OrcaSlicer/tests/slic3rutils/test_moonraker_printer_agent.cpp

#include <catch2/catch_message.hpp>
#include <catch2/catch_test_macros.hpp>
#include <catch2/generators/catch_generators.hpp>

#include <atomic>
#include <chrono>
#include <condition_variable>
#include <cstddef>
#include <functional>
#include <future>
#include <memory>
#include <mutex>
#include <string>
#include <thread>
#include <utility>
#include <vector>

#include <boost/asio/buffer.hpp>
#include <boost/asio/io_context.hpp>
#include <boost/asio/ip/address.hpp>
#include <boost/asio/ip/tcp.hpp>
#include <boost/beast/core/error.hpp>
#include <boost/beast/core/flat_buffer.hpp>
#include <boost/beast/http/message.hpp>
#include <boost/beast/http/read.hpp>
#include <boost/beast/http/status.hpp>
#include <boost/beast/http/string_body.hpp>
#include <boost/beast/http/write.hpp>
#include <boost/beast/websocket/error.hpp>
#include <boost/beast/websocket/rfc6455.hpp>
#include <boost/beast/websocket/stream.hpp>
#include <nlohmann/json.hpp>

#include "slic3r/Utils/IPrinterAgent.hpp"
#include "slic3r/Utils/MoonrakerPrinterAgent.hpp"

using namespace Slic3r;

namespace {

namespace beast     = boost::beast;
namespace http      = beast::http;
namespace websocket = beast::websocket;
namespace net       = boost::asio;
using tcp           = net::ip::tcp;
using namespace std::chrono_literals;

class DiscoveryAgent : public MoonrakerPrinterAgent
{
public:
    using MoonrakerPrinterAgent::MoonrakerPrinterAgent;
    using MoonrakerPrinterAgent::device_info;
    mutable unsigned fetch_count = 0;

protected:
    bool fetch_device_info(const ConnectionSettings&, MoonrakerDeviceInfo& info, std::string&) const override
    {
        ++fetch_count;
        info.dev_name = "Test printer";
        return true;
    }
};

class StreamAgent : public MoonrakerPrinterAgent
{
public:
    using MoonrakerPrinterAgent::MoonrakerPrinterAgent;
    using MoonrakerPrinterAgent::ws_timings;

    // Starts the stream the way a connection does, from a snapshot of the connection settings.
    void start_stream(const std::string& base_url, bool use_ssl = false)
    {
        const std::string dev_id = "printer";
        ConnectionSettings connection;
        connection.dev_id   = dev_id;
        connection.base_url = base_url;
        connection.use_ssl  = use_ssl;
        start_status_stream(dev_id, std::move(connection));
    }

    void stop_stream() { stop_status_stream(); }

protected:
    bool init_device_info(const PrinterConnectionParams& params) override
    {
        device_info          = MoonrakerDeviceInfo{};
        device_info.dev_id   = params.dev_id;
        device_info.dev_ip   = params.host;
        device_info.base_url = std::string(params.use_ssl ? "https://" : "http://") + params.host + ":" + params.port;
        return true;
    }

    bool fetch_device_info(const ConnectionSettings&, MoonrakerDeviceInfo& info, std::string&) const override
    {
        info.dev_name = "Test printer";
        return true;
    }
};

enum class Behavior {
    SilentBeforeHandshake,     // Accepts the connection and never answers
    Silent,                    // Completes the upgrade, then never reads or sends
    AnswersPings,              // Reports a status, then answers pings and sends nothing else
    AnswersPingsWithoutStatus, // Answers pings but never reports a status
    Chatters,                  // Reports a status, then answers pings and sends notifications that carry no status
    HangsUp,                   // Reports a status, then closes the websocket
    HangsUpWithoutAnswer,      // Closes the websocket without sending anything
    IgnoresClose,              // Reports Klippy disconnected, then never answers the close
    StallsObjectList,          // Never answers the stream's object list request
    StallsStatusQuery,         // Never answers HTTP requests until released, then hangs up on them
    RejectsTls,                // Records the first byte of the stream, which a TLS client sends as a handshake
};

class FakePrinter
{
public:
    explicit FakePrinter(Behavior behavior) : m_behavior(behavior)
    {
        m_accept_thread = std::thread([this] { accept_loop(); });
    }

