Design an Online Polling Service

To design a scalable online polling service for Meta's audience, I would first clarify that we expect millions of concurrent users during viral events, creating a classic read-heavy workload once polls are live. The system needs to handle rapid vote ingestion without blocking the user experience. I propose an architecture where the API Gateway routes requests to microservices. For the critical path of recording votes, we will utilize Redis. Since reading results is the dominant operation, storing vote counts in Redis hashes allows for O(1) access times, preventing database bottlenecks. We will use Redis's atomic INCR command to handle race conditions safely. For the write-heavy initial phase of creating a poll, we will write directly to a relational database like MySQL to ensure ACID compliance for poll metadata. However, actual vote tallies should be buffered in Redis and asynchronously flushed to the primary database in batches to reduce write amplification. This hybrid approach leverages Redis for its speed during the read-heavy lifecycle of a poll while maintaining data persistence. To handle global scale, we will shard our Redis instances by Poll ID, ensuring no single node becomes a hotspot. Finally, we must implement rate limiting at the gateway to mitigate DDoS attempts, a common concern for public-facing platforms.