|
|
@@ -5,6 +5,7 @@ import (
|
|
|
"errors"
|
|
|
"fmt"
|
|
|
"io"
|
|
|
+ "net"
|
|
|
"os"
|
|
|
"os/exec"
|
|
|
"path/filepath"
|
|
|
@@ -14,33 +15,42 @@ import (
|
|
|
"time"
|
|
|
)
|
|
|
|
|
|
+// 稳定运行的判定阈值：超过它则清零失败计数（退避复位）。
|
|
|
+const healthyUptime = 30 * time.Second
|
|
|
+
|
|
|
// Status 服务状态快照。
|
|
|
type Status struct {
|
|
|
Running bool
|
|
|
PID int
|
|
|
Restarts int
|
|
|
- StartedAt time.Time
|
|
|
+ Uptime time.Duration
|
|
|
LastError string
|
|
|
URL string
|
|
|
Updating bool
|
|
|
+ External bool // 端口被外部实例占用（非本程序启动）
|
|
|
}
|
|
|
|
|
|
-// Supervisor 守护 dsh web：隐藏终端启动，异常退出自动重启。
|
|
|
+// Supervisor 守护 dsh web：隐藏终端启动，异常退出自动重启（指数退避）。
|
|
|
type Supervisor struct {
|
|
|
mu sync.Mutex
|
|
|
+ probeMu sync.Mutex // 串行化入口探测，避免并发重复探测
|
|
|
cfg *Config
|
|
|
log *Logger
|
|
|
- cmd *exec.Cmd
|
|
|
pid int
|
|
|
running bool
|
|
|
desired bool
|
|
|
restarts int
|
|
|
+ failures int
|
|
|
startedAt time.Time
|
|
|
lastError string
|
|
|
updating bool
|
|
|
+ external bool
|
|
|
binCache string
|
|
|
nodeCache string
|
|
|
- webURL string // dsh web 输出的完整 URL（含访问 token）
|
|
|
+ webURL string // dsh web 输出的完整 URL（含访问 token，仅存内存）
|
|
|
+
|
|
|
+ autoMu sync.Mutex
|
|
|
+ autoLoop bool // 定期更新循环是否在运行（保证单例）
|
|
|
}
|
|
|
|
|
|
func NewSupervisor(cfg *Config, log *Logger) *Supervisor {
|
|
|
@@ -55,11 +65,12 @@ func (s *Supervisor) Start() {
|
|
|
return
|
|
|
}
|
|
|
s.desired = true
|
|
|
+ s.failures = 0
|
|
|
s.mu.Unlock()
|
|
|
go s.monitor()
|
|
|
}
|
|
|
|
|
|
-// Stop 停止守护并结束子进程树。
|
|
|
+// Stop 同步停止：结束进程树并等待其真正退出（最多 8 秒）。
|
|
|
func (s *Supervisor) Stop() {
|
|
|
s.mu.Lock()
|
|
|
s.desired = false
|
|
|
@@ -69,39 +80,49 @@ func (s *Supervisor) Stop() {
|
|
|
if pid > 0 {
|
|
|
s.log.Printf("停止 dsh web pid=%d", pid)
|
|
|
_ = killTree(pid)
|
|
|
+ s.waitStopped(pid, 8*time.Second)
|
|
|
}
|
|
|
}
|
|
|
|
|
|
-// Restart 重启服务。
|
|
|
+// Restart 重启并等待端口释放，避免新旧实例争抢端口。
|
|
|
func (s *Supervisor) Restart() {
|
|
|
s.log.Printf("重启 dsh web")
|
|
|
s.Stop()
|
|
|
- time.Sleep(500 * time.Millisecond)
|
|
|
+ s.waitPortFree(5 * time.Second)
|
|
|
s.Start()
|
|
|
}
|
|
|
|
|
|
func (s *Supervisor) Status() Status {
|
|
|
s.mu.Lock()
|
|
|
defer s.mu.Unlock()
|
|
|
- host := s.cfg.WebHost
|
|
|
- if host == "" {
|
|
|
- host = "127.0.0.1"
|
|
|
- }
|
|
|
- url := fmt.Sprintf("http://%s:%d", host, s.cfg.WebPort)
|
|
|
+ url := s.baseURL()
|
|
|
if s.webURL != "" {
|
|
|
- url = s.webURL // 优先使用 dsh web 输出的带 token URL
|
|
|
+ url = s.webURL // 优先使用带 token 的地址
|
|
|
+ }
|
|
|
+ var uptime time.Duration
|
|
|
+ if s.running && !s.startedAt.IsZero() {