    ~FakePrinter()
    {
        release();
        m_stopping = true;
        // Wakes the blocking accept.
        tcp::socket wake{m_accept_ioc};
        beast::error_code ec;
        wake.connect(m_acceptor.local_endpoint(), ec);
        m_accept_thread.join();
        for (std::thread& connection : m_connections)
            connection.join();
    }

    std::string host() const { return "127.0.0.1"; }
    std::string port() const { return std::to_string(m_acceptor.local_endpoint().port()); }
    std::string base_url() const { return "http://" + host() + ":" + port(); }

    // A connection is ready once it reaches the state its behavior waits in.
    bool wait_for_ready_connections(int count, std::chrono::milliseconds timeout = 10s)
    {
        std::unique_lock<std::mutex> lock(m_mutex);
        return m_changed.wait_for(lock, timeout, [&] { return m_ready >= count; });
    }

    bool wait_for_http_requests(int count)
    {
        std::unique_lock<std::mutex> lock(m_mutex);
        return m_changed.wait_for(lock, 10s, [&] { return m_http_requests >= count; });
    }

    // Waits until the printer, once released, has hung up on `count` stalled HTTP requests.
    bool wait_for_http_hang_ups(int count)
    {
        std::unique_lock<std::mutex> lock(m_mutex);
        return m_changed.wait_for(lock, 10s, [&] { return m_http_hung_up >= count; });
    }

    int close_frames()
    {
        std::lock_guard<std::mutex> lock(m_mutex);
        return m_close_frames;
    }

    // First byte the stream sent, or -1 until it has sent one.
    bool wait_for_first_byte()
    {
        std::unique_lock<std::mutex> lock(m_mutex);
        return m_changed.wait_for(lock, 10s, [&] { return m_first_byte >= 0; });
    }

    int first_byte()
    {
        std::lock_guard<std::mutex> lock(m_mutex);
        return m_first_byte;
    }

    // Hangs up every connection that is waiting.
    void release()
    {
        std::lock_guard<std::mutex> lock(m_mutex);
        m_released = true;
        m_changed.notify_all();
    }

private:
    void accept_loop()
    {
        for (;;) {
            // Each connection gets its own io_context, so a connection that runs one never runs another's handlers.
            auto ioc = std::make_unique<net::io_context>();
            tcp::socket socket{*ioc};
            beast::error_code ec;
            m_acceptor.accept(socket, ec);
            if (ec || m_stopping)
                return;
            m_connections.emplace_back(
                [this, ioc = std::move(ioc), socket = std::move(socket)]() mutable { serve(*ioc, std::move(socket)); });
        }
    }

    void serve(net::io_context& ioc, tcp::socket socket)
    {
        if (m_behavior == Behavior::SilentBeforeHandshake) {
            mark_ready();
            wait_for_release();
            return;
        }

        if (m_behavior == Behavior::RejectsTls) {
            char              first = 0;
            beast::error_code ec;
            socket.read_some(net::buffer(&first, 1), ec);
            if (!ec) {
                std::lock_guard<std::mutex> lock(m_mutex);
                m_first_byte = static_cast<unsigned char>(first);
                m_changed.notify_all();
            }
            mark_ready();
            wait_for_release();
            return;
        }

        beast::flat_buffer buffer;
        http::request<http::string_body> request;
        beast::error_code ec;
        http::read(socket, buffer, request, ec);
        if (ec)
            return;
        // The agent also queries the printer's status and object list over plain HTTP.
        if (!websocket::is_upgrade(request)) {
            serve_http(socket, request);
            return;
        }

