Amazon Web Services Updated DVA-C02 Exam Questions and Answers by arnie

AWS AppConfig , a capability of AWS Systems Manager, is specifically designed to manage feature flags , configuration changes, and application toggles safely and dynamically. AWS documentation highlights AppConfig as the recommended service for enabling and disabling application features without redeploying code.

Feature flags in AppConfig allow developers to control visibility, rollout strategies, and gradual releases across environments. AppConfig supports validation, monitoring, and rollback to prevent misconfigurations from impacting users.

Option A is incorrect because AWS AppSync is a GraphQL service and does not manage feature flags. Options B and D misuse data storage and synchronization services for configuration management, introducing unnecessary complexity and risk.

Using AWS AppConfig enables developers to hide features , activate them instantly when ready, and manage them centrally with minimal operational overhead.

Therefore, AWS AppConfig feature flags are the correct and AWS-recommended solution.

Amazon DynamoDB performance depends heavily on partition key design . AWS documentation warns against access patterns that repeatedly target the same partition key, as this causes hot partitions , even in on-demand mode.

In this scenario, most reads focus on the current day’s forecast for a small number of large cities. Using CityId alone as the partition key would cause hot partitions for those cities. Adding a calculated suffix (such as a hash or modulo value) to CityId spreads read requests across multiple partitions while still allowing efficient querying.

Using ForecastDate as the partition key (Options C and D) would concentrate traffic on a single value (today’s date), creating severe hot partition issues. Option B lacks a meaningful access pattern and complicates queries.

AWS documentation explicitly recommends key-suffix techniques for workloads with skewed access patterns to distribute traffic evenly. Therefore, using CityId with a calculated suffix as the partition key is the correct design.

The least implementation effort is to publish a custom CloudWatch metric from the existing Lambda function that records the number of orders in each batch, then create a CloudWatch alarm that sends a notification to the SNS topic when the metric value is 0. This avoids building a separate stream-processing pipeline or scheduled analytics job. A new Kinesis consumer and Lambda function would add unnecessary services and operational complexity. A 24-hour scheduled function is also slower than required because the notification must happen as soon as possible. Logging to Kinesis only creates more data movement and still requires separate analysis logic. CloudWatch supports custom metrics, and CloudWatch alarms can send notifications through Amazon SNS when metric thresholds are breached. ( AWS Documentation )

===============

The described behavior is a classic cache stampede (also called the “thundering herd” problem): a large set of hot keys share the same expiration time, so they all expire simultaneously. When that happens, many requests miss the cache at the same moment and fall back to the backend database, creating a sudden spike in database load and causing increased latency and poor user experience.

The most effective fix that preserves freshness is to introduce jitter —a small, randomized variance—into the TTL values so cached objects do not all expire at the same time. With TTL jitter, each item still has a bounded lifetime (so freshness is maintained), but expirations become spread out over time. That smooths cache-miss rates and prevents synchronized regeneration, reducing bursts of database traffic. This is a common best practice in high-load caching systems precisely to avoid mass expiration events.

Option A (increase TTL) might reduce how frequently a stampede occurs, but it does not solve the root cause: keys can still align and expire together, and a longer TTL can also reduce freshness if product data changes. Option B (more replicas, disable cluster mode) changes topology and read scaling, but it does not prevent synchronized expiry or the resulting backend spike; disabling cluster mode could also reduce scale for large keyspaces. Option D (more shards) can improve cache throughput capacity, but it does not address the core issue that backend load spikes occur when items expire simultaneously; lowering replica count could even reduce read availability.

Therefore, C is the correct solution: keep TTL-based freshness, but add a randomized component (for example, TTL = baseTTL + random(0..jitter)) so expirations are distributed and the backend database is protected from sudden bursts.