Why is the maxmemory-policy setting important for a cache-only Redis deployment?

maxmemory-policy is what actually determines Redis's behavior once the configured maxmemory limit is hit, and for a cache-only deployment, getting this setting wrong can turn a memory-pressure event into an application outage rather than a graceful, expected eviction.

maxmemory 4gb
maxmemory-policy allkeys-lfu

Left at the default noeviction, a cache-only instance under memory pressure starts rejecting every new write with an error, which most application code isn't written to handle gracefully — a caching layer failing writes often isn't treated as a soft, expected condition the way a cache miss is. Setting an appropriate allkeys-* policy instead means the instance quietly evicts less valuable entries to make room, keeping the cache functional and the application unaware anything unusual even happened. The choice between allkeys-lru (recency-based) and allkeys-lfu (frequency-based) matters too: LFU tends to perform better for workloads with a strong "hot set" of frequently-accessed keys, since LRU alone can be tricked by a burst of one-time accesses pushing genuinely hot keys out.