c++ mongodb driver 性能分析

系列数据库开发

---

---

前言

---

一、mongodb 性能测试环境准备

测试基准：

Centos7.5
cpu：4
memory：6g

安装Mongodb数据库，版本为4.4.3
g++ 4.8.5

mongo启动使用默认配置，无任何修改的环境下。

二、c++ 在centos 7.5 mongo create与query性能分析

1. 代码实现

本测试中通过glags修改insert线程数threaNum，与内存分配大小size

std::atomic_bool running{false};
std::atomic_int timestart(0);
std::atomic_int count(0);
char *buffer = nullptr;
const int EverySeconds = 5;
DEFINE_int32(threaNum, 10, "thread num");
DEFINE_int32(length, 2*1024*1024, "memory size");

void wirteDBTask(int taskid) 
{
    std::cout<< "create Task------>"<< taskid <<std::endl;
    std::thread([taskid](){
        auto client = dbclient();
        auto coll = (*client)["vtest"]["test"];  
        while (true) { 
            auto userid = butil::GenerateGUID();
            const auto &doc = bsoncxx::builder::stream::document{}
                                << "_id" << userid 
                                << "user" << butil::GenerateGUID()
                                << "pwd" << butil::GenerateGUID()
                                << "name" << "text"
                                << "jpegdata" << buffer
                                << finalize;      
            coll.insert_one(doc.view()); 

            ++count;
            auto now = time(NULL);
            if(now - timestart > EverySeconds){
                LOG(INFO)<< "every second average: "<< count/EverySeconds;
                count = 0;
                timestart = now;
            }            
        }
        std::cout<< "destroy Task------>"<< taskid <<std::endl;
    }).detach();    
}
void readDBTask(int taskid)
{
    std::cout<< "create Task------>"<< taskid <<std::endl;
    std::thread([taskid](){
        auto client = dbclient();
        auto coll = (*client)["vtest"]["test"];  
        std::string user[5] = {
            "7D1D01FD-41B7-FEC9-E0F9-FF64BF63E35F",
            "DDDDDDDDDDDDDDDDDDDDDDDDDEEEEEEEEEEEE",
            "7D1D01FD-41B7-FEC9-E0F9-FF64BF63E35F",
            "86680647-6966-24D8-00B3-FF38CAA608A7",
            "DDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDD"
        };
        int circle = 0;
        while (true) { 
            ++circle;
            if(circle%5==0){
                circle = 0;
            }            
            const auto &doc = bsoncxx::builder::stream::document{}
                                << "user" << user[circle]
                                << finalize;        
            auto resultdoc = coll.find_one(doc.view());           
            if(resultdoc){
                resultdoc->view()["name"].get_utf8().value.to_string();
            }     
            ++count;
            auto now = time(NULL);
            if(now - timestart > EverySeconds){
                LOG(INFO)<< "every second average: "<< count/EverySeconds;
                count = 0;
                timestart = now;
            }            
        }
        std::cout<< "destroy Task------>"<< taskid <<std::endl;
    }).detach();   
    return ;
}

bool InsertDataByMutilThreads()
{
    auto result = true;
    running = true;
    
    buffer = new char [FLAGS_length];
    memset((void*)buffer,1,FLAGS_length);

    timestart = time(NULL);
    for(auto taskid=0;taskid<FLAGS_threaNum;taskid++){
        wirteDBTask(taskid);
    }
    return res;
}

bool QueryDataByMutilThreads()
{
    auto result = true;
    running = true;
    
    buffer = new char [FLAGS_length];
    memset((void*)buffer,1,FLAGS_length);

    timestart = time(NULL);
    for(auto taskid=0;taskid<FLAGS_threaNum;taskid++){
        readDBTask(taskid);
    }
    return res;
}

2. 写性能测试

实验1：固定buffersize = 102410242=2M,通过增加多线程来insert，当线程达到一定数量后，tps不会持续增加。
a）
thread Num: 1
buffer alloc length: 2097152
tps：60-70

b）
buffer alloc length: 2097152
thread Num: 4
tps：160-170

c）
buffer alloc length: 2097152
thread Num: 8
tps：160-170

实验2：固定线程数4，通过insert不同大小的buffer，当buffer减少到一定大小，就会保持tps稳定，也就是说tps随着buffersize的增加会降低。

a）
thread Num: 4
buffer alloc length: 2097152 = 2M
tps：160-170

b）
thread Num: 4
buffer alloc length: 1048576 = 1M
tps：320-350

c）
thread Num: 4
buffer alloc length: 524288= 512K
tps： 600-700

d）
thread Num: 4
buffer alloc length: 262144= 256K
tps： 1100-1200

e）
thread Num: 4
buffer alloc length: 1024= 1K
tps： 5000-6000

f）
thread Num: 4
buffer alloc length: 1Byte
tps： 5000-6000

3. 以_id为查询条件的读性能测试

默认以_id为索引

std::atomic_bool running{false};
std::atomic_int timestart(0);
std::atomic_int count(0);
char *buffer = nullptr;
const int EverySeconds = 5;
DEFINE_int32(threaNum, 10, "thread num");
DEFINE_int32(length, 2*1024*1024, "memory size");


