Django项目Redis缓存性能异常：带缓存接口为何更慢？

测试数据对比 #

无缓存接口（person）测试结果 #

[Thu Mar 23 2023 17:48:01 GMT+0530 (India Standard Time)] INFO Max requests:        50
[Thu Mar 23 2023 17:48:01 GMT+0530 (India Standard Time)] INFO Concurrency level:   1
[Thu Mar 23 2023 17:48:01 GMT+0530 (India Standard Time)] INFO Agent:               none
[Thu Mar 23 2023 17:48:01 GMT+0530 (India Standard Time)] INFO 
[Thu Mar 23 2023 17:48:01 GMT+0530 (India Standard Time)] INFO Completed requests:  50
[Thu Mar 23 2023 17:48:01 GMT+0530 (India Standard Time)] INFO Total errors:        0
[Thu Mar 23 2023 17:48:01 GMT+0530 (India Standard Time)] INFO Total time:          0.40408834 s
[Thu Mar 23 2023 17:48:01 GMT+0530 (India Standard Time)] INFO Requests per second: 124
[Thu Mar 23 2023 17:48:01 GMT+0530 (India Standard Time)] INFO Mean latency:        8 ms

带缓存接口（person1）测试结果 #

[Thu Mar 23 2023 17:47:56 GMT+0530 (India Standard Time)] INFO Max requests:        50
[Thu Mar 23 2023 17:47:56 GMT+0530 (India Standard Time)] INFO Concurrency level:   1
[Thu Mar 23 2023 17:47:56 GMT+0530 (India Standard Time)] INFO Agent:               none
[Thu Mar 23 2023 17:47:56 GMT+0530 (India Standard Time)] INFO 
[Thu Mar 23 2023 17:47:56 GMT+0530 (India Standard Time)] INFO Completed requests:  50
[Thu Mar 23 2023 17:47:56 GMT+0530 (India Standard Time)] INFO Total errors:        0
[Thu Mar 23 2023 17:47:56 GMT+0530 (India Standard Time)] INFO Total time:          18.811554342 s
[Thu Mar 23 2023 17:47:56 GMT+0530 (India Standard Time)] INFO Requests per second: 3
[Thu Mar 23 2023 17:47:56 GMT+0530 (India Standard Time)] INFO Mean latency:        376.1 ms

接口实现代码 #

def person(request):       
    if 'auth' in request.session and  request.session['auth']:
        res_all = Person.objects.values().all()
        return render(request,'blog.html',context={"all_data":res_all})
    else:
        return render(request,'blog.html',context={})


def person1(request):
    cache_data=cache.get('cache_key_person')
    if cache_data:
        return render(request,'blog.html',context={"all_data":cache_data})
    else:
        res_all = Person.objects.values().all()            
        cache.set('cache_key_person', res_all, CACHE_TTL) 
        return render(request,'blog.html',context={"all_data":res_all})

问题根因分析 #

带缓存接口性能暴跌的核心原因是缓存了未执行的QuerySet对象：

• Django的Person.objects.values().all()返回的是惰性求值的QuerySet，它不是实际查询结果，而是包含数据库连接、查询参数等大量冗余信息的查询构造器。
• 调用cache.set存储该QuerySet时，Django会通过pickle对其序列化，这个过程耗时极长，且生成的序列化数据体积远大于实际查询结果。
• 后续从缓存获取数据时，还要对庞大的序列化对象反序列化，进一步拉长请求响应时间。
• 无缓存接口中，Person.objects.values().all()在传递给模板时才执行查询，直接返回字典列表，无额外序列化/反序列化开销，若数据库本身查询速度快，整体性能自然远超带缓存接口。

修复方案 #

修改带缓存接口，将QuerySet执行后的实际结果存入缓存：

def person1(request):
    cache_data=cache.get('cache_key_person')
    if cache_data:
        return render(request,'blog.html',context={"all_data":cache_data})
    else:
        # 将QuerySet转为列表，执行查询并获取实际结果
        res_all = list(Person.objects.values().all())            
        cache.set('cache_key_person', res_all, CACHE_TTL) 
        return render(request,'blog.html',context={"all_data":res_all})

额外优化检查点：

• 确认Redis连接配置合理性，避免每次请求新建连接（Django默认Redis缓存后端维护连接池，配置错误会导致连接开销过大）。
• 若Person表数据量较大，可考虑对缓存结果分页，避免一次性缓存过多数据。

内容的提问来源于stack exchange，提问作者Amit