Salesforce's Marketing Cloud Caching team, led by Paladi Sandhya Madhuri and others, successfully migrated its caching infrastructure from Memcached to Redis Cluster, achieving a remarkable 1.5 million cache events per second without any downtime. The existing Memcached system posed challenges, such as a lack of native replication and user authentication, which increased latency and stressed databases during failures. The move to Redis Cluster not only improved performance with its primary-replica replication and built-in security features but also aligned with broader modernization efforts within Salesforce's infrastructure. The team's zero-downtime strategy required meticulous planning and execution under live production traffic. They implemented a Dynamic Cache Router to facilitate traffic shifts between Memcached and Redis, used double-writes during warm-up to ensure key availability, and applied service grouping to prevent inconsistencies. These strategies helped maintain stable cache hit rates and predictable latency throughout the migration. To address differences in Time-to-Live (TTL) semantics and key-handling behaviors between the two systems, the team developed a compatibility layer that ensured Redis functioned seamlessly with existing applications. Handling hot keys was another challenge due to Redis Cluster's single-threaded shard model, which can cause bottlenecks when certain keys receive excessive traffic. The team had previously built detection tools in Memcached and extended this capability to Redis. They implemented a probabilistic model to track key access frequencies and refactored application code to use hybrid caching patterns. This approach, which included local in-memory caches and read replicas, alleviated pressure on Redis shards during peak traffic, ensuring stable performance across the migration.