Redis Data Types

Strings, hashes, sets, sorted sets.

Overview

Redis ships several data types beyond strings, and picking the right one for each workload turns Redis from a key-value cache into a precision tool. Hashes serialise objects without copying full strings; sets answer membership questions in O(1); sorted sets handle leaderboards natively. Defaulting to strings for everything wastes memory and forces application-level work that Redis could do faster in C.

• Strings. Simple values, counters, atomic increments. The default that should not always be the default.
• Hashes. Object fields under one key. Update individual fields without rewriting the whole serialised object.
• Sets. Membership testing in O(1). Tag systems, deduplication, “is this user in the cohort” checks.
• Sorted sets plus other types. Leaderboards and time-ranged queries via score; lists, streams, bitmaps, and HyperLogLog cover further specialised cases.

The approach

Three habits make Redis usage idiomatic: pick types per workload rather than per habit, use hashes for objects to save memory, and document the rationale so future engineers do not regress to strings.

• Workload-driven type choice. Each access pattern picks the type that fits, not the type that is familiar.
• Hashes for objects. HSET and HGET per field. Cuts memory and lets fields update independently.
• Sorted sets for rankings. ZADD with score plus ZRANGE for top-N queries. Leaderboards run at memory speed.
• Documented rationale plus per-key type. Per-key the type and the why; new engineers inherit the reasoning instead of guessing.

Why this compounds

Each correct type choice compounds across every access. Memory footprint shrinks; latency drops; the team’s caching mental model deepens. New services pick the right type from day one rather than defaulting to strings and migrating later.

• Performance improves. Right type matches workload. Single-millisecond operations stay sub-millisecond.
• Memory efficiency. Hashes and bitmaps use far less memory than serialised strings. Cluster size shrinks accordingly.
• Engineering culture matures. Workload-driven decisions replace tribal defaults. Code reviews start citing Redis types deliberately.
• Year-one investment, year-two habit. First deliberate type choice takes thought. By the third service, type selection is automatic.