|
|
|
+ uptime = time.Since(s.startedAt)
|
|
|
}
|
|
|
return Status{
|
|
|
Running: s.running,
|
|
|
PID: s.pid,
|
|
|
Restarts: s.restarts,
|
|
|
- StartedAt: s.startedAt,
|
|
|
+ Uptime: uptime,
|
|
|
LastError: s.lastError,
|
|
|
URL: url,
|
|
|
Updating: s.updating,
|
|
|
+ External: s.external,
|
|
|
}
|
|
|
}
|
|
|
|
|
|
+func (s *Supervisor) baseURL() string {
|
|
|
+ host := s.cfg.WebHost
|
|
|
+ if host == "" {
|
|
|
+ host = "127.0.0.1"
|
|
|
+ }
|
|
|
+ return fmt.Sprintf("http://%s:%d", host, s.cfg.WebPort)
|
|
|
+}
|
|
|
+
|
|
|
func (s *Supervisor) setError(err error) {
|
|
|
s.mu.Lock()
|
|
|
s.lastError = err.Error()
|
|
|
@@ -109,28 +130,63 @@ func (s *Supervisor) setError(err error) {
|
|
|
s.log.Printf("守护错误: %v", err)
|
|
|
}
|
|
|
|
|
|
-func (s *Supervisor) delay() time.Duration {
|
|
|
- sec := s.cfg.RestartDelaySec
|
|
|
- if sec <= 0 {
|
|
|
- sec = 5
|
|
|
+// backoff 指数退避：base * 2^failures（上限 2 分钟），避免故障时重启风暴。
|
|
|
+func (s *Supervisor) backoff() time.Duration {
|
|
|
+ s.mu.Lock()
|
|
|
+ failures := s.failures
|
|
|
+ base := s.cfg.RestartDelaySec
|
|
|
+ s.mu.Unlock()
|
|
|
+ if base <= 0 {
|
|
|
+ base = 5
|
|
|
}
|
|
|
- return time.Duration(sec) * time.Second
|
|
|
+ d := time.Duration(base) * time.Second
|
|
|
+ for i := 0; i < failures && i < 5; i++ {
|
|
|
+ d *= 2
|
|
|
+ }
|
|
|
+ if d > 2*time.Minute {
|
|
|
+ d = 2 * time.Minute
|
|
|
+ }
|
|
|
+ return d
|
|
|
}
|
|
|
|
|
|
-// monitor 守护循环：启动 -> 等待退出 -> 期望运行时延时重启。
|
|
|
+// portBusyBackoff 端口占用时的温和退避（上限 30 秒）：不是故障，只是等待外部实例释放。
|
|
|
+func (s *Supervisor) portBusyBackoff() time.Duration {
|
|
|
+ d := s.backoff()
|
|
|
+ if d > 30*time.Second {
|
|
|
+ d = 30 * time.Second
|
|
|
+ }
|
|
|
+ return d
|
|
|
+}
|
|
|
+
|
|
|
+// monitor 守护循环：端口预检 -> 启动 -> 等待退出 -> 期望运行时退避重启。
|
|
|
func (s *Supervisor) monitor() {
|
|
|
for {
|
|
|
- s.mu.Lock()
|
|
|
- desired := s.desired
|
|
|
- s.mu.Unlock()
|
|
|
- if !desired {
|
|
|
+ if !s.isDesired() {
|
|
|
return
|
|
|
}
|
|
|
|
|
|
+ // 端口预检：已有实例监听时不再启动，避免端口冲突导致的重启风暴
|
|
|
+ if isPortOpen(s.cfg.WebHost, s.cfg.WebPort) {
|
|
|
+ s.mu.Lock()
|
|
|
+ s.running = false
|
|
|
+ s.pid = 0
|
|
|
+ s.external = true
|
|
|
+ s.lastError = "端口已被占用（可能已有 dsh web 在运行）"
|
|
|
+ s.mu.Unlock()
|
|
|
+ s.log.Printf("端口 %s 已被占用，跳过启动（等待释放）", s.baseURL())
|
|
|
+ s.bumpFailure()
|
|
|
+ s.sleepInterruptible(s.portBusyBackoff())
|
|
|
+ continue
|
|
|
+ }
|
|
|
+ s.mu.Lock()
|
|
|
+ s.external = false
|
|
|
+ s.mu.Unlock()
|
|
|
+
|
|
|
exe, args, err := s.resolveCommand()
|
|
|
if err != nil {
|
|
|
