vdi-cloud/terminal/kiosk-shell/src/monitor.cpp

#include "monitor.h"

#include <signal.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <sys/wait.h>
#include <unistd.h>

#include <cerrno>
#include <chrono>
#include <cstdio>
#include <cstring>
#include <ctime>
#include <string>
#include <thread>
#include <vector>

namespace vdi {
namespace {

Monitor* g_monitor = nullptr;

void onSignal(int) {
    if (g_monitor) {
        g_monitor->requestStop();
    }
}

// 主进程写心跳文件；Monitor 读取其修改时间判断是否假死
bool heartbeatAgeMs(const std::string& path, long long& ageMs) {
    struct stat st {};
    if (stat(path.c_str(), &st) != 0) {
        return false;
    }
    const auto now = std::chrono::system_clock::now();
    const auto mtime = std::chrono::system_clock::from_time_t(st.st_mtime);
    ageMs = std::chrono::duration_cast<std::chrono::milliseconds>(now - mtime).count();
    return true;
}

}  // namespace

Monitor::Monitor(MonitorConfig config) : config_(std::move(config)) {}

void Monitor::requestStop() {
    stopRequested_ = true;
}

void Monitor::log(const std::string& message) const {
    if (!config_.verbose) {
        return;
    }
    const std::time_t now = std::time(nullptr);
    char timeBuf[32] = {0};
    std::strftime(timeBuf, sizeof(timeBuf), "%Y-%m-%d %H:%M:%S", std::localtime(&now));
    std::fprintf(stderr, "[monitor %s] %s\n", timeBuf, message.c_str());
    std::fflush(stderr);
}

bool Monitor::isProcessAlive(pid_t pid) const {
    if (pid <= 0) {
        return false;
    }
    return kill(pid, 0) == 0 || errno == EPERM;
}

// 结束主进程及其整棵进程树。
// 主进程通常还会再 fork 子进程（例如启动器拉起的串流客户端），
// 只杀直接子进程会留下孤儿进程继续占用显示设备和串流会话，
// 因此子进程启动时独立成组，结束时按进程组终止。
namespace {

void killTree(pid_t pid, int sig) {
    if (pid <= 0) {
        return;
    }
    if (kill(-pid, sig) != 0) {
        kill(pid, sig);
    }
}

}  // namespace

bool Monitor::isHung() const {
    long long ageMs = 0;
    if (!heartbeatAgeMs(config_.heartbeatFile, ageMs)) {
        // 没有心跳文件：主进程还没写过心跳
        return false;
    }
    return ageMs > config_.heartbeatTimeoutMs;
}

int Monitor::run(const std::string& command) {
    if (command.empty()) {
        log("用法: monitor -- <主进程命令行>");
        return 2;
    }

    g_monitor = this;
    signal(SIGTERM, onSignal);
    signal(SIGINT, onSignal);
    // 子进程退出时不发送 SIGPIPE，避免 Monitor 自身被杀
    signal(SIGPIPE, SIG_IGN);

    log("看门狗启动，监控: " + command);

    pid_t child = -1;
    auto startTime = std::chrono::steady_clock::now();

    while (!stopRequested_) {
        if (child <= 0) {
            // 启动主进程
            child = fork();
            if (child == 0) {
                // 子进程：恢复默认信号处理并 exec
                signal(SIGTERM, SIG_DFL);
                signal(SIGINT, SIG_DFL);
                signal(SIGPIPE, SIG_DFL);
                // 独立成组，便于结束整棵进程树
                setpgid(0, 0);
                execl("/bin/sh", "sh", "-c", command.c_str(), nullptr);
                _exit(127);
            }
            if (child < 0) {
                log(std::string("fork 失败: ") + std::strerror(errno));
                return 1;
            }
            // 父子都调用一次 setpgid，避免父子竞争导致进程组 ID 未及时生效
            setpgid(child, child);
            startTime = std::chrono::steady_clock::now();
            log("主进程已启动, pid=" + std::to_string(child));
        }

        std::this_thread::sleep_for(std::chrono::milliseconds(config_.checkIntervalMs));

        if (stopRequested_) {
            log("收到退出信号，终止主进程");
            if (child > 0) {
                killTree(child, SIGTERM);
                // 等待退出，超时则强杀
                for (int i = 0; i < 20 && isProcessAlive(child); ++i) {
                    std::this_thread::sleep_for(std::chrono::milliseconds(200));
                }
                if (isProcessAlive(child)) {
                    killTree(child, SIGKILL);
                }
                waitpid(child, nullptr, 0);
            }
            log("看门狗退出");
            return 0;
        }

        // 回收已退出的子进程
        int status = 0;
        const pid_t reaped = waitpid(child, &status, WNOHANG);
        if (reaped == child) {
            const int exitCode = WIFEXITED(status) ? WEXITSTATUS(status) : -1;
            const int signalNum = WIFSIGNALED(status) ? WTERMSIG(status) : 0;
            log("主进程已退出, code=" + std::to_string(exitCode) +
                (signalNum ? (", signal=" + std::to_string(signalNum)) : ""));

            if (++restartAttempts_ > config_.maxRestartAttempts) {
                log("连续重启超过 " + std::to_string(config_.maxRestartAttempts) +
                    " 次，放弃重启（熔断）");
                return 1;
            }
            log(std::to_string(config_.restartBackoffMs) + "ms 后重启");
            std::this_thread::sleep_for(std::chrono::milliseconds(config_.restartBackoffMs));
            child = -1;
            continue;
        }

        // 假死检测
        if (isHung()) {
            const auto elapsed =
                std::chrono::duration_cast<std::chrono::milliseconds>(
                    std::chrono::steady_clock::now() - startTime)
                    .count();
            // 启动宽限期内不判定假死
            if (elapsed > config_.startupGraceMs) {
                log("检测到主进程假死（心跳超时），执行 kill 并重启");
                killTree(child, SIGKILL);
                waitpid(child, &status, 0);
                child = -1;

                if (++restartAttempts_ > config_.maxRestartAttempts) {
                    log("连续重启超过 " + std::to_string(config_.maxRestartAttempts) +
                        " 次，放弃重启（熔断）");
                    return 1;
                }
                std::this_thread::sleep_for(std::chrono::milliseconds(config_.restartBackoffMs));
                continue;
            }
        } else {
            // 进程健康，重置重启计数
            restartAttempts_ = 0;
        }
    }

    return 0;
}

}  // namespace vdi