        websocket::stream<tcp::socket> ws{std::move(socket)};
        ws.accept(request, ec);
        // Without its object list the stream never sends the subscribe request.
        const int requests = m_behavior == Behavior::StallsObjectList ? 1 : 2;
        for (int i = 0; i < requests && !ec; ++i) {
            beast::flat_buffer message;
            ws.read(message, ec);
        }
        if (!ec && (m_behavior == Behavior::AnswersPings || m_behavior == Behavior::Chatters || m_behavior == Behavior::HangsUp)) {
            const std::string status =
                nlohmann::json{{"jsonrpc", "2.0"}, {"id", 1}, {"result", {{"status", {{"print_stats", {{"state", "standby"}}}}}}}}.dump();
            ws.write(net::buffer(status), ec);
        }
        if (ec)
            return;

        switch (m_behavior) {
        case Behavior::AnswersPings:
        case Behavior::AnswersPingsWithoutStatus:
        case Behavior::Chatters:
            mark_ready();
            answer_pings(ioc, ws, m_behavior == Behavior::Chatters);
            break;
        case Behavior::HangsUp:
        case Behavior::HangsUpWithoutAnswer:
            ws.close(websocket::close_code::normal, ec);
            mark_ready();
            break;
        case Behavior::IgnoresClose: {
            const std::string notification = nlohmann::json{{"jsonrpc", "2.0"}, {"method", "notify_klippy_disconnected"}}.dump();
            ws.write(net::buffer(notification), ec);
            mark_ready();
            wait_for_release();
            break;
        }
        case Behavior::StallsObjectList: answer_pings(ioc, ws, false); break;
        default: mark_ready(); wait_for_release();
        }
    }

    void serve_http(tcp::socket& socket, const http::request<http::string_body>& request)
    {
        {
            std::lock_guard<std::mutex> lock(m_mutex);
            ++m_http_requests;
            m_changed.notify_all();
        }
        if (m_behavior == Behavior::StallsObjectList) {
            mark_ready();
            wait_for_release();
            return;
        }
        if (m_behavior == Behavior::StallsStatusQuery) {
            wait_for_release();
            beast::error_code ec;
            socket.close(ec);
            std::lock_guard<std::mutex> lock(m_mutex);
            ++m_http_hung_up;
            m_changed.notify_all();
            return;
        }
        http::response<http::string_body> response{http::status::not_found, request.version()};
        response.prepare_payload();
        beast::error_code ec;
        http::write(socket, response, ec);
    }

    // Keeps a read pending, which answers pings, until released or closed, and with `chatter` also sends a notification every 50 ms.
    void answer_pings(net::io_context& ioc, websocket::stream<tcp::socket>& ws, bool chatter)
    {
        const std::string notification = nlohmann::json{{"jsonrpc", "2.0"},
                                                        {"method", "notify_proc_stat_update"},
                                                        {"params", nlohmann::json::array({nlohmann::json::object()})}}
                                             .dump();
        beast::flat_buffer message;
        bool reading                    = true;
        bool writing                    = false;
        std::function<void()> read_next = [&] {
            ws.async_read(message, [&](beast::error_code ec, std::size_t) {
                message.clear();
                if (ec == websocket::error::closed) {
                    std::lock_guard<std::mutex> lock(m_mutex);
                    ++m_close_frames;
                }
                if (ec)
                    reading = false;
                else
                    read_next();
            });
        };
        read_next();

        auto next_notification = std::chrono::steady_clock::now();
        while (reading && !wait_for_release(10ms)) {
            if (chatter && !writing && std::chrono::steady_clock::now() >= next_notification) {
                writing = true;
                next_notification += 50ms;
                ws.async_write(net::buffer(notification), [&](beast::error_code, std::size_t) { writing = false; });
            }
            ioc.restart();
            ioc.run_for(20ms);
        }
        beast::error_code ignored;
        ws.next_layer().close(ignored);
        while (reading || writing) {
            ioc.restart();
            ioc.run_for(20ms);
        }
    }

