Tars uses the [nghttp2 library] (https://www.nghttp2.org/) to support the http2 protocol. It encapsulates the protocol library with RPC, making it almost transparent to business development and very convenient to use. Because of reusing tars RPC, it also has the characteristics of synchronization, asynchrony and timeout, and can use tars stat to report call quality.
This article briefly introduces the methods and steps of using http2, including synchronous call and asynchronous call.
You can find demo: cpp/example/HttpDemo
By default, the tars framework does not turn on http2. Turn on http2:
cmake .. -DTARS_HTTP2=ON
rebuild tarscpp.
see demo: cpp/example/HttpDemo/Http2Server
Custom protocol with http2:
void HttpServer::initialize()
{
addServant<Http2Imp>(ServerConfig::Application + "." + ServerConfig::ServerName + ".Http2Obj");
addServantProtocol(ServerConfig::Application + "." + ServerConfig::ServerName + ".Http2Obj", &parseHttp2);
}
Protocol resolver function. Note that each connection creates and stores a TC_Http2server as the session controller of http2
TC_NetWorkBuffer::PACKET_TYPE parseHttp2(TC_NetWorkBuffer&in, vector<char> &out)
{
TC_Http2Server*sessionPtr = (TC_Http2Server*)(in.getContextData());
if(sessionPtr == NULL)
{
shared_ptr<TC_Http2Server> session(new TC_Http2Server());
in.setContextData(session.get());
session->settings(3000);
TC_EpollServer::Connection *connection = (TC_EpollServer::Connection *)in.getConnection();
Http2Imp::addHttp2(connection->getId(), session);
sessionPtr = session.get();
}
return sessionPtr->parse(in, out);
}
In the service implementation, the session (TC_Http2Server object) of http2 is obtained according to the connection ID and the protocol is parsed.
int Http2Imp::doRequest(TarsCurrentPtr current, vector<char> &buffer)
{
shared_ptr<TC_Http2Server> session = getHttp2(current->getUId());
vector<shared_ptr<TC_Http2Server::Http2Context>> contexts = session->decodeRequest();
for(size_t i = 0; i< contexts.size(); ++i)
{
shared_ptr<TC_Http2Server::Http2Context> context = contexts[i];
vector<char> data;
context->response.setHeader("X-Header", "TARS");
context->response.setResponse(200, "OK", context->request.getContent());
int ret = session->encodeResponse(context, data);
if(ret != 0)
{
cout << "encodeResponse error:" << session->getErrMsg() << endl;
}
buffer.insert(buffer.end(), data.begin(), data.end());
}
return 0;
}
int Http2Imp::doClose(TarsCurrentPtr current)
{
delHttp2(current->getUId());
return 0;
}
see demo: cpp/example/HttpDemo/Http2Client
CommunicatorPtr& comm = Application::getCommunicator();
ServantPrx prx = comm->stringToProxy<ServantPrx>("test.server.yourobj");
// set protocol
prox->tars_set_protocol(ServantProxy::PROTOCOL_HTTP2);
shared_ptr<TC_HttpResponse> rsp;
shared_ptr<TC_HttpRequest> req = std::make_shared<TC_HttpRequest>();
req->setPostRequest("http://domain.com/hello", string("helloworld-") + TC_Common::tostr(i), true);
prx->http_call("hello", req, rsp);
Note: the first parameter of http_call here has no actual function, but is used for monitoring. That is to say, the function name is "hello" in the interface statistics of web management platform
// 编写callback
class MyHttpCb : public HttpCallback
{
public:
virtual int onHttpResponse(const shared_ptr<TC_HttpResponse> &rsp)
{
return 0;
}
virtual int onHttpResponseException(int expCode)
{
return 0;
}
};
shared_ptr<TC_HttpRequest> req = std::make_shared<TC_HttpRequest>();
string buff = string("helloworld-") + TC_Common::tostr(i);
req->setPostRequest("http://domain.com/hello", buff, true);
prx->http_call_async("hello", req, p);
Note: the first parameter of http_call_async here has no actual function, but is used for monitoring. That is to say, the function name is "hello" in the interface statistics of web management platform