s.setError(err)
|
|
|
- s.sleepInterruptible(s.delay())
|
|
|
+ s.bumpFailure()
|
|
|
+ s.sleepInterruptible(s.backoff())
|
|
|
continue
|
|
|
}
|
|
|
|
|
|
@@ -140,21 +196,23 @@ func (s *Supervisor) monitor() {
|
|
|
stderr, errErr := cmd.StderrPipe()
|
|
|
if errOut != nil || errErr != nil {
|
|
|
s.setError(errors.New("无法创建输出管道"))
|
|
|
- s.sleepInterruptible(s.delay())
|
|
|
+ s.bumpFailure()
|
|
|
+ s.sleepInterruptible(s.backoff())
|
|
|
continue
|
|
|
}
|
|
|
if err := cmd.Start(); err != nil {
|
|
|
s.setError(err)
|
|
|
- s.sleepInterruptible(s.delay())
|
|
|
+ s.bumpFailure()
|
|
|
+ s.sleepInterruptible(s.backoff())
|
|
|
continue
|
|
|
}
|
|
|
|
|
|
pid := cmd.Process.Pid
|
|
|
+ startedAt := time.Now()
|
|
|
s.mu.Lock()
|
|
|
- s.cmd = cmd
|
|
|
s.pid = pid
|
|
|
s.running = true
|
|
|
- s.startedAt = time.Now()
|
|
|
+ s.startedAt = startedAt
|
|
|
s.lastError = ""
|
|
|
s.webURL = ""
|
|
|
s.mu.Unlock()
|
|
|
@@ -164,41 +222,77 @@ func (s *Supervisor) monitor() {
|
|
|
go s.pipeLog("err", stderr)
|
|
|
|
|
|
waitErr := cmd.Wait()
|
|
|
+ ranFor := time.Since(startedAt)
|
|
|
|
|
|
s.mu.Lock()
|
|
|
s.running = false
|
|
|
- s.cmd = nil
|
|
|
s.pid = 0
|
|
|
stillDesired := s.desired
|
|
|
+ if ranFor >= healthyUptime {
|
|
|
+ s.failures = 0 // 稳定运行过，退避复位
|
|
|
+ } else {
|
|
|
+ s.failures++
|
|
|
+ }
|
|
|
if stillDesired {
|
|
|
s.restarts++
|
|
|
}
|
|
|
s.mu.Unlock()
|
|
|
- s.log.Printf("dsh web 退出（err=%v），期望运行=%v", waitErr, stillDesired)
|
|
|
+ s.log.Printf("dsh web 退出（err=%v，运行 %s），期望运行=%v", waitErr, ranFor.Round(time.Second), stillDesired)
|
|
|
|
|
|
if !stillDesired {
|
|
|
return
|
|
|
}
|
|
|
- sleep := s.delay()
|
|
|
- s.log.Printf("%v 后自动重启", sleep)
|
|
|
- s.sleepInterruptible(sleep)
|
|
|
+ delay := s.backoff()
|
|
|
+ s.log.Printf("%s 后自动重启", delay)
|
|
|
+ s.sleepInterruptible(delay)
|
|
|
}
|
|
|
}
|
|
|
|
|
|
+func (s *Supervisor) isDesired() bool {
|
|
|
+ s.mu.Lock()
|
|
|
+ defer s.mu.Unlock()
|
|
|
+ return s.desired
|
|
|
+}
|
|
|
+
|
|
|
+// bumpFailure 增加失败计数（探测/启动阶段失败）。
|
|
|
+func (s *Supervisor) bumpFailure() {
|
|
|
+ s.mu.Lock()
|
|
|
+ s.failures++
|
|
|
+ s.mu.Unlock()
|
|
|
+}
|
|
|
+
|
|
|
func (s *Supervisor) sleepInterruptible(d time.Duration) {
|
|
|
deadline := time.Now().Add(d)
|
|
|
for time.Now().Before(deadline) {
|
|
|
- s.mu.Lock()
|
|
|
- desired := s.desired
|
|
|
- s.mu.Unlock()
|
|
|
- if !desired {
|
|
|
+ if !s.isDesired() {
|
|
|
return
|
|
|
}
|
|
|
time.Sleep(200 * time.Millisecond)
|
|
|
}
|
|
|
}
|
|
|
|
|
|
-// pipeLog 把子进程输出写入日志文件。
|
|
|
