Architecting Redis In-Memory Eviction Strategies (5691)

Technical Overview

Engineering breakdown of Architecting Redis In-Memory Eviction Strategies (5691). Bare-metal hardware performance requires isolated kernel parameters, tuned memory allocators, and precise eviction policy selection to meet sub-millisecond latency SLAs.

Eviction Policy Taxonomy

Redis offers eight eviction policies: noeviction, allkeys-lru, volatile-lru, allkeys-random, volatile-random, allkeys-ttl, volatile-ttl, and allkeys-lfu/volatile-lfu. Each policy trades off hit‑rate, memory overhead, and CPU cycles. The choice hinges on workload access patterns: LRU for temporal locality, LFU for frequency‑based retention, TTL for explicit expiration.

Memory Fragmentation & Allocator Tuning

Jemalloc (default) and allocators like tcmalloc or mimalloc exhibit different fragmentation profiles under high churn. Enabling activedefrag yes with active-defrag-threshold-lower 10 and active-defrag-threshold-upper 100 reduces RSS bloat by up to 30% in write‑heavy workloads.

Production Hardening Checklist

• Set maxmemory to 80% of available RAM.
• Enable lazyfree-lazy-eviction yes to offload key deletion to background threads.
• Monitor evicted_keys and expired_keys via Prometheus exporter.
• Use Redis 7.2+ LFU decay time (lfu-decay-time) to adapt to shifting hot sets.