Valkey

Valkey can serve as a cache, queue-like store, or durable in-memory database, but those roles have different correctness requirements. Before comparing replacements, state what must survive a restart.

Replay a small workload using your real key types, expiry values, and client commands. Test memory pressure and eviction, restart the primary, and compare the recovered keys and TTLs. If replication is used, interrupt the link and observe failover and write loss under your chosen configuration. Record p95 latency and memory use while the dataset is warming back up.

Valkey documents RDB and AOF persistence choices and asynchronous replication limits. A cache may tolerate lost keys; session or job state may not. Decide that boundary before optimizing a benchmark.