zlh-agent/internal/system/process.go

package system

import (
	"bytes"
	"errors"
	"fmt"
	"log"
	"os"
	"os/exec"
	"os/user"
	"path/filepath"
	"strconv"
	"strings"
	"sync"
	"syscall"
	"time"

	"zlh-agent/internal/provision"
	"zlh-agent/internal/provision/devcontainer"
	"zlh-agent/internal/runtime"
	"zlh-agent/internal/state"
)

/* --------------------------------------------------------------------------
   GLOBAL PROCESS STATE
----------------------------------------------------------------------------*/

var (
	mu              sync.Mutex
	serverCmd       *exec.Cmd
	serverPTY       *os.File
	serverStartTime time.Time

	devCmd *exec.Cmd
	devPTY *os.File
)

func GetServerPID() (int, bool) {
	mu.Lock()
	defer mu.Unlock()

	if serverCmd == nil || serverCmd.Process == nil {
		return 0, false
	}
	return serverCmd.Process.Pid, true
}

/* --------------------------------------------------------------------------
   StartServer (fixed)
----------------------------------------------------------------------------*/

func StartServer(cfg *state.Config) error {
	mu.Lock()
	defer mu.Unlock()

	// Already running?
	if serverCmd != nil {
		return fmt.Errorf("server already running")
	}

	dir := provision.ServerDir(*cfg)
	startScript := filepath.Join(dir, "start.sh")
	log.Printf("[process] vmid=%d server start requested dir=%s", cfg.VMID, dir)

	cmd := exec.Command("/bin/bash", startScript)
	cmd.Dir = dir

	ptmx, err := runtime.CreatePTY(cmd)
	if err != nil {
		return fmt.Errorf("start server: %w", err)
	}

	serverCmd = cmd
	serverPTY = ptmx
	serverStartTime = time.Now()

	if state.GetState() != state.StateCrashed {
		gameSupervisor.Reset()
	}
	state.ClearIntentionalStop()
	state.SetState(state.StateRunning)
	state.SetError(nil)
	state.SetReadyState(false, "", "")
	log.Printf("[process] vmid=%d server process started", cfg.VMID)

	go gameSupervisor.Watch(cfg, cmd, ptmx, serverStartTime)

	return nil
}

/* --------------------------------------------------------------------------
   StopServer
----------------------------------------------------------------------------*/

func StopServer() error {
	mu.Lock()
	defer mu.Unlock()

	if serverCmd == nil {
		return fmt.Errorf("server not running")
	}

	cfg, _ := state.LoadConfig()
	if cfg != nil {
		log.Printf("[process] vmid=%d stop requested", cfg.VMID)
	} else {
		log.Printf("[process] stop requested")
	}
	state.SetState(state.StateStopping)
	state.MarkIntentionalStop()
	state.SetReadyState(false, "", "")

	// Try graceful stop
	if serverPTY != nil {
		_ = runtime.Write(serverPTY, []byte("save-all\n"))
		time.Sleep(2 * time.Second)
		_ = runtime.Write(serverPTY, []byte("stop\n"))
	}

	// Wait a moment
	time.Sleep(4 * time.Second)

	// If still running, force kill
	if serverCmd.Process != nil {
		_ = serverCmd.Process.Kill()
	}

	return nil
}

func WaitForServerExit(timeout time.Duration) error {
	deadline := time.Now().Add(timeout)
	for {
		mu.Lock()
		running := serverCmd != nil
		mu.Unlock()

		if !running {
			return nil
		}
		if time.Now().After(deadline) {
			return fmt.Errorf("timeout waiting for server process to exit")
		}
		time.Sleep(200 * time.Millisecond)
	}
}

/* --------------------------------------------------------------------------
   RestartServer
----------------------------------------------------------------------------*/

func RestartServer(cfg *state.Config) error {
	if err := StopServer(); err != nil {
		// ignore if not running
	}

	return StartServer(cfg)
}

/* --------------------------------------------------------------------------
   SendConsoleCommand
----------------------------------------------------------------------------*/

func SendConsoleCommand(cmd string) error {
	mu.Lock()
	defer mu.Unlock()

	if serverPTY == nil {
		return fmt.Errorf("server console not available")
	}

	return runtime.Write(serverPTY, []byte(cmd+"\n"))
}

/* --------------------------------------------------------------------------
   Dev Shell PTY
----------------------------------------------------------------------------*/

func StartDevShell() (*os.File, error) {
	mu.Lock()
	defer mu.Unlock()

	if devPTY != nil && devCmd != nil {
		return devPTY, nil
	}
	if err := devcontainer.EnsureDevUserEnvironment(); err != nil {
		return nil, fmt.Errorf("prepare dev environment: %w", err)
	}

	shell := "/bin/bash"
	if _, err := os.Stat(shell); err != nil {
		shell = "/bin/sh"
	}

	var cmd *exec.Cmd
	if shell == "/bin/bash" {
		cmd = exec.Command(shell, "-l", "-i")
	} else {
		cmd = exec.Command(shell, "-i")
	}
	cmd.Dir = devcontainer.WorkspaceDir

	devUser, err := user.Lookup(devcontainer.DevUser)
	if err != nil {
		return nil, fmt.Errorf("lookup dev user: %w", err)
	}
	uid, err := strconv.Atoi(devUser.Uid)
	if err != nil {
		return nil, fmt.Errorf("parse dev uid: %w", err)
	}
	gid, err := strconv.Atoi(devUser.Gid)
	if err != nil {
		return nil, fmt.Errorf("parse dev gid: %w", err)
	}
	cmd.Env = append(os.Environ(),
		"HOME="+devcontainer.DevHome,
		"USER="+devcontainer.DevUser,
		"LOGNAME="+devcontainer.DevUser,
		"TERM=xterm-256color",
	)
	cmd.SysProcAttr = &syscall.SysProcAttr{
		Credential: &syscall.Credential{
			Uid: uint32(uid),
			Gid: uint32(gid),
		},
	}

