Redis热点Key问题的完整解决方案:

# 实时监控热点Key（Redis 4.0+）  
redis-cli --hotkeys  

# 输出示例  
[45.45%] Hot key 'product:1001:info' found: 120000 hits  

# Python伪代码:在客户端统计  
hot_keys = {}  

def before_execute(command):  
    key = command[1]  
    hot_keys[key] = hot_keys.get(key, 0) + 1  
    if hot_keys[key] > 5000:  
        alert_hot_key(key)  

LoadingCache<String, String> cache = Caffeine.newBuilder()  
    .maximumSize(10_000)  
    .expireAfterWrite(1, TimeUnit.SECONDS) // 短时缓存防穿透  
    .build(key -> redis.get(key));  

// 使用  
String value = cache.get("product:1001:info");  

import hashlib  

def decr_stock(item_id, count):  
    shard_id = hashlib.md5(user_id.encode()).hexdigest()[-1]  # 按用户分片  
    key = f"stock:{item_id}:shard{shard_id}"  
    return redis.decr(key, count)  

location /product {  
    set $bucket $arg_user_id % 10;  # 按用户ID分桶  
    rewrite ^/product/(.*) /$1:bucket$bucket last;  
}  

# 创建集群（3主3从）  
redis-cli --cluster create \  
  192.168.1.101:6379 192.168.1.102:6379 192.168.1.103:6379 \  
  192.168.1.104:6379 192.168.1.105:6379 192.168.1.106:6379 \  
  --cluster-replicas 1  

proxy_cache_path /data/nginx/cache levels=1:2 keys_zone=product_cache:10m;  

location /product {  
    proxy_cache product_cache;  
    proxy_cache_valid 200 5s;  # 缓存5秒  
    proxy_pass http://redis_backend;  
}  

from sklearn.ensemble import RandomForestRegressor  

# 历史访问数据训练模型  
model.fit(X_train, y_train)  

# 预测下一分钟热点  
hot_keys = model.predict(next_minute_features)  
preheat_cache(hot_keys)  # 主动预热  

// 双重检查锁  
public String getData(String key) {  
    String data = localCache.get(key);  
    if (data == null) {  
        synchronized (this) {  
            data = localCache.get(key);  
            if (data == null) {  
                data = redis.get(key);  
                localCache.put(key, data, 1+random(5)); // 随机过期防同时失效  
            }  
        }  
    }  
    return data;  
}  

# 一致性哈希分片  
shard_id = crc32(user_id) % 1024  # 扩大哈希空间  

# 熔断降级规则（Sentinel）  
rules:  
- resource: product:1001:info  
  threshold: 10000  # QPS>10000触发降级  
  fallback: "返回默认商品信息"