Files
HomeAgent/internal/sdk/plugin.go
JianFeeeee 02cc74ce11 fix(proc): 子进程崩溃自愈 + 集中台账 + 注册面摘除
根因:子进程插件被 kill 后,内核只发了一个无人订阅的事件,
工具/stage handler/IO 通道全留在注册表里指向死进程,
模型继续调用只吃 ErrProcessExited,没有任何路径把插件拉回来。

## 四层修复

### 1. 专职 waitLoop(进程收割)
- 每个子进程配一根 waitLoop goroutine,是 cmd.Wait() 的唯一调用点
- 不再依赖 stdout EOF 判定死亡(孙子进程继承 stdout 时 EOF 永不到来)
- 手工 os.Pipe 替代 cmd.StdinPipe/StdoutPipe,避免 waitLoop 与
  os/exec 的内部关闭竞争
- host.go: Host.Supervisor(),Host.Close() 先 StopAll 再拆段

### 2. 集中台账 Supervisor
- proc/supervisor.go: 插件 Spawn 握手成功即 track,进程退出即 untrack
- StopAll: 并发发 plugin.stop 走优雅路径,到期仍在的一律 Kill
- 关停后才完成握手的进程被立即结束,不会活过内核
- 消除「孤儿进程持共享段映射 → SIGBUS」的隐患

### 3. 注册面摘除(detachPlugin)
- 新增 StageHost.UnregisterPluginStages:摘除指定插件的全部 stage handler
- 新增 Registry.pluginChannels 台账:记录每个插件注册的 IO 通道
- 三条路径统一走 detachPlugin:Disable / ReloadOne / RemovePlugin
- StopAndUnload 漏了 IO 通道也一并补上

### 4. 自动重启
- onProcCrash 从「只发事件」改为「摘注册面 → 从注册表移除 → 异步排重启」
- scheduleProcRestart: 窗口 5 分钟内最多 3 次,线性退避 1s/2s/3s
- 超限停手留日志;重启前复核是否已被 Disable 或被其他路径加载
- 崩溃计数窗口过期自动归零

### 5. 主动停止 vs 崩溃的区分
- proc.Plugin 新增 stopping 标志:Stop()/Close() 里 Set(true)
- handleExit 读 stopping 标志,主动停止不上报 onCrash
- 防止重载/禁用/卸载被误判为崩溃触发多余重启

### 6. Linux Pdeathsig 兜底
- procattr_linux.go: SysProcAttr.Pdeathsig = SIGKILL
- 兜 homed 自身被 SIGKILL/OOM 时子进程变孤儿的场景
- macOS/Windows 无等价物,空实现

### 7. pluginmgr 升级
- PluginManager 接口新增 PluginRuntime / ListPluginRuntimes
- plugin_list 输出运行态:loaded / alive / pid / crash_count / channel
- 新增 plugin_status: 全量运行期快照 + dead/unhealthy 汇总
- 新增 plugin_restart: 无条件重启单个插件(plgreload 不动未改二进制的插件)

### 测试
- process_test.go: 3 例(grandchild stdout 感知 / Supervisor track-untrack /
  StopAll 无孤儿)
- crash_recovery_test.go: 8 例(detach 三项齐全 / 通道重注册 / 崩溃不阻塞 /
  退避阈值 / 窗口过期 / 关停中跳过 / PluginRuntime 通道识别)
- stages_plugin_test.go: 4 例(stage 按插件摘除 / 空 stage 清理 / 空名 no-op /
  工具+stage 双摘后可重新注册同名)
2026-09-03 12:37:58 +08:00