    void mark_ready()
    {
        std::lock_guard<std::mutex> lock(m_mutex);
        ++m_ready;
        m_changed.notify_all();
    }

    void wait_for_release()
    {
        std::unique_lock<std::mutex> lock(m_mutex);
        m_changed.wait(lock, [this] { return m_released; });
    }

    bool wait_for_release(std::chrono::milliseconds timeout)
    {
        std::unique_lock<std::mutex> lock(m_mutex);
        return m_changed.wait_for(lock, timeout, [this] { return m_released; });
    }

    const Behavior m_behavior;
    net::io_context m_accept_ioc;
    tcp::acceptor m_acceptor{m_accept_ioc, {net::ip::make_address("127.0.0.1"), 0}};
    std::atomic<bool> m_stopping{false};
    std::thread m_accept_thread;
    std::vector<std::thread> m_connections;
    std::mutex m_mutex;
    std::condition_variable m_changed;
    int m_ready         = 0;
    int m_http_requests = 0;
    int m_http_hung_up  = 0;
    int m_close_frames  = 0;
    int m_first_byte    = -1;
    bool m_released     = false;
};

// Records when the agent dispatches a status report.
class Reports
{
public:
    void attach(StreamAgent& agent)
    {
        agent.set_on_local_message_fn([this](std::string, std::string) {
            std::lock_guard<std::mutex> lock(m_mutex);
            m_times.push_back(std::chrono::steady_clock::now());
            m_changed.notify_all();
        });
    }

    bool wait_for(std::size_t count)
    {
        std::unique_lock<std::mutex> lock(m_mutex);
        return m_changed.wait_for(lock, 10s, [&] { return m_times.size() >= count; });
    }

    std::vector<std::chrono::steady_clock::time_point> times()
    {
        std::lock_guard<std::mutex> lock(m_mutex);
        return m_times;
    }

private:
    std::mutex m_mutex;
    std::condition_variable m_changed;
    std::vector<std::chrono::steady_clock::time_point> m_times;
};

// A printer and a stream agent connected to it.
struct StreamFixture
{
    explicit StreamFixture(Behavior behavior) : printer(behavior) {}

    // Runs before the members are destroyed, so the agent's destructor never joins a stream the printer still holds open.
    ~StreamFixture() { printer.release(); }

    void start() { agent.start_stream(printer.base_url()); }

    // Runs the stream as a wss client; the printer must be served on the same port.
    void start_secure() { agent.start_stream("https://" + printer.host() + ":" + printer.port(), true); }

    // Stops the stream, giving it `limit` to return before the printer hangs up on it.
    bool stops_within(std::chrono::milliseconds limit)
    {
        auto stopped       = std::async(std::launch::async, [this] { agent.stop_stream(); });
        const bool in_time = stopped.wait_for(limit) == std::future_status::ready;
        printer.release();
        stopped.wait();
        return in_time;
    }

    FakePrinter printer;
    StreamAgent agent{""};
};

} // namespace

TEST_CASE("Moonraker discovery waits for a complete device identity", "[MoonrakerPrinterAgent][Discovery]")
{
    DiscoveryAgent agent("");
    unsigned announcements = 0;
    nlohmann::json announcement;
    agent.set_on_ssdp_msg_fn([&](std::string payload) {
        ++announcements;
        announcement = nlohmann::json::parse(payload);
    });

    agent.start_discovery(true, false);
    REQUIRE(announcements == 0);
    REQUIRE(agent.fetch_count == 0);

    agent.device_info.dev_id = "127.0.0.1:7125";
    agent.device_info.dev_ip = agent.device_info.dev_id;
    agent.device_info.base_url = "http://127.0.0.1:7125";
    agent.start_discovery(true, false);
    REQUIRE(announcements == 1);
    REQUIRE(agent.fetch_count == 1);
    REQUIRE(announcement.at("dev_id") == agent.device_info.dev_id);
    REQUIRE(announcement.at("dev_ip") == agent.device_info.dev_ip);
    REQUIRE(announcement.at("dev_name") == "Test printer");
}

