Hash commands

Hashes map binary-safe field names to binary-safe values. Small hashes use one packed listpack; larger efficient hashes promote to arena storage plus a Swiss index, while the real-time implementation uses incrementally split linear-hash buckets. See the implementation guide for those layout and latency tradeoffs.

Missing keys behave as empty hashes. A key holding another type returns WRONGTYPE.

HMSET

HMSET key field value [field value ...]

HMSET is the legacy alias for multi-field HSET. It performs the same atomic write but returns OK, preserving the reply expected by older clients.

HINCRBYFLOAT

HINCRBYFLOAT key field increment

Treat a missing field as zero, add a finite floating-point increment, store the shortest round-trippable decimal result, and return that text as a bulk string. Invalid existing text, a non-finite increment, or a non-finite result is an error and leaves the field unchanged.

HRANDFIELD

HRANDFIELD key [count [WITHVALUES]]

Without count, return one field or nil for an empty hash. A positive count returns up to that many distinct fields. A negative count returns exactly its absolute value and may repeat fields. WITHVALUES returns a flat sequence of field/value pairs and is valid only when count is present. Sampling addresses the hash's dense field table directly rather than first materializing every value.

Implementations and scans

The hash family is also addressable through GOBLIN.RT.* and the historical GOBLIN.EFFICENT.* spelling. A qualified mutating command selects the layout when it creates a hash; subsequent commands operate on that stored layout. Standard names use --hash-implementation, or the real-time layout when --real-time is enabled.

For HSCAN key cursor [MATCH pattern] [COUNT count] [NOVALUES], including its cursor and mutation contract, see the iteration reference.