399 lines
12 KiB
Go
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

package sdk
import (
"log"
"sync"
agentIO "gitcode.com/JianFeeeee/HomeAgent/internal/agent/io"
"gitcode.com/JianFeeeee/HomeAgent/internal/events"
pubsdk "gitcode.com/JianFeeeee/homeagent-sdk/sdk"
)
// SDKVersion 是对外 SDK 版本号,与核心 meta.Version 保持一致。
var SDKVersion = pubsdk.SDKVersion
type Plugin interface {
Name() string
Start(sdk *PluginSDK) error
Stop() error
}
type ToolHandler = pubsdk.ToolHandler
type StageHandler = pubsdk.StageHandler
type Stage = pubsdk.Stage
const (
StageOnInput = pubsdk.StageOnInput
StagePreAction = pubsdk.StagePreAction
StagePostAction = pubsdk.StagePostAction
StageBeforeToolcall = pubsdk.StageBeforeToolcall
StageAfterToolcall = pubsdk.StageAfterToolcall
StageBeforeOutput = pubsdk.StageBeforeOutput
StageAfterOutput = pubsdk.StageAfterOutput
)
type StageContext = pubsdk.StageContext
type MemItem = pubsdk.MemItem
type ToolCall = pubsdk.ToolCall
type ToolResult = pubsdk.ToolResult
type ToolDef = pubsdk.ToolDef
type IOInjector = pubsdk.IOInjector
type ToolRegistrar = pubsdk.ToolRegistrar
type StageRegistrar = pubsdk.StageRegistrar
type StageScope = pubsdk.StageScope
const (
StageScopeGlobal = pubsdk.StageScopeGlobal
StageScopeOwnTools = pubsdk.StageScopeOwnTools
)
type APIRegistrar = pubsdk.APIRegistrar
type OutputChannelRegistrar = pubsdk.OutputChannelRegistrar
type InputChannelRegistrar = pubsdk.InputChannelRegistrar
type ChannelDef = pubsdk.ChannelDef
type DisabledPluginInfo struct {
Name string `json:"name"`
DisabledAt string `json:"disabled_at"`
DisabledBy string `json:"disabled_by"`
}
// PluginMeta is the display-name metadata for a plugin (from plg.json / RegisterPluginMeta).
type PluginMeta struct {
NameZh string `json:"name_zh"`
NameEn string `json:"name_en"`
}
// PluginRuntimeInfo 是插件的**运行期**状态,与 plugin.json 里的静态元数据相对。
//
// 为何需要:子进程插件的进程可能已经死了而注册表里还有条目(或反过来,
// 崩溃摘除后注册表已无条目但目录还在)。此前 plugin_list / GET /plugins
// 只读 plugin.json无论插件死活都返回同一份内容——WebUI 与模型都看不出
// 「已安装」与「正在运行」的区别,插件被 kill 后只表现为工具静默失败。
type PluginRuntimeInfo struct {
Name string `json:"name"`
// Loaded 表示注册表中存在该插件实例。
Loaded bool `json:"loaded"`
// Disabled 表示插件被显式禁用(不该运行)。
Disabled bool `json:"disabled"`
// Builtin 表示编译期内置插件(无独立进程)。
Builtin bool `json:"builtin"`
// Channel 是加载通道proc子进程/ lua / builtin。
Channel string `json:"channel"`
// PID 是子进程插件的进程号;非子进程或已退出为 0。
PID int `json:"pid"`
// Alive 表示子进程仍存活;非子进程插件与 Loaded 同值。
Alive bool `json:"alive"`
// CrashCount 是最近窗口内的崩溃次数0 表示健康)。
CrashCount int `json:"crash_count"`
// AutoRestart 表示崩溃后内核是否会自动拉起。
AutoRestart bool `json:"auto_restart"`
// Tools 是该插件当前注册在内核里的工具名。
Tools []string `json:"tools,omitempty"`
}
type PluginManager interface {
ListLoadedPlugins() []string
ListDisabledPlugins() []DisabledPluginInfo
IsPluginDisabled(name string) bool
// IsBuiltinPlugin 判断插件是否为内置插件编译期工厂init() 自注册)。
// 内置插件只能禁用/启用,不能卸载。
IsBuiltinPlugin(name string) bool
DisablePlugin(name, by string) error
EnablePlugin(name string) error
// RemovePlugin 卸载插件先停止stop handlers + Stop再执行插件注册的
// onRemove 回调RegisterOnRemoveHandler最后从注册表移除并清理配置表。
// 目录删除由调用方负责。
RemovePlugin(name string) error
// StopAndUnload 停止并从注册表移除插件但保留配置表,供更新/升级流程使用:
// 换产物不动配置,重装后配置原样生效。不触发 onRemove 回调。
StopAndUnload(name string) error
ReloadPlugins() (string, error)
// ReloadOne 重载单个插件(停止后重新加载,处理 dlclose/dynamic 句柄)。
ReloadOne(name string) error
PluginMetas() map[string]PluginMeta
PluginDir() string
// PluginRuntime 返回单个插件的运行期状态(进程存活 / PID / 崩溃计数)。
// 未安装的插件返回零值 + false。
