feat(acp): supervisor skeleton + AcpConfig wiring

2026-05-07 12:12:05 +08:00
parent dee20bb618
commit bd95fd33d6
3 changed files with 176 additions and 0 deletions
@@ -0,0 +1,126 @@
+// supervisor.go 管理 ACP agent 子进程池。每 session 一个 Process，spawn / kill /
+// monitor 三类操作通过 mu 保护 procs map。
+//
+// 生命周期 invariants（spec §6.8）：
+//   - procs 写锁仅在 spawn / Kill / monitorProcess delete 时持有
+//   - done channel 严格 close-once（monitor 唯一持有）
+//   - KilledByUs atomic 防 race（Kill / monitor 并发）
+//   - onExit 不能再 Lock supervisor.mu（避免死锁）
+package acp
+
+import (
+	"io"
+	"log/slog"
+	"os/exec"
+	"sync"
+	"sync/atomic"
+	"time"
+
+	"github.com/google/uuid"
+
+	"github.com/yan1h/agent-coding-workflow/internal/audit"
+	"github.com/yan1h/agent-coding-workflow/internal/infra/crypto"
+	"github.com/yan1h/agent-coding-workflow/internal/infra/notify"
+	"github.com/yan1h/agent-coding-workflow/internal/workspace"
+)
+
+// SupervisorConfig 是从 config.AcpConfig 派生的 supervisor 子集。
+type SupervisorConfig struct {
+	SpawnTimeout       time.Duration
+	KillGrace          time.Duration
+	StderrBufferLines  int
+	StderrTailBytes    int
+	EventMaxPayload    int
+	EventTruncateField int
+	ShutdownGrace      time.Duration
+}
+
+// Supervisor 是 ACP 子进程总管。
+type Supervisor struct {
+	mu    sync.RWMutex
+	procs map[uuid.UUID]*Process
+
+	repo   Repository
+	audit  audit.Recorder
+	notify *notify.Dispatcher
+	wtSvc  workspace.WorktreeService
+	wsRepo workspace.Repository
+	crypto *crypto.Encryptor
+	cfg    SupervisorConfig
+	log    *slog.Logger
+}
+
+// NewSupervisor 构造空 Supervisor。Start 不需要 — supervisor 是被动的（spawn 时
+// 起 goroutine，无主循环）。Stop 在 ShutdownAll 中实现。
+func NewSupervisor(repo Repository, rec audit.Recorder, disp *notify.Dispatcher,
+	wtSvc workspace.WorktreeService, wsRepo workspace.Repository,
+	enc *crypto.Encryptor, cfg SupervisorConfig, log *slog.Logger) *Supervisor {
+	if log == nil {
+		log = slog.Default()
+	}
+	return &Supervisor{
+		procs:  map[uuid.UUID]*Process{},
+		repo:   repo,
+		audit:  rec,
+		notify: disp,
+		wtSvc:  wtSvc,
+		wsRepo: wsRepo,
+		crypto: enc,
+		cfg:    cfg,
+		log:    log,
+	}
+}
+
+// Process 是单个 agent 子进程的运行时句柄。
+type Process struct {
+	SessionID   uuid.UUID
+	WorkspaceID uuid.UUID
+	UserID      uuid.UUID
+	IsMain      bool       // 主 worktree 占用 → release 走不同路径
+	WorktreeID  *uuid.UUID // 子 worktree 时填，用于 release
+
+	Cmd    *exec.Cmd
+	Stdin  io.WriteCloser
+	Stdout io.ReadCloser
+	Stderr io.ReadCloser
+
+	StderrBuf  *stderrRing  // 排障 ring buffer
+	StartedAt  time.Time
+	KilledByUs atomic.Bool  // true → exited，false → crashed
+	done       chan struct{} // monitor 退出时 close
+}
+
+// stderrRing 是固定行数 ring buffer，monitor 退出时取 tail 写 last_error。
+type stderrRing struct {
+	mu    sync.Mutex
+	lines []string
+	cap   int
+}
+
+func newStderrRing(cap int) *stderrRing { return &stderrRing{cap: cap} }
+
+func (r *stderrRing) Append(line string) {
+	r.mu.Lock()
+	defer r.mu.Unlock()
+	if len(r.lines) >= r.cap {
+		r.lines = r.lines[1:]
+	}
+	r.lines = append(r.lines, line)
+}
+
+func (r *stderrRing) Tail(maxBytes int) string {
+	r.mu.Lock()
+	defer r.mu.Unlock()
+	out := ""
+	for i := len(r.lines) - 1; i >= 0 && len(out) < maxBytes; i-- {
+		out = r.lines[i] + "\n" + out
+	}
+	if len(out) > maxBytes {
+		out = out[len(out)-maxBytes:]
+	}
+	return out
+}
+
+// Relay field + GetRelay method added in Task D7 when Relay type is wired.
+
+// Spawn / Kill / ShutdownAll / monitorProcess / handshake are implemented in subsequent tasks (E2).
@@ -41,6 +41,7 @@ type Config struct {
 	Storage   Storage         `mapstructure:"storage"`
 	Jobs      JobsConfig      `mapstructure:"jobs"`
 	Notify    NotifyConfig    `mapstructure:"notify"`
+	Acp       AcpConfig       `mapstructure:"acp"`
 }

