行深膳食-核心查询接口 RT 稳定控制在 200ms 以内

一:我的-饮食模式设置-食物组构成-查询

wrk -t2 -c50 -d30s --latency \
    -H "Authorization: Bearer eyJhbGc..." \
"http://192.168.31.109:30289/xs-json/health/food-group/user-composition?eating_pattern_id=3&estimated_calorie_needs_id=8"
Running 30s test @ http://192.168.31.109:30289/xs-json/health/food-group/user-composition?eating_pattern_id=3&estimated_calorie_needs_id=8
  2 threads and 50 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency    44.38ms   16.67ms 138.28ms   68.18%
    Req/Sec   566.50     83.91     1.06k    71.24%
  Latency Distribution
     50%   42.03ms
     75%   52.73ms
     90%   65.56ms
     99%   91.01ms
  33910 requests in 30.08s, 234.66MB read
Requests/sec:   1127.35
Transfer/sec:      7.80MB

我用 wrk 做基准压测,摘掉限流、带上合法 JWT,并发 50 压 30 秒;P99 延迟 91ms,远低于 200ms 目标,QPS 稳定 1k+;这条 SQL 本身执行仅 2ms,大部分耗时消耗在 GORM 数据扫描、结构体序列化、网络和协程调度上。

加了redis缓存继续wrk压测结果如下:

Running 30s test @ http://192.168.31.109:30289/xs-json/health/food-group/user-composition?eating_pattern_id=3&estimated_calorie_needs_id=8
  2 threads and 50 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency    12.04ms   12.08ms 141.93ms   87.19%
    Req/Sec     2.69k   575.70     3.75k    56.78%
  Latency Distribution
     50%    7.65ms
     75%   13.07ms
     90%   30.00ms
     99%   52.19ms
  160493 requests in 30.06s, 1.08GB read
Requests/sec:   5339.70
Transfer/sec:     36.95MB

针对膳食食物组推荐读多写少接口,引入 Redis 实现 Cache Aside 缓存方案,通过分布式锁防止缓存击穿、TTL 随机偏移规避雪崩、空值缓存解决穿透;基准压测对比:优化前 QPS 约 1100、P99 91ms,接入缓存后 QPS 提升至 5300+,P99 RT 降至 52ms,显著降低 MySQL 多表关联查询压力。