C++ map数据清理clear效率问题

1. 现实问题

生产中接收历史股票行情做一些别的计算时。整个计算系统的大概流程如下：

因为json解析比较耗时，所以采用了分割vector后多线程解析json成Json::Value对象数据，并通过日期时间作为key插入保存在同一个map中，这样解析后的数据也是按key 日期时间有序。从而保存后续的计算没有问题。

遇到的问题是，当map的value保存的是对象且map很大时，clear释放的时候比较耗时。63W条数据，要26秒，太慢了。影响了整个系统的效率。

2. 测试环境

2.1 计算机情况

2.2 IDE运行情况

使用VS2019社区版，编译x64.Release版本直接在vs下运行。项目属性，优化选项采用默认的，最大优化。

3. 测试代码

模拟实际生产情况的测试代码，map中的value保存的是对象。

#include<map>
#include<vector>
#include<string>
#include <iostream>
#include <time.h>
using namespace std;

class useTest
{
public:
    useTest() {};
    map<string,string> testMap;
    vector<string> testVertor;
    string id;
};

void clearData(map<int, useTest>& needClearMap)
{
    clock_t  startt = clock();

    //分别通过去注释测试下面四种情况

    //使用clear
    needClearMap.clear();

    //使用swap
    //map<int, useTest> uu;
    //needClearMap.swap(uu);

    //使用erase
    //needClearMap.erase(needClearMap.begin(), needClearMap.end());

    //使用for erase
    //for (auto iter = needClearMap.begin(); iter != needClearMap.end(); iter = needClearMap.erase(iter)) {}
    double sec = double(clock() - startt) / CLOCKS_PER_SEC;
    std::cout << "In Clear Cost Time:" << sec << endl;
}

void test()
{
    map<int, useTest> needClearMap;
    for (size_t i = 0; i <= 10000; ++i)
    {
        useTest uT;
        for (size_t ii = 0; ii <= 1000; ++ii)
        {
            uT.testMap[to_string(ii)] = "我是测试，我是测试，我就是测试string";
            uT.testVertor.push_back("我也是测试，我也是测试，我就是测试string");
        }
        uT.id = to_string(i);
        //cout << i << endl;
        needClearMap[i] = uT;
    }
    clock_t  startt = clock();
    clearData(needClearMap);
    double sec = double(clock() - startt) / CLOCKS_PER_SEC;
    std::cout << "clearData Cost Time:" << sec << endl;
}

int main()
{
    for (size_t i = 0; i < 10; ++i)
    {
        test();
    }
    getchar();
}

4. 测试结果

for循环10次测试情况

4.1 使用clear

In Clear Cost Time:2.141
clearData Cost Time:2.142
In Clear Cost Time:2.522
clearData Cost Time:2.523
In Clear Cost Time:2.633
clearData Cost Time:2.633
In Clear Cost Time:2.834
clearData Cost Time:2.836
In Clear Cost Time:2.748
clearData Cost Time:2.749
In Clear Cost Time:2.919
clearData Cost Time:2.919
In Clear Cost Time:4.114
clearData Cost Time:4.116
In Clear Cost Time:3.734
clearData Cost Time:3.735
In Clear Cost Time:4.147
clearData Cost Time:4.147
In Clear Cost Time:3.949
clearData Cost Time:3.949

4.2 使用swap

In Clear Cost Time:0
clearData Cost Time:1.962
In Clear Cost Time:0
clearData Cost Time:2.102
In Clear Cost Time:0
clearData Cost Time:2.22
In Clear Cost Time:0
clearData Cost Time:2.258
In Clear Cost Time:0
clearData Cost Time:2.457
In Clear Cost Time:0
clearData Cost Time:2.908
In Clear Cost Time:0
clearData Cost Time:3.358
In Clear Cost Time:0
clearData Cost Time:3.352
In Clear Cost Time:0
clearData Cost Time:3.236
In Clear Cost Time:0
clearData Cost Time:3.048

4.3 使用erase

In Clear Cost Time:2.005
clearData Cost Time:2.007
In Clear Cost Time:2.317
clearData Cost Time:2.363
In Clear Cost Time:3.113
clearData Cost Time:3.113
In Clear Cost Time:2.45
clearData Cost Time:2.45
In Clear Cost Time:2.52
clearData Cost Time:2.521
In Clear Cost Time:2.937
clearData Cost Time:2.938
In Clear Cost Time:3.825
clearData Cost Time:3.825
In Clear Cost Time:3.646
clearData Cost Time:3.646
In Clear Cost Time:3.195
clearData Cost Time:3.196
In Clear Cost Time:3.554
clearData Cost Time:3.554

4.4 使用for erase

In Clear Cost Time:1.944
clearData Cost Time:1.944
In Clear Cost Time:2.091
clearData Cost Time:2.091
In Clear Cost Time:1.923
clearData Cost Time:1.924
In Clear Cost Time:1.985
clearData Cost Time:1.986
In Clear Cost Time:1.917
clearData Cost Time:1.918
In Clear Cost Time:2.248
clearData Cost Time:2.249
In Clear Cost Time:2.259
clearData Cost Time:2.259
In Clear Cost Time:2.649
clearData Cost Time:2.65
In Clear Cost Time:2.566
clearData Cost Time:2.566
In Clear Cost Time:2.56
clearData Cost Time:2.561

5. 测试结论

就单单实现某个map清空来说，swap效率最高，几乎是0耗时。但是当退出整个函数，释放swap转移到的临时对象要耗一定的时间。
erase效率稍微比clear高。通过for循环erase好似效率又高点。

6. 最后问题解决

基于测试的情况，采用swap，并把clearData函数放到另外一个线程中，这样它释放局部变量的耗时并不影响当前线程，加一个状态变量判定要释放的map已经被释放了就行。

修改后的系统流程如下：