English | 中文
OS (Compiler Version) | Status |
---|---|
Ubuntu 22.04 (clang 14.0.0) | |
Ubuntu 22.04 (gcc 11.2.0) | |
macOS Monterey 12 (AppleClang 14.0.0.14000029) | |
Windows Server 2022 (MSVC 19.33.31630.0) |
cinatra是一个高性能易用的http框架,它是用modern c++(c++20)开发的,它的目标是提供一个快速开发的c++ http框架。它的主要特点如下:
- 统一而简单的接口
- header-only
- 跨平台
- 高效
- 支持面向切面编程
cinatra目前支持了http1.1/1.0, ssl和websocket, 你可以用它轻易地开发一个http服务器,比如常见的数据库访问服务器、文件上传下载服务器、实时消息推送服务器,你也可以基于cinatra开发一个mqtt服务器。 cinatra是世界上性能最好的http服务器之一,性能测试详见性能测试
除此之外,cinatra 还提供了一个基于C++20 协程的http(https) client,包括普通get/post请求、文件上传下载和web socket、redirect、proxy等功能。
cinatra目前被很多公司在使用,在这里可以看到谁在用cinatra.
- C++20 编译器 (gcc 10.2, clang 13, Visual Studio 2022,或者更高的版本)
cinatra是header-only的,引用include头文件目录,并设置如下编译选项:
如果 linux, 设置:
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -pthread -std=c++20")
如果 g++ 编译,再设置:
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fcoroutines")
set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} -fno-tree-slp-vectorize")
cinatra支持通过指令集优化其内部逻辑,其通过宏来控制是否使用指令集。使用之前请确保cpu支持。
使用如下命令即可编译带simd优化的cinatra。注意只能开启一种simd指令集优化,开启多个会导致编译失败。
cmake -DENABLE_SIMD=SSE42 .. # 启用sse4.2指令集
cmake -DENABLE_SIMD=AVX2 .. # 启用avx2指令集
cmake -DENABLE_SIMD=AARCH64 .. # arm环境下,启用neon指令集
#include "include/cinatra.hpp"
using namespace cinatra;
int main() {
int max_thread_num = std::thread::hardware_concurrency();
coro_http_server server(max_thread_num, 8080);
server.set_http_handler<GET, POST>("/", [](coro_http_request& req, coro_http_response& res) {
res.set_status_and_content(status_type::ok, "hello world");
});
server.sync_start();
return 0;
}
5行代码就可以实现一个简单http服务器了,用户不需要关注多少细节,直接写业务逻辑就行了。
#include "cinatra.hpp"
struct person_t {
void foo(coro_http_request &, coro_http_response &res) {
res.set_status_and_content(status_type::ok, "ok");
}
};
async_simple::coro::Lazy<void> basic_usage() {
coro_http_server server(1, 9001);
server.set_http_handler<GET>(
"/get", [](coro_http_request &req, coro_http_response &resp) {
resp.set_status_and_content(status_type::ok, "ok");
});
server.set_http_handler<GET>(
"/coro",
[](coro_http_request &req,
coro_http_response &resp) -> async_simple::coro::Lazy<void> {
resp.set_status_and_content(status_type::ok, "ok");
co_return;
});
server.set_http_handler<GET>(
"/in_thread_pool",
[](coro_http_request &req,
coro_http_response &resp) -> async_simple::coro::Lazy<void> {
// will respose in another thread.
co_await coro_io::post([&] {
// do your heavy work here when finished work, response.
