feat: M3 mosh 挂起→恢复<3s(唤醒脉冲 + 会话状态机三相)

- SessionMachine 加 mosh 三相(moshActive/moshParked/moshResuming)+看门狗兜底,
  mosh 接管即拆 SSH、后台冻结不重建、前台唤醒脉冲等 SSP 续(+5 单测)
- bridge.go: Node.WakeUp()(tsnet InjectEvent+MagicSock.Rebind/ReSTUN) +
  MoshRelay.Rebind()(同端口重开 loopback、泵按 socket 生成容错)
- iosclient.cc: g_mosh_last_heard 探针(导出 mosh_last_heard_ms) + SIGCONT 全屏重绘
- MoshSession: nudge()(SIGCONT) + lastHeardMs();SSHTerminalModel 前台脉冲三连
  + healthy 轮询 + beginBackgroundTask;URL scheme 供无头前台唤醒
- 验证(vohive-vm over tsnet):15s 冻结→前台 wake pulse→1.3s recovered,<3s

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
kid
2026-07-24 17:04:33 +08:00
parent 32520f09e0
commit 866df1a323
23 changed files with 489 additions and 96 deletions

View File

@@ -22,6 +22,13 @@ public struct SessionMachine: Sendable, Equatable {
case failed(reason: String)
/// 用户主动关闭。
case closed
// —— M3 mosh 会话相(mosh 接管后,SSH 已拆,生命周期由 mosh SSP 承载)——
/// mosh 已接管并活跃(底层 SSH 已拆除;transportClosed 不再影响)。
case moshActive
/// 进入后台,mosh 冻结(SSP 会在前台自动续传,不重建)。
case moshParked
/// 前台恢复中:已发唤醒脉冲,等 mosh SSP resync(记录第几次尝试,超时兜底)。
case moshResuming(attempt: Int)
}
public enum Event: Sendable, Equatable {
@@ -35,6 +42,11 @@ public struct SessionMachine: Sendable, Equatable {
case enteredForeground
case closeRequested
case retryRequested // 用户手动重试(从 failed 复位)
// —— M3 mosh ——
case moshEstablished // mosh 接管会话成功(此后拆 SSH)
case moshHealthy // 前台恢复后收到服务器合法包(SSP 已续上)
case moshExited(rc: Int) // mosh 主循环退出(会话结束)
case resumeWatchdogFired // 恢复看门狗到点(本次唤醒脉冲未见 healthy)
}
public enum Effect: Sendable, Equatable {
@@ -43,6 +55,11 @@ public struct SessionMachine: Sendable, Equatable {
case cancelReconnectTimer
case teardownTransport
case notify(String) // UI 提示
// —— M3 mosh ——
case nudgeResume // 打唤醒脉冲:tsnet WakeUp + relay Rebind + mosh SIGCONT
case scheduleResumeWatchdog(ms: Int) // 排程恢复看门狗
case cancelResumeWatchdog
case teardownMosh // 拆除 mosh 会话(relay + 线程)
}
public private(set) var phase: Phase = .idle
@@ -52,11 +69,18 @@ public struct SessionMachine: Sendable, Equatable {
/// 退避基数与上限(毫秒)。
private let baseDelayMS: Int
private let maxDelayMS: Int
/// M3:单次唤醒脉冲后等 mosh SSP 续上的看门狗窗口;超时则再脉冲一次。
public let resumeWatchdogMS: Int
/// M3:最多打几次唤醒脉冲,仍未 healthy 则兜底全量重建。
public let maxResumePulses: Int
public init(maxReconnectAttempts: Int = 6, baseDelayMS: Int = 500, maxDelayMS: Int = 16000) {
public init(maxReconnectAttempts: Int = 6, baseDelayMS: Int = 500, maxDelayMS: Int = 16000,
resumeWatchdogMS: Int = 3000, maxResumePulses: Int = 2) {
self.maxReconnectAttempts = maxReconnectAttempts
self.baseDelayMS = baseDelayMS
self.maxDelayMS = maxDelayMS
self.resumeWatchdogMS = resumeWatchdogMS
self.maxResumePulses = maxResumePulses
}
/// 指数退避:base * 2^(attempt-1),封顶 maxDelayMS。
@@ -135,10 +159,60 @@ public struct SessionMachine: Sendable, Equatable {
phase = .backgroundParked
return [.cancelReconnectTimer]
// —— M3 mosh 会话相 ——
// mosh 接管成功(引导完成):拆掉底层 SSH,进入 mosh 自持会话。
case (.connected, .moshEstablished), (.reconnecting, .moshEstablished):
phase = .moshActive
return [.teardownTransport, .notify("mosh 已接管")]
// mosh 活跃期间底层 SSH 残连断开:忽略(mosh 不依赖 SSH)。
case (.moshActive, .transportClosed):
return []
// mosh 活跃 → 后台:冻结,不重建。
case (.moshActive, .enteredBackground):
phase = .moshParked
return []
// mosh 后台 → 前台:打唤醒脉冲 + 排看门狗,等 SSP 续上。
case (.moshParked, .enteredForeground):
phase = .moshResuming(attempt: 1)
return [.nudgeResume, .scheduleResumeWatchdog(ms: resumeWatchdogMS), .notify("恢复中…")]
// mosh 后台期间的杂项(底层断/重复进后台):保持 parked。
case (.moshParked, .transportClosed), (.moshParked, .enteredBackground):
return []
// 恢复中收到服务器合法包(SSP 已续):回到活跃。
case (.moshResuming, .moshHealthy):
phase = .moshActive
return [.cancelResumeWatchdog, .notify("已恢复")]
// 恢复看门狗到点:未 healthy → 再脉冲;达上限 → 兜底全量重建。
case (.moshResuming(let attempt), .resumeWatchdogFired):
if attempt < maxResumePulses {
phase = .moshResuming(attempt: attempt + 1)
return [.nudgeResume, .scheduleResumeWatchdog(ms: resumeWatchdogMS)]
}
phase = .reconnecting(attempt: 1)
return [.teardownMosh, .startConnect, .notify("mosh 恢复失败,重建会话…")]
// 恢复中又进后台:取消看门狗,回 parked。
case (.moshResuming, .enteredBackground):
phase = .moshParked
return [.cancelResumeWatchdog]
// mosh 主循环退出(会话结束):终态失败,可手动重试。
case (.moshActive, .moshExited(let rc)),
(.moshResuming, .moshExited(let rc)),
(.moshParked, .moshExited(let rc)):
phase = .failed(reason: "mosh 已结束(rc=\(rc))")
return [.cancelResumeWatchdog, .teardownMosh, .notify("mosh 会话结束")]
// —— 关闭 ——
case (_, .closeRequested):
phase = .closed
return [.cancelReconnectTimer, .teardownTransport]
return [.cancelReconnectTimer, .cancelResumeWatchdog, .teardownTransport, .teardownMosh]
// 其它组合不改变状态
default:

