SynthesisAgent.OpenclawPlugin/core/process-manager.ts

import { spawn, type ChildProcess } from 'node:child_process'
import type { SynthesisConfig } from './config.js'
import type { SessionMapping } from './session-mapping.js'

export interface ProcessHandle {
  pid: number
  openclawSessionId: string
  claudeSessionUuid: string
  proc: ChildProcess
  startedAt: number
  lastActiveAt: number
  /** Resolved when the ClaudePlugin's hello frame has been seen on the bridge. */
  ready: Promise<void>
  markReady: () => void
}

export interface ProcessManagerDeps {
  config: SynthesisConfig
  mapping: SessionMapping
}

/**
 * Owns the pool of `claude` subprocesses. One per openclaw_session, lazily
 * spawned on first message and reaped after `idleKillMs` of inactivity.
 *
 * Hand-off to ClaudePlugin happens via env vars + a shared WebSocket bridge:
 * we know a process is "live" when its plugin connects back and sends `hello`.
 *
 * Responsibilities NOT handled here (delegated):
 *   - Routing inbound messages to a specific process → done by bridge-server,
 *     which holds the ws connection per pid.
 *   - Tool catalog → bridge-server sends in hello_ack.
 */
export class ProcessManager {
  private byOpenclawSession = new Map<string, ProcessHandle>()
  private idleSweeper: ReturnType<typeof setInterval> | null = null
  private shuttingDown = false

  constructor(private deps: ProcessManagerDeps) {
    this.idleSweeper = setInterval(() => this.sweepIdle(), 60_000)
  }

  /** Returns the live handle, spawning if needed. Awaits hello handshake. */
  async ensure(openclawSessionId: string): Promise<ProcessHandle> {
    const existing = this.byOpenclawSession.get(openclawSessionId)
    if (existing && !existing.proc.killed) {
      existing.lastActiveAt = Date.now()
      return existing
    }

    if (this.byOpenclawSession.size >= this.deps.config.maxProcesses) {
      this.evictOldestIdle()
    }

    const rec = this.deps.mapping.ensure(openclawSessionId)
    const handle = this.spawn(openclawSessionId, rec.claudeSessionUuid)
    this.byOpenclawSession.set(openclawSessionId, handle)
    // Wait for ClaudePlugin → bridge → bridge-server → handle.markReady().
    await Promise.race([
      handle.ready,
      new Promise<void>((_, rej) => setTimeout(() => rej(new Error('claude spawn ready timeout')), 15_000)),
    ])
    this.deps.mapping.touch(openclawSessionId)
    return handle
  }

  /** Bridge server calls this once it has paired a connection to a pid. */
  markReady(pid: number): void {
    for (const h of this.byOpenclawSession.values()) {
      if (h.pid === pid) {
        h.markReady()
        return
      }
    }
  }

  private spawn(openclawSessionId: string, claudeSessionUuid: string): ProcessHandle {
    const { config } = this.deps
    const args = [
      '--channels', config.claudePluginRef,
      '--resume', claudeSessionUuid,
      '--permission-mode', config.permissionMode,
      // Tool allowlist scoped to our MCP namespace plus the basics models need
      // for any non-trivial assist (Read, Edit, Bash). Tighten if/when policy
      // demands.
      '--allowed-tools', 'Read Edit Bash mcp__synthesis__*',
    ]

    const proc = spawn('claude', args, {
      env: {
        ...process.env,
        SYNTHESIS_BRIDGE_URL: `ws://127.0.0.1:${config.bridgePort}/bridge`,
        SYNTHESIS_BRIDGE_TOKEN: config.bridgeToken,
        SYNTHESIS_OPENCLAW_SESSION: openclawSessionId,
        SYNTHESIS_CLAUDE_SESSION: claudeSessionUuid,
      },
      stdio: ['ignore', 'pipe', 'pipe'],
      detached: false,
    })

    let resolveReady!: () => void
    const ready = new Promise<void>(r => { resolveReady = r })

    const handle: ProcessHandle = {
      pid: proc.pid ?? -1,
      openclawSessionId,
      claudeSessionUuid,
      proc,
      startedAt: Date.now(),
      lastActiveAt: Date.now(),
      ready,
      markReady: () => resolveReady(),
    }

    proc.on('exit', code => {
      this.byOpenclawSession.delete(openclawSessionId)
      // mapping is preserved so the next message resumes via --resume
      process.stderr.write(`synthesis: claude exit session=${openclawSessionId} pid=${handle.pid} code=${code}\n`)
    })

    proc.stderr?.on('data', chunk => {
      process.stderr.write(`[claude:${handle.pid}] ${chunk}`)
    })

    return handle
  }

  private sweepIdle(): void {
    if (this.shuttingDown) return
    const cutoff = Date.now() - this.deps.config.idleKillMs
    for (const h of this.byOpenclawSession.values()) {
      if (h.lastActiveAt < cutoff) {
        process.stderr.write(`synthesis: idle-killing session=${h.openclawSessionId} pid=${h.pid}\n`)
        h.proc.kill('SIGTERM')
      }
    }
  }

  private evictOldestIdle(): void {
    let oldest: ProcessHandle | null = null
    for (const h of this.byOpenclawSession.values()) {
      if (!oldest || h.lastActiveAt < oldest.lastActiveAt) oldest = h
    }
    if (oldest) {
      process.stderr.write(`synthesis: evicting session=${oldest.openclawSessionId} (max processes reached)\n`)
      oldest.proc.kill('SIGTERM')
    }
  }

  list(): ProcessHandle[] {
    return [...this.byOpenclawSession.values()]
  }

  async shutdown(): Promise<void> {
    this.shuttingDown = true
    if (this.idleSweeper) clearInterval(this.idleSweeper)
    for (const h of this.byOpenclawSession.values()) {
      try { h.proc.kill('SIGTERM') } catch { /* ignore */ }
    }
    this.byOpenclawSession.clear()
  }
}