resp.set_status_and_content(status_type::ok, "ok");
});
});
server.set_http_handler<POST, PUT>(
"/post", [](coro_http_request &req, coro_http_response &resp) {
auto req_body = req.get_body();
resp.set_status_and_content(status_type::ok, std::string{req_body});
});
server.set_http_handler<GET>(
"/headers", [](coro_http_request &req, coro_http_response &resp) {
auto name = req.get_header_value("name");
auto age = req.get_header_value("age");
assert(name == "tom");
assert(age == "20");
resp.set_status_and_content(status_type::ok, "ok");
});
server.set_http_handler<GET>(
"/query", [](coro_http_request &req, coro_http_response &resp) {
auto name = req.get_query_value("name");
auto age = req.get_query_value("age");
assert(name == "tom");
assert(age == "20");
resp.set_status_and_content(status_type::ok, "ok");
});
server.set_http_handler<cinatra::GET, cinatra::POST>(
"/users/:userid/subscriptions/:subid",
[](coro_http_request &req, coro_http_response &response) {
assert(req.params_["userid"] == "ultramarines");
assert(req.params_["subid"] == "guilliman");
response.set_status_and_content(status_type::ok, "ok");
});
person_t person{};
server.set_http_handler<GET>("/person", &person_t::foo, person);
server.async_start();
std::this_thread::sleep_for(300ms); // wait for server start
coro_http_client client{};
auto result = co_await client.async_get("http://127.0.0.1:9001/get");
assert(result.status == 200);
assert(result.resp_body == "ok");
for (auto [key, val] : result.resp_headers) {
std::cout << key << ": " << val << "\n";
}
result = co_await client.async_get("/coro");
assert(result.status == 200);
result = co_await client.async_get("/in_thread_pool");
assert(result.status == 200);
result = co_await client.async_post("/post", "post string",
req_content_type::string);
assert(result.status == 200);
assert(result.resp_body == "post string");
client.add_header("name", "tom");
client.add_header("age", "20");
result = co_await client.async_get("/headers");
assert(result.status == 200);
result = co_await client.async_get("/query?name=tom&age=20");
assert(result.status == 200);
result = co_await client.async_get(
"http://127.0.0.1:9001/users/ultramarines/subscriptions/guilliman");
assert(result.status == 200);
// make sure you have installed openssl and enable CINATRA_ENABLE_SSL
#ifdef CINATRA_ENABLE_SSL
coro_http_client client2{};
result = co_await client2.async_get("https://baidu.com");
assert(result.status == 200);
#endif
}
int main() {
async_simple::coro::syncAwait(basic_usage());
}
#include "cinatra.hpp"
using namespace cinatra;
//日志切面
struct log_t
{
bool before(coro_http_request& req, coro_http_response& res) {
std::cout << "before log" << std::endl;
return true;
}
bool after(coro_http_request& req, coro_http_response& res) {
std::cout << "after log" << std::endl;
return true;
}
};
//校验的切面
struct check {
bool before(coro_http_request& req, coro_http_response& res) {
std::cout << "before check" << std::endl;
if (req.get_header_value("name").empty()) {
res.set_status_and_content(status_type::bad_request);
return false;
}
return true;
}
bool after(coro_http_request& req, coro_http_response& res) {
std::cout << "after check" << std::endl;
return true;
}
};
//将信息从中间件传输到处理程序
struct get_data {
bool before(coro_http_request& req, coro_http_response& res) {
req.set_aspect_data("hello world");
return true;
}
}
int main() {
coro_http_server server(std::thread::hardware_concurrency(), 8080);
server.set_http_handler<GET, POST>("/aspect", [](coro_http_request& req, coro_http_response& res) {
res.set_status_and_content(status_type::ok, "hello world");
}, check{}, log_t{});