PluginRuntime(name string) (PluginRuntimeInfo, bool)
// ListPluginRuntimes 返回全部已加载插件的运行期状态。
ListPluginRuntimes() []PluginRuntimeInfo
}
type PluginSDK struct {
*pubsdk.PluginSDK
settings SettingsAPI
memory MemoryAPI
textMem TextMemoryAPI
docMem DocMemoryAPI
know KnowledgeAPI
llm LLMAPI
iom *agentIO.IOManager
eventBus *events.Bus
logger *log.Logger
pluginMgr PluginManager
status StatusAPI
supervisor SupervisorAPI
adapter AdapterAPI
tracker TrackerAPI
config ConfigAPI
tool ToolAPI
indexer IndexerAPI
selftestMu sync.Mutex
selftest *VirtualInstance
}
func (s *PluginSDK) PluginMgr() PluginManager { return s.pluginMgr }
// 以下访问器遮蔽公共 SDK 的同名方法,返回内置插件可用的全量接口。
func (s *PluginSDK) Settings() SettingsAPI { return s.settings }
func (s *PluginSDK) Memory() MemoryAPI { return s.memory }
func (s *PluginSDK) TextMemory() TextMemoryAPI { return s.textMem }
func (s *PluginSDK) DocMemory() DocMemoryAPI { return s.docMem }
func (s *PluginSDK) Knowledge() KnowledgeAPI { return s.know }
func (s *PluginSDK) LLM() LLMAPI { return s.llm }
// ioAdapter 桥接 IOManager 到公共 SDK 的 IOInjector 接口,
// 确保外部插件通过 s.InjectText() 等方法的调用能被路由到内核 IO 层。
type ioAdapter struct{ iom *agentIO.IOManager }
func (a ioAdapter) InjectInterruptText(source, channel, text string) {
if a.iom != nil {
a.iom.InjectInterrupt(source, channel, map[string]interface{}{"type": "text", "content": text})
}
}
// InjectInputSync 同步注入输入并等待回复(阻塞直至 agent 处理完成),返回回复文本。
func (a ioAdapter) InjectInputSync(source, channel, text string) string {
if a.iom == nil {
return ""
}
out := a.iom.InjectInputSyncTo(source, channel, "text", map[string]interface{}{
"content": text,
})
if out == nil {
return ""
}
reply, _ := out.Payload["content"].(string)
return reply
}
func (a ioAdapter) InjectText(source, channel, text string) {
if a.iom != nil {
a.iom.InjectInputTo(source, channel, "text", map[string]interface{}{"content": text})
}
}
func (a ioAdapter) InjectTextNoMemory(source, channel, text string) {
if a.iom != nil {
a.iom.InjectInputTo(source, channel, "text", map[string]interface{}{"content": text, "no_memory": true})
}
}
// SDKConfig holds all dependencies for creating a PluginSDK.
type SDKConfig struct {
IOManager *agentIO.IOManager
EventBus *events.Bus
Memory MemoryAPI
TextMemory TextMemoryAPI
DocMemory DocMemoryAPI
Knowledge KnowledgeAPI
LLM LLMAPI
Settings SettingsAPI
RegTool ToolRegistrar
RegStage StageRegistrar
RegAPI APIRegistrar
RegOutput OutputChannelRegistrar
RegInput InputChannelRegistrar
PluginMgr PluginManager
Status StatusAPI
Supervisor SupervisorAPI
Adapter AdapterAPI
Tracker TrackerAPI
Config ConfigAPI
Tool ToolAPI
Indexer IndexerAPI
}
func New(name string, cfg SDKConfig) *PluginSDK {
base := pubsdk.New(name, cfg.Settings, cfg.RegTool, cfg.RegStage, cfg.RegAPI, cfg.RegOutput)
if cfg.IOManager != nil {
base.SetIOInjector(ioAdapter{iom: cfg.IOManager})
}
if cfg.RegInput != nil {
base.SetInputChannelRegistrar(cfg.RegInput)
}
base.SetMemoryAPI(cfg.Memory)
base.SetTextMemoryAPI(cfg.TextMemory)
base.SetDocMemoryAPI(cfg.DocMemory)
base.SetKnowledgeAPI(cfg.Knowledge)
base.SetLLMAPI(cfg.LLM)
return &PluginSDK{
PluginSDK: base,
settings: cfg.Settings,
memory: cfg.Memory,
textMem: cfg.TextMemory,
docMem: cfg.DocMemory,
know: cfg.Knowledge,
llm: cfg.LLM,
iom: cfg.IOManager,
eventBus: cfg.EventBus,
logger: log.Default(),
pluginMgr: cfg.PluginMgr,
status: cfg.Status,
supervisor: cfg.Supervisor,
adapter: cfg.Adapter,
tracker: cfg.Tracker,
config: cfg.Config,
tool: cfg.Tool,
indexer: cfg.Indexer,
}
}
// Selftest 返回一个隔离的虚拟自检实例healthcheck 等内置插件用),
