Cache topology, consistency models, invalidation strategies, and distributed state management for low-latency financial applications. The patterns that keep p99 latency in the millisecond range and survive cache stampedes.
Cache patterns (cache-aside, write-through, write-behind), Redis topology (Cluster vs Sentinel), invalidation strategies, the cache stampede that has felled more banking platforms than any disk failure, and operational patterns for sub-millisecond latency.
Embedded grid topology, partition awareness, near-cache configuration, and the stateful financial workloads where collapsing the network hop between your JVM and its data beats a shared Redis cluster — plus the real costs of embedding state inside application pods.
Driving cache invalidation from a Debezium DB2 stream: connector config, Redis consumer with Lua tag-invalidation, SAMA audit continuity, and the only pattern that survives DBA-level schema changes the application doesn’t know about.
Spring Session + Redis for stateless banking apps, OAuth2/FAPI 2.0 PKCE session lifecycle, SAMA timeout requirements, concurrent session control, and the session fixation and CSRF patterns that survive a regulator examination.
Caffeine L1 and Redis L2 CompositeCacheManager, keyspace notification for L1 coherence across pods, cache warming on startup, TTL budget allocation, and the stampede prevention patterns that keep p99 latency under 5 ms at banking scale.