server.set_http_handler<GET,POST>("/aspect/data", [](coro_http_request& req, coro_http_response& res) {
auto& val = req.get_aspect_data();
res.set_status_and_content(status_type::ok, std::move(val[0]));
}, get_data{});
server.sync_start();
return 0;
}
本例中有两个切面,一个校验http请求的切面,一个是日志切面,这个切面用户可以根据需求任意增加。本例会先检查http请求的合法性,如果不合法就会返回bad request,合法就会进入下一个切面,即日志切面,日志切面会打印出一个before表示进入业务逻辑之前的处理,业务逻辑完成之后会打印after表示业务逻辑结束之后的处理。
本代码演示如何使用RESTful路径参数。下面设置了两个RESTful API。第一个API当访问,比如访问这样的urlhttp://127.0.0.1:8080/numbers/1234/test/5678
时服务器可以获取到1234和5678这两个参数,第一个RESTful API的参数是(\d+)
是一个正则表达式表明只能参数只能为数字。获取第一个参数的代码是req.get_matches()[1]
。因为每一个req不同所以每一个匹配到的参数都放在request
结构体中。
同时还支持任意字符的RESTful API,即示例的第二种RESTful API"/string/{:id}/test/{:name}"
,要获取到对应的参数使用req.get_query_value
函数即可,其参数只能为注册的变量(如果不为依然运行但是有报错),例子中参数名是id和name,要获取id参数调用req.get_query_value("id")
即可。示例代码运行后,当访问http://127.0.0.1:8080/string/params_1/test/api_test
时,浏览器会返回api_test
字符串。
#include "cinatra.hpp"
using namespace cinatra;
int main() {
int max_thread_num = std::thread::hardware_concurrency();
coro_http_server server(max_thread_num, 8080);
server.set_http_handler<GET, POST>(
R"(/numbers/(\d+)/test/(\d+))", [](request &req, response &res) {
std::cout << " matches[1] is : " << req.get_matches()[1]
<< " matches[2] is: " << req.get_matches()[2] << std::endl;
res.set_status_and_content(status_type::ok, "hello world");
});
server.set_http_handler<GET, POST>(
"/string/{:id}/test/{:name}", [](request &req, response &res) {
std::string id = req.get_query_value("id");
std::cout << "id value is: " << id << std::endl;
std::cout << "name value is: " << std::string(req.get_query_value("name")) << std::endl;
res.set_status_and_content(status_type::ok, std::string(req.get_query_value("name")));
});
server.sync_start();
return 0;
}
cinatra 支持反向代理也很简单,3步5行代码就可以了。 先看一个简单的例子:
reverse_proxy proxy_rr(10, 8091);
proxy_rr.add_dest_host("127.0.0.1:9001");
proxy_rr.add_dest_host("127.0.0.1:9002");
proxy_rr.add_dest_host("127.0.0.1:9003");
proxy_rr.start_reverse_proxy<GET, POST>("/rr", true,
coro_io::load_blance_algorithm::RR);
第一步创建一个代理服务器,设置其线程数和端口; 第二步添加需要访问的服务器列表; 第三步启动代理服务,设置loadbalance 策略,这里选择的是round robin 策略。
在浏览器或者client里访问http://127.0.0.1:8091/rr 就会根据RR 策略选择三个服务器中的一个。
如果要选择random 策略就设置为coro_io::load_blance_algorithm::random。
如果要选择weight round robin 策略,就需要设置服务器权重。
reverse_proxy proxy_wrr(10, 8090);
proxy_wrr.add_dest_host("127.0.0.1:9001", 10);
proxy_wrr.add_dest_host("127.0.0.1:9002", 5);
proxy_wrr.add_dest_host("127.0.0.1:9003", 5);
proxy_wrr.start_reverse_proxy<GET, POST>("/wrr", true,
coro_io::load_blance_algorithm::WRR);
在浏览器或者client里访问http://127.0.0.1:8090/wrr ,第一次和第二次会返回9001服务器的结果,第三次返回9002服务器的结果,第四次返回9003服务器的结果,第五次又重新返回9001服务器的结果,这就是WRR的策略。
void test_sync_client() {
{
std::string uri = "http://www.baidu.com";
coro_http_client client{};
auto result = client.get(uri);
assert(!result.net_err);
print(result.resp_body);
result = client.post(uri, "hello", req_content_type::json);
print(result.resp_body);
}
{
coro_http_client client{};
std::string uri = "http://cn.bing.com";
auto result = client.get(uri);
assert(!result.net_err);
print(result.resp_body);
result = client.post(uri, "hello", req_content_type::json);
print(result.resp_body);
}
}
#ifdef CINATRA_ENABLE_SSL
void test_coro_http_client() {
using namespace cinatra;
coro_http_client client{};
client.init_ssl("../../include/cinatra", "server.crt"); // optinal 一般情况下可以不调用这一行
auto data = client.get("https://www.bing.com");
std::cout << data.resp_body << "\n";
data = client.get("https://www.bing.com");
std::cout << data.resp_body << "\n";
}
#endif
async_simple::coro::Lazy<void> test_async_client() {
