Before you start

Make sure that you can compile and run core tests as described at Configure, Build and Install.

Step by step guide

Let's write a simple TCP server that accepts incoming connections, and as long as the client sends "hi" responds with greeting from configuration file.

TCP server

Derive from components::TcpAcceptorBase and override the ProcessSocket function to get the new sockets:

#include <userver/components/minimal_component_list.hpp>
#include <userver/components/tcp_acceptor_base.hpp>
#include <userver/utils/daemon_run.hpp>
#include <userver/yaml_config/merge_schemas.hpp>
 
namespace samples::tcp {
 
class Hello final : public components::TcpAcceptorBase {
public:
    static constexpr std::string_view kName = "tcp-hello";
 
    // Component is valid after construction and is able to accept requests
    Hello(const components::ComponentConfig& config, const components::ComponentContext& context)
        : TcpAcceptorBase(config, context), greeting_(config["greeting"].As<std::string>("hi")) {}
 
    void ProcessSocket(engine::io::Socket&& sock) override;
 
    static yaml_config::Schema GetStaticConfigSchema();
 
private:
    const std::string greeting_;
};
 
}  // namespace samples::tcp
 

Warning: ProcessSocket functions are invoked concurrently on the same instance of the class. Use synchronization primitives or do not modify shared data in ProcessSocket.

Static config

Our new "tcp-hello" component should support the options of the components::TcpAcceptorBase and the "greeting" option. To achieve that we would need the following implementation of the GetStaticConfigSchema function:

yaml_config::Schema Hello::GetStaticConfigSchema() {
    return yaml_config::MergeSchemas<TcpAcceptorBase>(R"(
    type: object
    description: |
      Component for accepting incoming TCP connections and responding with some
      greeting as long as the client sends 'hi'.
    additionalProperties: false
    properties:
      greeting:
          type: string
          description: greeting to send to client
          defaultDescription: hi
  )");
}

Now lets configure our component in the components section:

        # yaml
        tcp-hello:
            task_processor: main-task-processor         # Run socket accepts on CPU bound task processor
            sockets_task_processor: main-task-processor # Run ProcessSocket() for each new socket on CPU bound task processor
            port: 8180
            greeting: hello

ProcessSocket

It's time to deal with new sockets. The code is quite straightforward:

void Hello::ProcessSocket(engine::io::Socket&& sock) {
    std::string data;
    data.resize(2);
 
    while (!engine::current_task::ShouldCancel()) {
        const auto read_bytes = sock.ReadAll(data.data(), 2, {});
        if (read_bytes != 2 || data != "hi") {
            sock.Close();
            return;
        }
 
        const auto sent_bytes = sock.SendAll(greeting_.data(), greeting_.size(), {});
        if (sent_bytes != greeting_.size()) {
            return;
        }
    }
}

int main()

Finally, add the component to the components::MinimalComponentList(), and start the server with static configuration file passed from command line.

int main(int argc, const char* const argv[]) {
    const auto component_list = components::MinimalComponentList().Append<samples::tcp::Hello>();
 
    return utils::DaemonMain(argc, argv, component_list);
}

Build and Run

To build the sample, execute the following build steps at the userver root directory:

mkdir build_release
cd build_release
cmake -DCMAKE_BUILD_TYPE=Release ..
make userver-samples-tcp_service

The sample could be started by running make start-userver-samples-tcp_service. The command would invoke testsuite start target that sets proper paths in the configuration files and starts the service.

To start the service manually run ./samples/tcp_service/userver-samples-tcp_service -c </path/to/static_config.yaml>.

Now you can send a request to your server from another terminal:

bash
$ nc localhost 8180
hi
hello

Functional testing

Functional tests for the service could be implemented using the testsuite in the following way:

import socket
 
import pytest
 
 
async def test_basic(service_client, loop, tcp_service_port):
    sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
    sock.connect(('localhost', tcp_service_port))
 
    await loop.sock_sendall(sock, b'hi')
    hello = await loop.sock_recv(sock, 5)
    assert hello == b'hello'
 
    await loop.sock_sendall(sock, b'whats up?')
    with pytest.raises(ConnectionResetError):
        await loop.sock_recv(sock, 1)

Note that in this case testsuite requires some help to detect that the service is ready to accept requests. To do that, override the pytest_userver.plugins.service.service_non_http_health_checks :

import pytest
from pytest_userver.utils import net
 
pytest_plugins = ['pytest_userver.plugins.core']
 
 
@pytest.fixture(name='tcp_service_port', scope='session')
def _tcp_service_port(service_config) -> int:
    components = service_config['components_manager']['components']
    tcp_hello = components.get('tcp-hello')
    assert tcp_hello, 'No "tcp-hello" component found'
    return int(tcp_hello['port'])
 
 
@pytest.fixture(scope='session')
def service_non_http_health_checks(
    service_config, tcp_service_port,
) -> net.HealthChecks:
    checks = net.get_health_checks_info(service_config)
    checks.tcp.append(net.HostPort(host='localhost', port=tcp_service_port))
    return checks

Full sources

See the full example at:

⇦ Production configs and best practices | TCP full-duplex server with metrics and Spans ⇨