TEST_CASE("Stopping the status stream gives a silent printer no more than its timeouts", "[MoonrakerPrinterAgent][StatusStream]")
{
    // An open websocket gets the 200 ms close timeout, and a stalled object list request up to 1 s of curl progress to see the stop.
    const auto [behavior, limit] = GENERATE(table<Behavior, std::chrono::milliseconds>({
        {Behavior::SilentBeforeHandshake, 500ms},
        {Behavior::Silent, 1000ms},
        {Behavior::StallsObjectList, 2000ms},
    }));
    CAPTURE(static_cast<int>(behavior));

    StreamFixture f(behavior);
    f.agent.ws_timings.close = 200ms;
    f.start();
    REQUIRE(f.printer.wait_for_ready_connections(1));

    CHECK(f.stops_within(limit));
}

TEST_CASE("A secure status stream opens with a TLS handshake", "[MoonrakerPrinterAgent][StatusStream]")
{
    StreamFixture f(Behavior::RejectsTls);
    f.start_secure();
    REQUIRE(f.printer.wait_for_first_byte());
    // 0x16 is a TLS handshake record; a plaintext client would send "GET".
    CHECK(f.printer.first_byte() == 0x16);
    CHECK(f.stops_within(3s));
}

TEST_CASE("Stopping the status stream sends the printer a close frame", "[MoonrakerPrinterAgent][StatusStream]")
{
    // A stop while the stream reads, and one while it waits on the object list request.
    const Behavior behavior = GENERATE(Behavior::AnswersPings, Behavior::StallsObjectList);
    CAPTURE(static_cast<int>(behavior));

    StreamFixture f(behavior);
    f.start();
    REQUIRE(f.printer.wait_for_ready_connections(1));

    CHECK(f.stops_within(3s));
    CHECK(f.printer.close_frames() == 1);
}

TEST_CASE("Stopping the status stream does not wait out the reconnect delay", "[MoonrakerPrinterAgent][StatusStream]")
{
    StreamFixture f(Behavior::HangsUp);
    f.agent.ws_timings.reconnect_delay = 10s;
    f.start();
    REQUIRE(f.printer.wait_for_ready_connections(1));
    // Lets the stream reach its reconnect wait.
    std::this_thread::sleep_for(200ms);

    CHECK(f.stops_within(3s));
}

TEST_CASE("The status stream reconnects when the printer stops answering", "[MoonrakerPrinterAgent][StatusStream]")
{
    // An unanswered close ends at whichever of the handshake and close timeouts is shorter.
    const auto [behavior, handshake, close] = GENERATE(table<Behavior, std::chrono::milliseconds, std::chrono::milliseconds>({
        {Behavior::Silent, 300ms, 1000ms},
        {Behavior::IgnoresClose, 300ms, 1000ms},
        {Behavior::IgnoresClose, 1000ms, 200ms},
    }));
    CAPTURE(static_cast<int>(behavior), handshake.count(), close.count());

    StreamFixture f(behavior);
    f.agent.ws_timings.handshake       = handshake;
    f.agent.ws_timings.close           = close;
    f.agent.ws_timings.idle            = 400ms;
    f.agent.ws_timings.reconnect_delay = 20ms;
    f.start();

    CHECK(f.printer.wait_for_ready_connections(2));
    CHECK(f.stops_within(3s));
}

TEST_CASE("The status stream reconnects at the shortest delay while the printer answers", "[MoonrakerPrinterAgent][StatusStream]")
{
    StreamFixture f(Behavior::HangsUp);
    f.agent.ws_timings.reconnect_delay = 50ms;
    f.start();