+func (s *Supervisor) waitStopped(pid int, timeout time.Duration) {
|
|
|
+ deadline := time.Now().Add(timeout)
|
|
|
+ for time.Now().Before(deadline) {
|
|
|
+ if !processAlive(pid) {
|
|
|
+ return
|
|
|
+ }
|
|
|
+ time.Sleep(100 * time.Millisecond)
|
|
|
+ }
|
|
|
+ s.log.Printf("等待 pid=%d 退出超时（继续）", pid)
|
|
|
+}
|
|
|
+
|
|
|
+func (s *Supervisor) waitPortFree(timeout time.Duration) {
|
|
|
+ deadline := time.Now().Add(timeout)
|
|
|
+ for time.Now().Before(deadline) {
|
|
|
+ if !isPortOpen(s.cfg.WebHost, s.cfg.WebPort) {
|
|
|
+ return
|
|
|
+ }
|
|
|
+ time.Sleep(150 * time.Millisecond)
|
|
|
+ }
|
|
|
+}
|
|
|
+
|
|
|
+// pipeLog 把子进程输出写入日志（token 由 Logger 统一脱敏），并捕获访问地址。
|
|
|
func (s *Supervisor) pipeLog(tag string, r io.ReadCloser) {
|
|
|
if r == nil {
|
|
|
return
|
|
|
@@ -208,25 +302,34 @@ func (s *Supervisor) pipeLog(tag string, r io.ReadCloser) {
|
|
|
for sc.Scan() {
|
|
|
line := sc.Text()
|
|
|
s.log.Printf("[dsh %s] %s", tag, line)
|
|
|
- // 捕获 dsh web 启动时输出的访问地址（含 token），供托盘直接打开
|
|
|
- if idx := strings.Index(line, "http://"); idx >= 0 {
|
|
|
- url := strings.TrimSpace(line[idx:])
|
|
|
- if i := strings.IndexAny(url, " \t"); i > 0 {
|
|
|
- url = url[:i]
|
|
|
- }
|
|
|
- if strings.HasPrefix(url, "http://") && len(url) > len("http://") {
|
|
|
- s.mu.Lock()
|
|
|
- if url != s.webURL {
|
|
|
- s.webURL = url
|
|
|
- s.log.Printf("捕获 Web 访问地址（含 token）")
|
|
|
- }
|
|
|
- s.mu.Unlock()
|
|
|
+ if url := extractHTTPURL(line); url != "" {
|
|
|
+ s.mu.Lock()
|
|
|
+ if url != s.webURL {
|
|
|
+ s.webURL = url
|
|
|
+ s.log.Printf("捕获 Web 访问地址（含 token，仅存内存）")
|
|
|
}
|
|
|
+ s.mu.Unlock()
|
|
|
}
|
|
|
}
|
|
|
}
|
|
|
|
|
|
-// resolveCommand 解析启动命令：优先 node + dsh 入口 JS（无 shell 包装，最小化窗口风险）。
|
|
|
+// extractHTTPURL 从一行输出里取出 http(s) URL（去除尾随分隔符）。
|
|
|
+func extractHTTPURL(line string) string {
|
|
|
+ idx := strings.Index(line, "http")
|
|
|
+ if idx < 0 {
|
|
|
+ return ""
|
|
|
+ }
|
|
|
+ url := strings.TrimSpace(line[idx:])
|
|
|
+ if i := strings.IndexAny(url, " \t"); i > 0 {
|
|
|
+ url = url[:i]
|
|
|
+ }
|
|
|
+ if strings.HasPrefix(url, "http://") || strings.HasPrefix(url, "https://") {
|
|
|
+ return url
|
|
|
+ }
|
|
|
+ return ""
|
|
|
+}
|
|
|
+
|
|
|
+// resolveCommand 解析启动命令：优先 node + dsh 入口 JS（无 shell 包装，避免弹窗与注入面）。
|
|
|
func (s *Supervisor) resolveCommand() (string, []string, error) {
|
|
|
node := s.cfg.NodePath
|
|
|
if node == "" {
|
|
|
@@ -236,17 +339,15 @@ func (s *Supervisor) resolveCommand() (string, []string, error) {
|
|
|
if binJS == "" {
|
|
|
binJS = s.dshBinJS()
|
|
|
}
|
|
|
+ port := strconv.Itoa(s.cfg.WebPort)