View File

@@ -125,4 +125,78 @@ struct SessionMachineTests {
#expect(m2.reduce(.retryRequested) == [.startConnect])
#expect(m2.phase == .connecting)
}
// MARK: - M3 mosh 生命周期
private func connectedThenMosh() -> SessionMachine {
var m = SessionMachine()
_ = m.reduce(.connectRequested)
_ = m.reduce(.established)
return m
}
@Test("mosh 接管拆 SSH,之后忽略 transportClosed")
func moshTakeover() {
var m = connectedThenMosh()
let e = m.reduce(.moshEstablished)
#expect(m.phase == .moshActive)
#expect(e.contains(.teardownTransport))
// 底层 SSH 残连断开,不影响 mosh
#expect(m.reduce(.transportClosed(reason: "ssh eof")) == [])
#expect(m.phase == .moshActive)
}
@Test("mosh 后台→前台:唤醒脉冲 + 看门狗,收到 healthy 即恢复")
func moshSuspendResumeHealthy() {
var m = connectedThenMosh()
_ = m.reduce(.moshEstablished)
#expect(m.reduce(.enteredBackground) == [])
#expect(m.phase == .moshParked)
let e = m.reduce(.enteredForeground)
#expect(m.phase == .moshResuming(attempt: 1))
#expect(e.contains(.nudgeResume))
#expect(e.contains(.scheduleResumeWatchdog(ms: 3000)))
// SSP 续上
let e2 = m.reduce(.moshHealthy)
#expect(m.phase == .moshActive)
#expect(e2.contains(.cancelResumeWatchdog))
}
@Test("恢复看门狗:一次超时再脉冲,两次仍失败则兜底全量重建")
func moshResumeWatchdogFallback() {
var m = SessionMachine(resumeWatchdogMS: 3000, maxResumePulses: 2)
_ = m.reduce(.connectRequested)
_ = m.reduce(.established)
_ = m.reduce(.moshEstablished)
_ = m.reduce(.enteredBackground)
_ = m.reduce(.enteredForeground) // moshResuming(1)
// 第一次看门狗到点 → 再脉冲,进 moshResuming(2)
let e1 = m.reduce(.resumeWatchdogFired)
#expect(m.phase == .moshResuming(attempt: 2))
#expect(e1.contains(.nudgeResume))
// 第二次看门狗到点 → 兜底全量重建
let e2 = m.reduce(.resumeWatchdogFired)
#expect(m.phase == .reconnecting(attempt: 1))
#expect(e2.contains(.teardownMosh))
#expect(e2.contains(.startConnect))
}
@Test("mosh 主循环退出转 failed 并拆除")
func moshExited() {
var m = connectedThenMosh()
_ = m.reduce(.moshEstablished)
let e = m.reduce(.moshExited(rc: 1))
if case .failed = m.phase {} else { Issue.record("期望 failed,实际 \(m.phase)") }
#expect(e.contains(.teardownMosh))
}
@Test("mosh 恢复中进后台取消看门狗")
func moshResumingToBackground() {
var m = connectedThenMosh()
_ = m.reduce(.moshEstablished)
_ = m.reduce(.enteredBackground)
_ = m.reduce(.enteredForeground) // moshResuming(1)
#expect(m.reduce(.enteredBackground) == [.cancelResumeWatchdog])
#expect(m.phase == .moshParked)
}
}