    // Twelve 50 ms waits; a backoff that kept doubling would need over 10 s.
    CHECK(f.printer.wait_for_ready_connections(12, 3s));
    CHECK(f.stops_within(3s));
}

TEST_CASE("The status stream keeps retrying a printer that never answers", "[MoonrakerPrinterAgent][StatusStream]")
{
    StreamFixture f(Behavior::HangsUpWithoutAnswer);
    f.agent.ws_timings.reconnect_delay = 1ms;
    f.start();

    // The backoff stops doubling at 32 ms, so 20 attempts take well under the wait.
    CHECK(f.printer.wait_for_ready_connections(20));
    CHECK(f.stops_within(3s));
}

TEST_CASE("A callback on the stream thread can stop the status stream", "[MoonrakerPrinterAgent][StatusStream]")
{
    std::promise<void> stopped;
    std::once_flag stopped_once;
    StreamFixture f(Behavior::AnswersPings);
    f.agent.set_on_local_message_fn([&](std::string, std::string) {
        std::call_once(stopped_once, [&] {
            f.agent.stop_stream();
            stopped.set_value();
        });
    });
    f.start();

    REQUIRE(stopped.get_future().wait_for(5s) == std::future_status::ready);
    CHECK(f.stops_within(3s));
}

TEST_CASE("A quiet printer that is still up keeps reporting its status", "[MoonrakerPrinterAgent][StatusStream]")
{
    Reports reports;
    StreamFixture f(Behavior::AnswersPings);
    f.agent.ws_timings.idle      = 400ms;
    f.agent.ws_timings.heartbeat = 300ms;
    reports.attach(f.agent);
    f.start();

    // The status the printer sent, then two heartbeats.
    REQUIRE(reports.wait_for(3));
    const auto times = reports.times();
    // The steady clock is compared in whole milliseconds, so allow 1 ms of rounding.
    CHECK(times[1] - times[0] >= 299ms);
    CHECK(times[2] - times[1] >= 299ms);
    CHECK(f.stops_within(3s));
}

TEST_CASE("The status stream sends no heartbeat while the printer has nothing new to report", "[MoonrakerPrinterAgent][StatusStream]")
{
    const Behavior behavior = GENERATE(Behavior::Chatters, Behavior::AnswersPingsWithoutStatus);
    CAPTURE(static_cast<int>(behavior));

    Reports reports;
    StreamFixture f(behavior);
    f.agent.ws_timings.idle      = 400ms;
    f.agent.ws_timings.heartbeat = 800ms;
    reports.attach(f.agent);
    f.start();
    REQUIRE(f.printer.wait_for_ready_connections(1));

    // The chattering printer's status, and nothing from the one that never sent a status.
    const std::size_t expected = behavior == Behavior::Chatters ? 1 : 0;
    std::this_thread::sleep_for(1500ms);
    CHECK(reports.times().size() == expected);
    CHECK(f.stops_within(3s));
}

TEST_CASE("A disconnect during the connection queries keeps the status stream from starting", "[MoonrakerPrinterAgent][StatusStream]")
{
    StreamFixture f(Behavior::StallsStatusQuery);
    PrinterConnectionParams params;
    params.dev_id = "printer";
    params.host   = f.printer.host();
    params.port   = f.printer.port();
    f.agent.connect_printer(params);
    REQUIRE(f.printer.wait_for_http_requests(1));

    // disconnect_printer joins the worker running the stalled query, so it is called aside and the
    // printer is released to let that query return.
    auto disconnecting = std::async(std::launch::async, [&] { f.agent.disconnect_printer(); });
    f.printer.release();
    // The query has returned once the printer hangs up on it, and a stream would connect right after.
    REQUIRE(f.printer.wait_for_http_hang_ups(1));
    CHECK_FALSE(f.printer.wait_for_ready_connections(1, 500ms));
    disconnecting.wait();
}