 // HTTPConfig 描述 HTTP 服务监听参数。
@@ -165,6 +166,24 @@ type WebhookCfg struct {
 	Topics  []string      `mapstructure:"topics"`
 }

+// AcpConfig 控制 ACP 模块的并发限流、子进程超时、WS 心跳与事件保留。
+type AcpConfig struct {
+	Enabled            bool          `mapstructure:"enabled"`
+	MaxActiveGlobal    int           `mapstructure:"max_active_global"`
+	MaxActivePerUser   int           `mapstructure:"max_active_per_user"`
+	SpawnTimeout       time.Duration `mapstructure:"spawn_timeout"`
+	KillGrace          time.Duration `mapstructure:"kill_grace"`
+	StderrBufferLines  int           `mapstructure:"stderr_buffer_lines"`
+	StderrTailBytes    int           `mapstructure:"stderr_tail_bytes"`
+	WSPingInterval     time.Duration `mapstructure:"ws_ping_interval"`
+	WSPongTimeout      time.Duration `mapstructure:"ws_pong_timeout"`
+	WSSendBuffer       int           `mapstructure:"ws_send_buffer"`
+	EventMaxPayload    int           `mapstructure:"event_max_payload_bytes"`
+	EventTruncateField int           `mapstructure:"event_truncate_field_bytes"`
+	EventRetentionDays int           `mapstructure:"event_retention_days"`
+	ShutdownGrace      time.Duration `mapstructure:"shutdown_grace"`
+}
+
 // Load 从（按优先级递增）默认值 → 配置文件（如有）→ 环境变量 加载配置。
 // configFile 可为空字符串，仅用环境变量。
 func Load(configFile string) (*Config, error) {
@@ -210,6 +229,20 @@ func Load(configFile string) (*Config, error) {
 	v.SetDefault("jobs.job_purge.completed_retention", "168h")
 	v.SetDefault("notify.webhook.enabled", false)
 	v.SetDefault("notify.webhook.timeout", "10s")
+	v.SetDefault("acp.enabled", true)
+	v.SetDefault("acp.max_active_global", 20)
+	v.SetDefault("acp.max_active_per_user", 3)
+	v.SetDefault("acp.spawn_timeout", "30s")
+	v.SetDefault("acp.kill_grace", "5s")
+	v.SetDefault("acp.stderr_buffer_lines", 100)
+	v.SetDefault("acp.stderr_tail_bytes", 2000)
+	v.SetDefault("acp.ws_ping_interval", "30s")
+	v.SetDefault("acp.ws_pong_timeout", "60s")
+	v.SetDefault("acp.ws_send_buffer", 256)
+	v.SetDefault("acp.event_max_payload_bytes", 65536)
+	v.SetDefault("acp.event_truncate_field_bytes", 4096)
+	v.SetDefault("acp.event_retention_days", 7)
+	v.SetDefault("acp.shutdown_grace", "30s")

 	if configFile != "" {
 		v.SetConfigFile(configFile)