deangrant Redis Scaling
Use this skill when Redis must sustain high throughput or large memory.
1. Topology
Start Sentinel + single primary until ~100k ops/s or ~25 GB. Move to Cluster when throughput/RAM limits hit. Plan hash tags for multi-key ops. Prefer latency-aware replica reads when available.
2. Hot keys
Monitor per-shard CPU. Read-heavy → short in-process LRU. Writable hot keys → N-way key replication across slots. At scale → client-side caching / BCAST.
3. Throughput and memory
Pipeline batch reads (slot-grouped in Cluster). Size pools by
conns_per_pod × pods; stagger restarts; fail-fast timeouts.
maxmemory ≈ 50% RAM; allkeys-lru (cache) or volatile-lru (mixed). Target
hit rate >95%; watch evicted_keys; replication lag low.
4. Quick checklist
- Sentinel vs Cluster decision explicit.
- Hash tags planned for multi-key.
- Hot-key mitigations.
- Pipelining + pooled clients + timeouts.
- maxmemory policy set.
See reference.md and examples.md.