// 完全独立于生产存储,不产生任何污染。首次调用创建,复用已存在实例;
// 每轮自检前调用 SelftestReset 重建以清空上轮测试数据。
func (s *PluginSDK) Selftest(scope string) (*VirtualInstance, error) {
s.selftestMu.Lock()
defer s.selftestMu.Unlock()
if s.selftest == nil {
vi, err := NewVirtualInstance(scope)
if err != nil {
return nil, err
}
s.selftest = vi
}
return s.selftest, nil
}
// SelftestReset 清理并重建隔离自检实例,用于每轮健康检查前重置状态。
func (s *PluginSDK) SelftestReset(scope string) error {
s.selftestMu.Lock()
defer s.selftestMu.Unlock()
if s.selftest == nil {
vi, err := NewVirtualInstance(scope)
if err != nil {
return err
}
s.selftest = vi
return nil
}
vi, err := s.selftest.Reset(scope)
if err != nil {
return err
}
s.selftest = vi
return nil
}
func (s *PluginSDK) Status() StatusAPI { return s.status }
func (s *PluginSDK) Supervisor() SupervisorAPI { return s.supervisor }
func (s *PluginSDK) Adapter() AdapterAPI { return s.adapter }
func (s *PluginSDK) Tracker() TrackerAPI { return s.tracker }
func (s *PluginSDK) Config() ConfigAPI { return s.config }
func (s *PluginSDK) Tool() ToolAPI { return s.tool }
func (s *PluginSDK) Indexer() IndexerAPI { return s.indexer }
func (s *PluginSDK) InjectInput(source, channel, eventType string, payload map[string]interface{}) {
if s.iom != nil {
s.iom.InjectInputTo(source, channel, eventType, payload)
}
}
func (s *PluginSDK) InjectInputSync(source, channel, eventType string, payload map[string]interface{}) *agentIO.OutputEvent {
if s.iom != nil {
return s.iom.InjectInputSyncTo(source, channel, eventType, payload)
}
return nil
}
func (s *PluginSDK) InjectInterrupt(source, channel, eventType string, payload map[string]interface{}) {
if s.iom != nil {
if payload == nil {
payload = map[string]interface{}{}
}
payload["type"] = eventType
s.iom.InjectInterrupt(source, channel, payload)
}
}
func (s *PluginSDK) InjectTextSync(source, channel, text string) *agentIO.OutputEvent {
return s.InjectInputSync(source, channel, "text", map[string]interface{}{"content": text})
}
func (s *PluginSDK) InjectTextSyncNoMemory(source, channel, text string) *agentIO.OutputEvent {
return s.InjectInputSync(source, channel, "text", map[string]interface{}{"content": text, "no_memory": true})
}
func (s *PluginSDK) OutputChan() <-chan *agentIO.OutputEvent {
if s.iom != nil {
return s.iom.OutputChan()
}
return nil
}
func (s *PluginSDK) RegisterChannel(name string, dev agentIO.Device) error {
if s.iom != nil {
return s.iom.RegisterDevice(dev)
}
return nil
}
func (s *PluginSDK) UnregisterChannel(name string) {
if s.iom != nil {
s.iom.UnregisterDevice(name)
}
}
func (s *PluginSDK) ListChannels() []agentIO.ChannelInfo {
if s.iom != nil {
return s.iom.ListChannels()
}
return nil
}
func (s *PluginSDK) Publish(evt *events.Event) {
if s.eventBus != nil {
s.eventBus.Publish(evt)
}
}
func (s *PluginSDK) Subscribe(eventType events.EventType, handler events.Handler) func() {
if s.eventBus != nil {
return s.eventBus.Subscribe(eventType, handler)
}
return func() {}
}
// SetToolBlocks 桥接到 IOManager插件工具注入多模态块process.go 消费。
func (a ioAdapter) SetToolBlocks(blocks []pubsdk.ContentBlock) {
if a.iom == nil {
return
}
ifaces := make([]interface{}, len(blocks))
for i, b := range blocks {
ifaces[i] = b
}
a.iom.SetToolBlocks(ifaces)
}
// SetToolBlocks 注入多模态内容块(图片/音频),内核在下一条 tool message
// 的 content 数组里带上这些块,让模型在后续轮次看到图/听到音频。
func (s *PluginSDK) SetToolBlocks(blocks []pubsdk.ContentBlock) {
if s.iom != nil {
ifaces := make([]interface{}, len(blocks))
for i, b := range blocks {
ifaces[i] = b
}
s.iom.SetToolBlocks(ifaces)
}
}