When “One in a Billion” Happens Every Day: Scaling Redis at Report URI

Report URI processes telemetry at such scale that one-in-a-billion edge cases occur daily. The post details several Redis optimizations made to handle this load: increasing the replication backlog from 64MB to 2GB to prevent full resyncs during failovers, switching from phpredis connect() to pconnect() to reuse persistent connections and cut connection overhead by ~99.97%, replacing blocking hGetAll() calls with incremental hScan() to avoid stalling Redis's single-threaded command loop, and upgrading primary/replica servers to 16GB RAM with 4 vCPUs. The HA Redis Sentinel setup enabled zero-downtime upgrades via controlled failovers. Future considerations include read replicas, though write-heavy volatile data and atomic lock requirements make that complex.