redis

Use this skill for Redis 相关设计、性能与运维（DBA/中间件）任务。

Defaults / assumptions to confirm

• Redis mode: single instance / Sentinel / Cluster
• Version and deployment (bare metal, Docker, managed)
• Persistence: RDB / AOF / both
• Memory policy and eviction strategy

Workflow

1. Understand use-cases

• Cache vs primary store vs queue/stream.
• Data size, TTL distribution, QPS, latency SLO.
• Consistency requirements and acceptable staleness.

2. Key design

• Namespacing: {app}:{domain}:{entity}:{id} (or similar)
• Avoid overly long keys; ensure stable prefixes for metrics.
• Plan for multi-tenant isolation if needed.

3. Data structures & patterns

• Strings/Hashes for objects, Sets/ZSets for membership/ranking, Streams for event pipelines.
• Avoid large values; prefer hashes for many small fields.
• Choose one cache pattern explicitly: Cache-Aside / Write-Through / Write-Behind.
• Prevent cache stampede: singleflight/mutex, request coalescing, jittered TTL.

4. Performance & reliability

• Identify hot keys, big keys, slow commands.
• Use pipelining where safe; avoid blocking commands on large collections.
• Track latency with LATENCY DOCTOR / slowlog; instrument at client.

5. Memory management

• Set maxmemory and an eviction policy suitable for workload (allkeys-lru, volatile-ttl, etc.).
• Watch fragmentation and RSS vs used_memory.
• Use key TTLs and size controls to avoid unbounded growth.

6. Persistence & durability

• RDB: snapshot intervals, fork time, disk IO impact.
• AOF: fsync policy, rewrite, size growth.
• Define recovery objectives (RPO/RTO) explicitly.

7. HA / scaling

• Sentinel: failover behavior, client reconnection strategy.
• Cluster: hash slots, resharding plan, multi-key operations constraints.
• Plan for multi-AZ and network partitions.

8. Operations checklist

• Backups and restore drills (test in staging).
• Capacity planning: memory headroom, CPU, network bandwidth.
• Upgrade playbook and rollback plan.

Outputs

• Key/TTL design doc (prefixes, structures, TTL, max size).
• Config recommendations (maxmemory, persistence, replication).
• Troubleshooting report (symptoms → evidence → root cause → fixes).