|
|
|
if node != "" && binJS != "" {
|
|
|
- args := []string{binJS, "web", "--no-open", "--host", s.cfg.WebHost, "--port", strconv.Itoa(s.cfg.WebPort)}
|
|
|
- return node, args, nil
|
|
|
+ return node, []string{binJS, "web", "--no-open", "--host", s.cfg.WebHost, "--port", port}, nil
|
|
|
}
|
|
|
if p, err := exec.LookPath("dsh.cmd"); err == nil {
|
|
|
- args := []string{"/c", p, "web", "--no-open", "--port", strconv.Itoa(s.cfg.WebPort)}
|
|
|
- return "cmd.exe", args, nil
|
|
|
+ return "cmd.exe", []string{"/c", p, "web", "--no-open", "--port", port}, nil
|
|
|
}
|
|
|
if p, err := exec.LookPath("dsh"); err == nil {
|
|
|
- args := []string{"web", "--no-open", "--port", strconv.Itoa(s.cfg.WebPort)}
|
|
|
- return p, args, nil
|
|
|
+ return p, []string{"web", "--no-open", "--port", port}, nil
|
|
|
}
|
|
|
return "", nil, errors.New("未找到 dsh 入口：请确认已全局安装 @deepseek-ai/dsh，或在配置中指定 nodePath / dshBinJs")
|
|
|
}
|
|
|
@@ -269,8 +370,11 @@ func (s *Supervisor) nodePath() string {
|
|
|
return p
|
|
|
}
|
|
|
|
|
|
-// dshBinJS 探测 dsh 的 bin.js：APPDATA\npm 与 npm root -g。
|
|
|
+// dshBinJS 探测 dsh 的 bin.js（APPDATA\npm 与 npm root -g）；probeMu 保证并发只探测一次。
|
|
|
func (s *Supervisor) dshBinJS() string {
|
|
|
+ s.probeMu.Lock()
|
|
|
+ defer s.probeMu.Unlock()
|
|
|
+
|
|
|
s.mu.Lock()
|
|
|
if s.binCache != "" {
|
|
|
v := s.binCache
|
|
|
@@ -298,6 +402,13 @@ func (s *Supervisor) dshBinJS() string {
|
|
|
return ""
|
|
|
}
|
|
|
|
|
|
+// invalidateProbe 使入口缓存失效（更新后调用）。
|
|
|
+func (s *Supervisor) invalidateProbe() {
|
|
|
+ s.mu.Lock()
|
|
|
+ s.binCache = ""
|
|
|
+ s.mu.Unlock()
|
|
|
+}
|
|
|
+
|
|
|
// installedVersion 读取已安装 dsh 版本。
|
|
|
func (s *Supervisor) installedVersion() string {
|
|
|
binJS := s.cfg.DshBinJS
|
|
|
@@ -307,11 +418,23 @@ func (s *Supervisor) installedVersion() string {
|
|
|
if binJS == "" {
|
|
|
return ""
|
|
|
}
|
|
|
- pkg := filepath.Join(filepath.Dir(filepath.Dir(binJS)), "package.json")
|
|
|
- return readPackageVersion(pkg)
|
|
|
+ return readPackageVersion(filepath.Join(filepath.Dir(filepath.Dir(binJS)), "package.json"))
|
|
|
}
|
|
|
|
|
|
func fileExists(p string) bool {
|
|
|
st, err := os.Stat(p)
|
|
|
return err == nil && !st.IsDir()
|
|
|
}
|
|
|
+
|
|
|
+// isPortOpen 检测端口是否有监听者。
|
|
|
+func isPortOpen(host string, port int) bool {
|
|
|
+ if host == "" {
|
|
|
+ host = "127.0.0.1"
|
|
|
+ }
|
|
|
+ conn, err := net.DialTimeout("tcp", fmt.Sprintf("%s:%d", host, port), 1200*time.Millisecond)
|
|
|
+ if err != nil {
|
|
|
+ return false
|
|
|
+ }
|
|
|
+ _ = conn.Close()
|
|
|
+ return true
|
|
|
+}
|
