Answer

Server-Side Latency, to determine whether slow query execution is causing the 429 responses

Answer

Data Usage, to confirm whether the container has exceeded its storage quota

Answer

Total Request Units, averaged over the full account, to verify overall provisioning is adequate

Question 14

A blog platform stores each Post document and separate Comment documents, all partitioned by postId in the same container. Product wants each Post to display a commentCount without running an aggregate query on every page load. You need the commentCount on the Post to stay accurate whenever a comment is inserted, and the update must occur within the same logical operation. Which approach best satisfies this requirement?

Accepted Answer

Attach a post-trigger to the comment insert operation that reads the parent Post and increments its commentCount, since both documents share the postId partition key.

Answer

Attach a pre-trigger to the comment insert that queries all comments for the post and recalculates commentCount before the insert completes.

Answer

Create a UDF that computes commentCount at query time so the Post document never needs to store the value.

Answer

Use a change feed processor to asynchronously recompute and patch commentCount on the Post after comments are written.

Question 15

A developer notices that a frequently executed query filtering on 'c.status = "active"' has a high RU charge. Using the Query Metrics from the SDK response, they observe that the 'Index Hit Document Count' is much lower than the 'Retrieved Document Count', and 'Index Lookup Time' is minimal while 'Document Load Time' dominates. What is the most likely cause and appropriate remediation?

Accepted Answer

The query engine is performing a full scan because the 'status' path is not indexed; add a range index for '/status/?' to the indexing policy.

Answer

The integrated cache is disabled; provision a dedicated gateway to cache the query results and reduce RU charge.

Answer

The container uses autoscale throughput; switch to manual provisioned throughput to lower the per-query RU cost.

Answer

The partition key is poorly chosen; change the partition key to '/status' so the query targets a single partition.

Question 16

You are designing a Cosmos DB data model for a blogging platform. Each author writes many blog posts (an author can have thousands of posts). Author profile data (display name, bio, avatar URL) changes rarely, but when it does change it must be reflected consistently everywhere. Blog posts are read very frequently, and queries almost never need the full author bio — only the display name is shown next to each post. Which modeling approach best balances read performance and update consistency?

Accepted Answer

Store authors and posts in separate containers, reference the author by ID from each post, and denormalize only the author's display name into the post.

Answer

Embed the complete author profile document inside every blog post so no additional lookup is needed.

Answer

Store everything in a single document per author that embeds an array of all that author's blog posts.

Answer

Store authors and posts as separate documents but perform a cross-container JOIN at query time to always retrieve the latest profile fields.

Question 17

A financial services team is configuring an Azure Cosmos DB account that must be deployed across three Azure regions (East US, West Europe, and Southeast Asia). Their compliance team demands that every read return the most recent committed write with a linearizable guarantee. During design, an architect proposes enabling multi-region writes so that clients in each region write to their nearest region for lower write latency. What should you advise the team?

Accepted Answer

Strong consistency cannot be combined with multi-region writes; they must choose either strong consistency with a single write region or a weaker consistency level to enable multi-region writes.

Answer

Enable multi-region writes and select strong consistency, since strong consistency is supported across all write regions when automatic failover is enabled.

Answer

Enable multi-region writes with bounded staleness, which provides the same linearizable guarantee as strong consistency across regions.

Answer

Enable multi-region writes with session consistency, which guarantees globally that all readers see the most recent committed write.

Question 18

You maintain a Cosmos DB for NoSQL container of blog posts. Each document has a 'tags' array (up to 50 string entries) and a 'comments' array (up to 200 nested objects, each with an 'author' and 'likes' property). A dashboard query must return each post's title plus only the comments where 'likes' is greater than 100. The current query uses JOIN over the comments array and filters in the WHERE clause, but it returns a Cartesian-product row per matching comment and consumes high RU. What is the most efficient way to rewrite the query so each post yields a single result with the filtered comments?

Accepted Answer

Use a subquery in the SELECT projection: SELECT p.title, (SELECT VALUE c FROM c IN p.comments WHERE c.likes > 100) AS topComments FROM p

Answer

Keep the JOIN but add DISTINCT to collapse the duplicate rows into one row per post

Answer

Use ARRAY_CONTAINS(p.comments, {'likes': 100}, true) in the WHERE clause to filter posts

Answer

Add a composite index on (/comments/[]/likes ASC) so the JOIN no longer produces Cartesian rows

Question 19

A logistics company stores shipment events in a Cosmos DB container. Each event has a warehouseId (only 8 distinct values, causing hot partitions) and a timestamp. Most queries filter by warehouseId AND a date range (e.g., all events for a warehouse in a given month). The team wants to spread writes more evenly while still allowing efficient date-range queries scoped to a single partition. Which partition key design best meets these needs?

Accepted Answer

Use a synthetic key that concatenates warehouseId and the year-month, e.g. "WH03-2024-05"

Answer

Use warehouseId alone as the partition key

Answer

Use a synthetic key that concatenates warehouseId and a random number from 1 to 100

Answer

Use the raw timestamp as the partition key

Question 20

You are designing a Cosmos DB container for an IoT telemetry workload. Devices write readings continuously, and there are only 12 device types. You initially chose deviceType as the partition key, but during load testing you observe severe throttling (429 errors) concentrated on a few logical partitions, while total provisioned throughput is far from being consumed. The application always queries telemetry filtered by a specific deviceType over a time range. Which change to the partition key strategy best resolves the hot-partition problem while keeping queries efficient?

Accepted Answer

Create a synthetic partition key by concatenating deviceType with a bucket suffix (e.g., deviceType_0 through deviceType_9) to spread writes across more logical partitions.

Answer

Switch the partition key to a random GUID generated per reading so every document lands in a unique logical partition.

Answer

Keep deviceType as the partition key but increase provisioned throughput with autoscale to eliminate the 429 errors.

Answer

Change the partition key to the reading's timestamp so writes are always distributed by time.