The workload is a real-time leaderboard that needs extremely low latency: microsecond reads and single-digit millisecond writes. It also needs rapid failover (accepting writes in less than a minute after a primary node failure) and minimal operational overhead. Additionally, the data must be durable enough to persist and be used later for analytics through a data pipeline.

Amazon MemoryDB for Redis is a Redis-compatible, durable, in-memory database service designed for microsecond read latency and low-millisecond write latency while also providing durability through a distributed transaction log. MemoryDB is designed to be a primary database rather than a cache and supports Multi-AZ configurations with fast failover. These characteristics match the leaderboard’s latency requirements and the availability requirement to resume writes quickly after a failure. Because MemoryDB is fully managed, it has less operational overhead than running Redis on EC2.

Option C is the best choice because MemoryDB provides Redis compatibility, in-memory performance, Multi-AZ high availability, and durability that supports data persistence for downstream processing. The durability aspect is especially important for feeding analytics pipelines, because the data is not only cached but is retained as a database record.

Option A is not the best choice because Amazon ElastiCache for Redis is primarily used as a cache. While it can be configured for replication and can achieve low latency, it is not positioned as a durable primary data store in the same way as MemoryDB for Redis. For requirements that explicitly combine extremely low latency with persistence and database-like durability, MemoryDB is the more appropriate managed service with lower operational overhead than building custom durability mechanisms.

Option B is not suitable because Amazon RDS for MySQL typically cannot meet microsecond read latency, and write latencies are usually higher than single-digit milliseconds for non-trivial workloads. Also, read replicas do not provide multi-writer capability and do not ensure sub-minute write availability in all failure scenarios without additional architecture and failover handling.

Option D has the highest operational overhead because it requires the company to operate Redis on EC2: instance management, patching, scaling, replication, failover automation, monitoring, and backup/restore processes. This does not meet the requirement for the least operational overhead compared to fully managed services.

Therefore, an Amazon MemoryDB cluster in Multi-AZ mode provides the required ultra-low latency, rapid failover for write availability, durability for persistence, and least operational overhead.