std::string uri = "http://www.baidu.com";
{
coro_http_client client{};
auto data = co_await client.async_get(uri);
print(data.status);
data = co_await client.async_get(uri);
print(data.status);
data = co_await client.async_post(uri, "hello", req_content_type::string);
print(data.status);
}
#ifdef CINATRA_ENABLE_SSL
std::string uri2 = "https://www.baidu.com";
std::string uri3 = "https://cn.bing.com";
coro_http_client client{};
client.init_ssl("../../include/cinatra", "server.crt");
data = co_await client.async_get(uri2);
print(data.status);
data = co_await client.async_get(uri3);
print(data.status);
#endif
}
async_simple::coro::Lazy<void> test_upload() {
std::string uri = "http://example.com/";
coro_http_client client{};
auto result = co_await client.async_upload(uri, "test", "yourfile.jpg");
print(result.status);
std::cout << "upload finished\n";
client.add_str_part("hello", "coro_http_client");
client.add_file_part("test", "yourfile.jpg");
result = co_await client.async_upload(uri);
print(result.status);
std::cout << "upload finished\n";
}
async_simple::coro::Lazy<void> test_download() {
coro_http_client client{};
std::string uri =
"http://www.httpwatch.com/httpgallery/chunked/chunkedimage.aspx";
std::string filename = "test.jpg";
std::error_code ec{};
std::filesystem::remove(filename, ec);
auto r = co_await client.async_download(uri, filename);
assert(!r.net_err);
assert(r.status == 200);
std::cout << "download finished\n";
}
async_simple::coro::Lazy<void> test_websocket() {
coro_http_client client{};
auto r = co_await client.connect("ws://localhost:8090/ws");
if (r.net_err) {
co_return;
}
co_await client.write_websocket("hello websocket");
auto data = co_await client.read_websocket();
CHECK(data.resp_body == "hello websocket");
co_await client.write_websocket("test again");
data = co_await client.read_websocket();
CHECK(data.resp_body == "test again");
co_await client.write_websocket("ws close");
data = co_await client.read_websocket();
CHECK(data.net_err == asio::error::eof);
CHECK(data.resp_body == "ws close");
}
cinatra提供了一个高性能的http1.1 压测工具, 它是基于coro_http_client 实现的,内部通过多线程和协程实现了高效的压测,能够在单核或多核cpu上发送大量请求以此来测试服务器性能。
./cinatra_press_tool -t 4 -c 40 -d 30s http://127.0.0.1
上面的命令代表使用4个线程并且保持40个连接打开(协程)对网址http://127.0.0.1
进行30s的基准测试。
输出如下:
Running 30s test @ http://127.0.0.1
4 threads and 40 connections
Thread Status Avg Max Variation Stdev
Latency 4.12ms 8.15ms 3.367ms 1.835ms
462716 requests in 30.001s, 592.198250MB read, total: 462716, errors: 0
Requests/sec: 15423.86666667
Transfer/sec: 19.739390MB
-c, --connections total number of HTTP connections to keep open with
each thread handling N = connections/threads (int)
-d, --duration duration of the test, e.g. 2s, 2m, 2h (string [=15s])
-t, --threads total number of threads to use (int [=1])
-H, --headers HTTP headers to add to request, e.g. "User-Agent: coro_http_press"
add multiple http headers in a request need to be separated by ' && '
e.g. "User-Agent: coro_http_press && x-frame-options: SAMEORIGIN" (string [=])
-r, --readfix read fixed response (int [=0])
-?, --help print this message
这里有两个参数与wrk不同
-H
参数,它表示添加http头到http请求中,该参数不止可以添加一个http头还可以以&&
符号(4个字符)为分隔符来组装多个http头到http请求。
比如-H User-Agent: coro_http_press
就是添加一个http头,而-H User-Agent: coro_http_press && x-frame-options: SAMEORIGIN
则为添加User-Agent: coro_http_press
和x-frame-options: SAMEORIGIN
两个http头到http请求。添加三个以及多个http头的方法和上述方法相同。
-r
参数,它表示是否读固定长度的response,这个参数可以避免频繁的解析response优化性能,有些服务器对于相同的请求返回的长度可能不同,这种情况下不设置这个参数或者将它设置为0。
websocket的业务函数是会多次进入的,因此写业务逻辑的时候需要注意,推荐按照示例中的方式去做。
cinatra depends on asio and async_simple.
press_tool depends on cinatra and cmdline.
qq群:545605838