Redis Object Encoding Optimization Prompt

You are a senior engineer who reduces Redis memory by keeping values in their compact internal encodings.

I will provide:
- `INFO memory` (used_memory, dataset size), a few representative keys, and their sizes
- The access patterns on the biggest namespaces
- Current `*-max-*` config thresholds

Your job:

1. **Explain the encodings**: each type has a compact form for small values and a full form for large ones:
   - Hash/Set/List/ZSet → **listpack** (small) vs hashtable/skiplist/quicklist (large).
   - Set of integers → **intset** (very compact) until it exceeds `set-max-intset-entries` or gains a non-integer.
   - String → **int** (integers), **embstr** (≤44 bytes), **raw** (longer).
2. **Read the thresholds** and their effect: `hash-max-listpack-entries`/`-value`, `list-max-listpack-size`, `set-max-listpack-entries`/`-value`, `set-max-intset-entries`, `zset-max-listpack-entries`/`-value`. Crossing any threshold converts the whole key to the memory-heavier encoding — and Redis does **not** convert back automatically.
3. **Verify with `OBJECT ENCODING`**: for each hot key, show whether it is on the compact or heavy encoding and quantify the memory difference with `MEMORY USAGE`.
4. **Reshape to stay compact**: the biggest win is often splitting one huge hash into many small hashes (bucketing by a hash of the field) so each stays under the listpack threshold — the classic "hash-field sharding" trick that stores millions of small values densely.
5. **Tune thresholds deliberately**: raising a `*-max-listpack-*` limit keeps larger collections compact but makes operations on them O(N) (listpack is linear); balance memory vs CPU. Note new keys pick up new limits; existing keys keep their encoding until rewritten.
6. **Confirm the win**: compare `used_memory` and `MEMORY USAGE` before/after, and check `latency` didn't regress from larger listpacks.

Mark DESTRUCTIVE or risky: raising listpack limits so high that O(N) ops stall the event loop, rewriting millions of keys on the primary without rate limiting, and assuming a threshold change re-compacts existing keys (it does not).

---

INFO memory + sample keys: [PASTE]
Access patterns: [DESCRIBE]
Current thresholds: [PASTE]