void readDBTask(int taskid)
{
    std::cout<< "create Task------>"<< taskid <<std::endl;
    std::thread([taskid](){
        auto client = dbclient();
        auto coll = (*client)["vtest"]["test"];  
        std::string id[5] = {
            "7D1D01FD-41B7-FEC9-E0F9-FF64BF63E35F",
            "DDDDDDDDDDDDDDDDDDDDDDDDDEEEEEEEEEEEE",
            "7D1D01FD-41B7-FEC9-E0F9-FF64BF63E35F",
            "86680647-6966-24D8-00B3-FF38CAA608A7",
            "DDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDD"
        };
        int circle = 0;
        while (true) { 
            ++circle;
            if(circle%5==0){
                circle = 0;
            }            
            const auto &doc = bsoncxx::builder::stream::document{}
                                << "_id" << id[circle]
                                << finalize;      
            auto resultdoc = coll.find_one(doc.view());           
            if(resultdoc){
                resultdoc->view()["name"].get_utf8().value.to_string();
            }     
            ++count;
            auto now = time(NULL);
            if(now - timestart > EverySeconds){
                LOG(INFO)<< "every second average: "<< count/EverySeconds;
                count = 0;
                timestart = now;
            }            
        }
        std::cout<< "destroy Task------>"<< taskid <<std::endl;
    }).detach();   
    return ;
}



bool QueryDataByMutilThreads()
{
    auto result = true;
    running = true;
    
    buffer = new char [FLAGS_length];
    memset((void*)buffer,1,FLAGS_length);

    timestart = time(NULL);
    for(auto taskid=0;taskid<FLAGS_threaNum;taskid++){
        readDBTask(taskid);
    }
    return res;
}

实验1：通过增加多线程来query，当线程达到一定数量后，tps不会持续增加。
本实验默认以_id为查询条件。
a）
thread Num: 1
tps：800-900

b）
thread Num: 4
tps：950-1050

c）
thread Num: 4
tps：950-1050

4. 无索引与有索引读性能测试

std::atomic_bool running{false};
std::atomic_int timestart(0);
std::atomic_int count(0);
char *buffer = nullptr;
const int EverySeconds = 5;
DEFINE_int32(threaNum, 10, "thread num");
DEFINE_int32(length, 2*1024*1024, "memory size");


void readDBTask(int taskid)
{
    std::cout<< "create Task------>"<< taskid <<std::endl;
    std::thread([taskid](){
        auto client = dbclient();
        auto coll = (*client)["vtest"]["test"];  
        std::string user[5] = {
            "7D1D01FD-41B7-FEC9-E0F9-FF64BF63E35F",
            "DDDDDDDDDDDDDDDDDDDDDDDDDEEEEEEEEEEEE",
            "7D1D01FD-41B7-FEC9-E0F9-FF64BF63E35F",
            "86680647-6966-24D8-00B3-FF38CAA608A7",
            "DDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDD"
        };
        int circle = 0;
        while (true) { 
            ++circle;
            if(circle%5==0){
                circle = 0;
            }            
            const auto &doc = bsoncxx::builder::stream::document{}
                                << "user" << user[circle]
                                << finalize;      
            auto resultdoc = coll.find_one(doc.view());           
            if(resultdoc){
                resultdoc->view()["name"].get_utf8().value.to_string();
            }     
            ++count;
            auto now = time(NULL);
            if(now - timestart > EverySeconds){
                LOG(INFO)<< "every second average: "<< count/EverySeconds;
                count = 0;
                timestart = now;
            }            
        }
        std::cout<< "destroy Task------>"<< taskid <<std::endl;
    }).detach();   
    return ;
}



bool QueryDataByMutilThreads()
{
    auto result = true;
    running = true;
    
    buffer = new char [FLAGS_length];
    memset((void*)buffer,1,FLAGS_length);

    timestart = time(NULL);
    for(auto taskid=0;taskid<FLAGS_threaNum;taskid++){
        readDBTask(taskid);
    }
    return res;
}

实验1：
固定线程数，通过有索引查询与无索引query的tps，有索引的tps远远大于索引查询tps
本实验默认以_id为查询条件。
a）
thread Num: 4
tps：0-1

b）
db.test.createIndex({user:1});
thread Num: 4
tps：950-1050

---

总结

以上就是今天要讲的内容，希望你对mongodb性能有个大概的了解。