Fast Warm Reads
L1 memory caching with optional Window TinyLFU admission. Benchmark actual throughput and memory use on your workload.
L1 RAM → L1.5 disk or /dev/shm spill → L2 Redis/Valkey. Bounded local tiers, per-process singleflight, and configurable distributed invalidation.

[Incoming Read Request]
│
▼
┌──────────────────────────────────────────────────┐
│ L1: Smart Memory (W-TinyLFU + CMS) │ Bounded packed data
│ Window LRU (1%) + Segmented Main SLRU (99%) │ Workload-dependent latency
└────────────────────────┬─────────────────────────┘
│ Miss / Eviction Spill
▼
┌──────────────────────────────────────────────────┐
│ L1.5: Off-Heap /dev/shm & NVMe Spill Tier │ Bounded write concurrency
│ tmpfs counts against container memory usage │ Writable mount required
└────────────────────────┬─────────────────────────┘
│ Miss
▼
┌──────────────────────────────────────────────────┐
│ L2: Redis / Valkey Distributed Tier │ ~1.2 ms (Network)
│ AES-256-GCM AEAD at-rest + msgpackr records │ Pipelined MGET/MSET
└────────────────────────┬─────────────────────────┘
│ Complete Cache Miss
▼
┌──────────────────────────────────────────────────┐
│ Upstream Database / API (Singleflight) │ Per process/key
│ Inflight Promise Map coalesces concurrent work│ Not a fleet-wide lock
└──────────────────────────────────────────────────┘First-class drivers and turnkey adapters built for zero serialization overhead across modern runtimes.