Recommendation
Profile cache: use a Hash per user. Leaderboard: use a Sorted Set per board.
User profile: Hash
The JSON string forces fetch, parse, mutate and rewrite for every change. A Hash lets you touch one field at a time:
HSET user:1001:profile name "Ada" email "[email protected]" level 42 country "DE"
HSET user:1001:profile email "[email protected]" # update one field, no read needed
HGET user:1001:profile email # read one field
HGETALL user:1001:profile # read everything
HINCRBY user:1001:profile level 1 # atomic numeric update
EXPIRE user:1001:profile 3600 # TTL, since this is a cache
This is atomic, needs no read-modify-write cycle, and avoids lost updates when two writers race. It also sends far less data over the wire.
Hashes hold flat field/value pairs. If the profile has nested data, such as arrays of items or sub-objects, use the JSON type (JSON.SET user:1001:profile $.address.city '"Berlin"'). Otherwise a Hash is simpler and lighter. You can also keep rarely changed nested parts in a separate key, for example user:1001:settings.
Leaderboard: Sorted Set
ZADD game:space-invaders:leaderboard 12500 user:1001 # set/update score
ZINCRBY game:space-invaders:leaderboard 300 user:1001 # add points atomically
ZREVRANGE game:space-invaders:leaderboard 0 9 WITHSCORES # top 10
ZREVRANK game:space-invaders:leaderboard user:1001 # a player's rank
The member should be the user ID, not the whole profile. To render the board, take the top N IDs and fetch each profile with HMGET or HGETALL, pipelined.
Key naming
Use lowercase, colon-separated keys in the form {entity}:{id}:{attribute}. User_1001_Profile breaks that convention in casing, separators and order.
| Purpose | Key |
|---|---|
| Profile | user:1001:profile |
| Settings | user:1001:settings |
| Global leaderboard | game:space-invaders:leaderboard |
| Periodic board | game:space-invaders:leaderboard:2026-w43 |
| Session | session:abc123 |
Further tips:
- Keep keys short, because they live in memory and appear in every command.
- Pick one convention and apply it everywhere.
- If you're multi-tenant, prefix with the tenant (
tenant:42:user:1001:profile). That makes scans and ACLs easy to scope. - Time-bucketed leaderboard keys such as weekly or daily boards can be expired or rotated with
EXPIRE.
Migration
- Write a one-off script that reads each
User_*_Profilestring, parses it and doesHSET user:{id}:profile .... - Have the app write to the new keys, or dual-write briefly.
- Delete the old keys once reads are fully switched over. Use
SCANfor this, neverKEYS.
I can sketch the migration script or the client code in Python or Java if that would help.
Pure-instruction skill, so the trial gave the model only the Skill and Read tools: no network, no shell, no Redis server. The request mixes a data-model question with a key-renaming question.
What it does
Helps the model model data in Redis the way Redis, Inc. recommends: match the data type to the access pattern, and use short, lowercase, colon-separated key names such as user:1001:profile.
How it works
- A decision table maps use cases (counters, objects, queues, unique sets, rankings, nested data, event logs, vectors) to a Redis type.
- It flags the common anti-pattern of storing a flat object as one serialized string, and shows the Hash alternative in Python (redis-py) and Java (Jedis).
- It gives key-naming rules:
{entity}:{id}:{attribute}, no spaces or mixed case, no URLs as keys, a tenant prefix for multi-tenant apps.
Good for
Designing a Redis data model, caching objects or sessions, building counters and leaderboards, and cleaning up inconsistent key names.
Limits
A short skill (about 7 KB with two reference files). It covers data types and key names only; it does not cover clustering, persistence or search.
Low risk: pure instruction files from the official Redis repository (MIT). No scripts, no network access, no file writes, no account or key needed. It only advises; it does not connect to your Redis. Check any migration advice against your own data before running it.