	ptmx, err := runtime.CreatePTY(cmd)
	if err != nil {
		return nil, fmt.Errorf("start dev shell: %w", err)
	}

	devCmd = cmd
	devPTY = ptmx

	state.SetState(state.StateRunning)
	state.SetError(nil)

	go func() {
		err := cmd.Wait()

		mu.Lock()
		defer mu.Unlock()

		if devPTY != nil {
			_ = devPTY.Close()
		}

		if err != nil {
			state.RecordCrash(err)
		} else {
			state.SetState(state.StateIdle)
		}

		devCmd = nil
		devPTY = nil
	}()

	return devPTY, nil
}

func GetConsolePTY(cfg *state.Config) (*os.File, error) {
	if cfg.ContainerType == "dev" {
		return StartDevShell()
	}

	mu.Lock()
	defer mu.Unlock()

	if serverPTY == nil {
		return nil, fmt.Errorf("server console not available")
	}
	return serverPTY, nil
}

func WriteConsoleInput(cfg *state.Config, input string) error {
	if strings.HasSuffix(input, "\n") {
		input = strings.TrimSuffix(input, "\n")
	}
	payload := []byte(input + "\n")

	if cfg.ContainerType == "dev" {
		mu.Lock()
		defer mu.Unlock()
		if devPTY == nil {
			return fmt.Errorf("dev shell not available")
		}
		return runtime.Write(devPTY, payload)
	}

	mu.Lock()
	defer mu.Unlock()
	if serverPTY == nil {
		return fmt.Errorf("server console not available")
	}
	return runtime.Write(serverPTY, payload)
}

/* --------------------------------------------------------------------------
   Stop Dev Shell
----------------------------------------------------------------------------*/

func StopDevShell() error {
	mu.Lock()
	defer mu.Unlock()

	if devCmd == nil {
		return nil
	}

	if devPTY != nil {
		_ = runtime.Write(devPTY, []byte("exit\n"))
	}

	time.Sleep(1 * time.Second)

	if devCmd.Process != nil {
		_ = devCmd.Process.Kill()
	}

	if devPTY != nil {
		_ = devPTY.Close()
		devPTY = nil
	}

	devCmd = nil
	state.SetState(state.StateIdle)
	return nil
}

func buildCrashInfo(cfg *state.Config, waitErr error, startedAt time.Time) *state.CrashInfo {
	exitCode, signal := extractExitDetails(waitErr)
	uptime := int64(0)
	if !startedAt.IsZero() {
		uptime = int64(time.Since(startedAt).Seconds())
		if uptime < 0 {
			uptime = 0
		}
	}

	logTail := tailLogLines(cfg, 40)

	return &state.CrashInfo{
		Time:           time.Now().UTC(),
		ExitCode:       exitCode,
		Signal:         signal,
		UptimeSeconds:  uptime,
		Classification: classifyCrash(waitErr, exitCode, signal, logTail),
		LogTail:        logTail,
	}
}

func classifyCrash(waitErr error, exitCode, signal int, logTail []string) string {
	errText := strings.ToLower(fmt.Sprint(waitErr))
	for _, line := range logTail {
		lower := strings.ToLower(line)
		switch {
		case strings.Contains(lower, "outofmemoryerror"),
			strings.Contains(lower, "java heap space"),
			strings.Contains(lower, "gc overhead limit exceeded"):
			return "oom"
		case strings.Contains(lower, "mixin"),
			strings.Contains(lower, "failed to load mod"),
			strings.Contains(lower, "loading errors encountered"),
			strings.Contains(lower, "mod resolution encountered"),
			strings.Contains(lower, "plugin"):
			return "mod_or_plugin_error"
		case strings.Contains(lower, "classnotfoundexception"),
			strings.Contains(lower, "noclassdeffounderror"),
			strings.Contains(lower, "could not find or load main class"),
			strings.Contains(lower, "no such file or directory"):
			return "missing_dependency"
		}
	}

	switch {
	case strings.Contains(errText, "ready"):
		return "readiness_failure"
	case signal == 9:
		return "killed"
	case signal != 0:
		return "signaled_exit"
	case exitCode != 0:
		return "nonzero_exit"
	default:
		return "unexpected_exit"
	}
}

func extractExitDetails(err error) (int, int) {
	exitCode := 0
	signal := 0

	if err == nil {
		return exitCode, signal
	}

	var exitErr *exec.ExitError
	if !errors.As(err, &exitErr) {
		return exitCode, signal
	}

	exitCode = exitErr.ExitCode()
	if status, ok := exitErr.Sys().(syscall.WaitStatus); ok && status.Signaled() {
		signal = int(status.Signal())
	}
	return exitCode, signal
}

func tailLogLines(cfg *state.Config, maxLines int) []string {
	buf, err := TailLatestLog(cfg, 64*1024)
	if err != nil || len(buf) == 0 {
		return nil
	}

	rawLines := bytes.Split(buf, []byte{'\n'})
	lines := make([]string, 0, len(rawLines))
	for _, raw := range rawLines {
		line := strings.TrimSpace(string(raw))
		if line == "" {
			continue
		}
		lines = append(lines, line)
	}
	return lastLines(lines, maxLines)
}

func lastLines(lines []string, maxLines int) []string {
	if len(lines) <= maxLines {
		return append([]string(nil), lines...)
	}
	return append([]string(nil), lines[len(lines)-maxLines:]...)
}