AWS DVA-C02Free AWS Certified Developer — Associate practice test

Correct answer: B. Use a composite partition key such as tenantId#shardSuffix (e.g., tenantId#1 through tenantId#10) and createdAt as the sort key, then scatter-gather across shards at read time.

Write sharding by appending a random or deterministic suffix to the partition key (tenantId#N) distributes writes across multiple physical partitions, preventing hot-partition throttling for high-volume tenants. At read time you issue parallel Query requests for each shard and merge results — a well-known scatter-gather pattern. Option A creates a hot partition for any tenant with very high write throughput. Option C requires a full-table Scan, which is expensive and does not scale. Option D co-locates all tenant data under a single partition key value 'SERVICE', which creates the most extreme hot-partition scenario possible.

Correct answer: B. After a node failure the cache is cold for all keys that were resident on that node; the application must re-fetch from the origin database and re-populate the cache on each cache miss.

In the cache-aside pattern, the application checks the cache first; on a miss it reads from the primary data store and writes the result back to the cache. After a node failure (or primary failover), the replacement node starts empty for the keys it owned, so the first request for each evicted key is a cache miss that pays the full database round-trip cost. Option A is incorrect — ElastiCache Redis Cluster Mode Enabled and Multi-AZ both support automatic failover. Option C describes a client-side retry delay that is not a default behavior of the ElastiCache client. Option D is wrong because only keys on the failed shard are lost, not cluster-wide.

Correct answer: B. Generate a pre-signed URL for s3:PutObject using the AWS SDK with an expiration of 1800 seconds and share it with the partner.

Pre-signed URLs embed scoped, time-limited credentials derived from the signing IAM identity, allowing the holder to perform a specific S3 operation (PutObject) without needing AWS credentials of their own. A single pre-signed PUT URL supports objects up to 5 GB, which is exactly the S3 single-PUT object size limit. Option A exposes the bucket to any internet actor for the window and requires manual remediation. Option C introduces IAM user lifecycle management overhead and still requires the partner to hold credentials. Option D (Transfer Acceleration) routes data through CloudFront edge nodes to speed transfers but does not provide any access control or time-limiting mechanism on its own.

Correct answer: B. Another concurrent TransactWriteItems or TransactGetItems call is in progress on any of the same items at the same time.

DynamoDB Transactions implement optimistic concurrency; a TransactionConflictException is raised when two transactions attempt to modify or read the same item concurrently and one of them detects a conflict during its commit phase. Option A is incorrect — TransactWriteItems supports Put operations that create new items. Option C describes a ValidationException (the maximum total payload for a transaction is 4 MB; the per-item limit is 400 KB). Option D is incorrect because DynamoDB Transactions work within a single Region — Global Tables replicate asynchronously, so cross-region transactional guarantees are not available via this API.

Correct answer: A. Strongly consistent reads consume twice the read capacity units (RCUs) compared to eventually consistent reads.

DynamoDB charges one RCU per 4 KB of data for a strongly consistent read and 0.5 RCU for an eventually consistent read of the same item — effectively doubling the read cost. This is the documented trade-off for guaranteeing that a read reflects all writes that received a successful response. Option B is the opposite of the truth: strongly consistent reads are supported on base tables and LSIs, but NOT on GSIs (GSIs only support eventual consistency). Option C is fabricated — there is no fixed delay. Option D is false; DynamoDB Streams operation is independent of read consistency settings.

Correct answer: B. Place Amazon RDS Proxy between Lambda and RDS; configure the Proxy to pool and reuse database connections.

RDS Proxy is designed precisely for this scenario: it maintains a warm pool of persistent database connections and multiplexes thousands of short-lived application connections (from Lambda invocations) onto a much smaller set of actual DB connections, preventing max_connections exhaustion. Option A is wrong — Multi-AZ standby does not serve read traffic; it is purely for failover. Option C worsens the problem by keeping connections open longer, increasing the connection count per unit time. Option D (Aurora Serverless v1) scales compute capacity, not the max_connections limit relative to Lambda concurrency; RDS Proxy is still the correct tool even with Aurora Serverless.

Correct answer: B. Enable DynamoDB Streams and use a Lambda function to index new or updated items into Amazon OpenSearch Service; direct search queries to OpenSearch.

Amazon OpenSearch Service provides inverted-index-based full-text search with relevance scoring, fuzzy matching, and compound queries — capabilities DynamoDB is not designed for. Streaming changes via DynamoDB Streams → Lambda → OpenSearch keeps the search index near-real-time. Option A only supports exact or range partition-key lookups, not free-text relevance search. Option C (S3 Select) uses SQL predicates against columnar data and does not provide relevance ranking or stemming. Option D is incorrect — DAX is a read-through/write-through cache that accelerates DynamoDB API calls; it does not add search or ranking functionality.

Correct answer: B. S3 Multipart Upload; minimum part size is 5 MB.

S3 Multipart Upload splits a large object into independently uploaded parts (minimum 5 MB each, except the last part which has no minimum size requirement). Each part receives its own ETag; a CompleteMultipartUpload call assembles them. If the upload fails, only the in-flight part needs to be retried, enabling resume semantics. Option A (Transfer Acceleration) routes data through CloudFront edge nodes to speed transfers but does not itself enable resumable uploads. Option C overstates the minimum — 5 MB is the documented minimum per part. Option D (Byte-Range Fetch) is a download feature for parallel reads, not uploads.

Correct answer: B. S3 Select, which allows the application to issue a SQL expression against the object and retrieve only the matching fields server-side.

S3 Select evaluates a SQL expression (SELECT userId, email, planTier FROM S3Object) against JSON, CSV, or Parquet objects server-side and returns only the matching data, dramatically reducing transfer size and Lambda parsing overhead. Option A (Intelligent-Tiering) manages storage cost based on access frequency and has no concept of field-level projection. Option C (S3 Object Lambda) can transform responses on the fly but requires writing, deploying, and maintaining a Lambda function — S3 Select is simpler and purpose-built for server-side field projection. Option D (Batch Operations) is an asynchronous bulk job tool, not a per-request query mechanism.

Correct answer: B. DynamoDB TTL deletions are asynchronous and may lag the expiration timestamp by up to 48 hours; expired items can still be returned until the background deletion process removes them.

DynamoDB TTL uses a background process to identify and delete expired items; AWS documentation states this process can take up to 48 hours after the expiration timestamp. During that window, expired items remain visible to reads unless the application explicitly filters them (e.g., by comparing the TTL attribute value to the current epoch time in a FilterExpression). Option A is false — TTL is enabled per-table via the UpdateTimeToLive API call on any numeric attribute name; no GSI is required. Option C is false — you can choose any attribute name when enabling TTL on a table. Option D is false — TTL has no effect on write access; it is purely a background delete mechanism.