Design a Shopping Cart Service

To design a highly available shopping cart service, I first define the core constraints: it must be stateless to allow easy scaling, support sub-second latency for mobile users, and handle millions of concurrent writes. For storage, I would recommend a hybrid approach. We use Redis as the primary cache for active carts because its in-memory nature guarantees the lowest latency, which is critical for the checkout experience. However, since Redis can lose data on restarts, we asynchronously persist snapshots to DynamoDB. This leverages DynamoDB's massive throughput and durability. Regarding consistency, we accept eventual consistency between Redis and DynamoDB to prioritize availability and speed. If a user adds an item, we update Redis immediately and return success. The write to DynamoDB happens in the background. To handle conflicts when a user edits a cart on two devices simultaneously, we use vector clocks or last-write-wins with versioning to merge changes safely. Finally, to ensure statelessness, the API gateway handles routing based on a UserID hash, directing requests to the correct shard. This architecture aligns with Amazon's focus on high availability and customer-centric performance.