fix(pty): await unpublished spawn teardown

This commit is contained in:
Tianyi Cui
2026-07-23 00:11:20 +08:00
parent 306681b53b
commit 672f2e5ec5
6 changed files with 130 additions and 27 deletions
@@ -2,5 +2,5 @@
# side as of the last confirmed-consistent state. Both languages carry equal authority;
# after editing either side, bring the other along and re-record with:
# pnpm run verify-translation-pairing --write
2026-07-16-persistent-pty-sessions.md: afa6f1771931ed437d8c2c34204ad4a59d02248e
2026-07-16-persistent-pty-sessions.zh.md: 2af97c3d126f8639e6b52685fb7fea1292343c9b
2026-07-16-persistent-pty-sessions.md: a73b8d0247243cc0d9cc93f27d84a8540b3b271d
2026-07-16-persistent-pty-sessions.zh.md: b0dc4010f69206b7aaeab4164199ffd935165955
@@ -34,7 +34,7 @@ Idle detection is backend behavior, not a second public seam. A remote or contai
There are no plugin-load auto-start sessions. `terminal_open` creates a session only during an agent tool call, when ownership and the owning event-sourced session are known. A future declarative startup feature must compose through unpublished agent setup rather than create shared global terminals.
Agent-scope disposal closes registrations first, then awaits quiescent teardown of every owned PTY. Backend or tool-plugin reload does not orphan sessions: ownership lives in `PtyService` until the agent ends, following the same service-owned-record pattern as [`ctx.tasks`](../../../../packages/tasks/tasks/README.md). The service reserves the session synchronously for one active send before returning its operation, including before a background task id becomes visible; a second send fails with `SEND_ACTIVE`, so output and cancellation cannot cross operation ownership.
Agent-scope disposal closes registrations first, then awaits quiescent teardown of every owned PTY. Unpublished backend setup is a tracked lifecycle operation: owner or service disposal aborts its service-owned signal, waits for backend settlement and rollback, and only then returns. Caller cancellation retains its exact `AbortSignal.reason` even when the backend rejects in response. Backend or tool-plugin reload does not orphan sessions: ownership lives in `PtyService` until the agent ends, following the same service-owned-record pattern as [`ctx.tasks`](../../../../packages/tasks/tasks/README.md). The service reserves the session synchronously for one active send before returning its operation, including before a background task id becomes visible; a second send fails with `SEND_ACTIVE`, so output and cancellation cannot cross operation ownership.
### Security and process boundary
@@ -152,7 +152,7 @@ The package ships concise tool guidance explaining persistent state, owner isola
## Verification
- Per-file coverage pins owner fencing, concurrent reservations, sandbox-mode change rejection, retriable lifecycle cleanup, readiness tiers, sanitizer carry state, complete UTF-8 bounds, task integration, schemas, and exact render intents.
- Per-file coverage pins owner fencing, concurrent reservations, unpublished-spawn cancellation and awaited teardown, sandbox-mode change rejection, retriable lifecycle cleanup, readiness tiers, sanitizer carry state, complete UTF-8 bounds, task integration, schemas, and exact render intents.
- Linux process fixtures cover non-leader and non-main-thread stdin waits, zombie quiescence, unreadable process state, supported syscall tables, unsupported architectures, and false-positive rejection; macOS inspector logic is injected into the same unit suite.
- Real `node-pty` tests exercise shell state, shared sandbox policy, environment scrubbing, raw-mode foreground `SIGINT`, a TERM-ignoring descendant, and immediate post-disposal quiescence on supported hosts.
- A Loader-driven `cordis.yml` test mounts the real three-package composition, while ACP and headless snapshots pin the six schemas, bounded results, error rendering, and terminal/generic cards through opt-in overlays.
@@ -34,7 +34,7 @@ idle 检测属于后端行为,不是第二条公共 seam。远程或容器后
实现不提供插件加载期 auto-start 会话。`terminal_open` 只在 agent 工具调用期间创建会话,此时所有权和所属的事件溯源会话都已确定。未来的声明式启动功能必须通过尚未发布的 agent setup 组合,而不能创建全局共享终端。
agent scope dispose 时先关闭注册,再等待全部所属 PTY 静默退出。后端或工具插件 reload 不会遗留会话:所有权持续存放在 `PtyService` 中,直到 agent 结束,与 [`ctx.tasks`](../../../../packages/tasks/tasks/README.md) 的服务持有记录模式一致。服务会先同步把会话预留给一次活跃发送,再返回该操作;后台发送同样会在 task id 对外可见前完成预留。第二次发送会以 `SEND_ACTIVE` 失败,因此输出与取消无法跨越操作所有权。
agent scope dispose 时先关闭注册,再等待全部所属 PTY 静默退出。未发布的后端 setup 同样是受追踪的生命周期操作:owner 或服务 dispose 会中止服务自有的 signal,等待后端结算与回滚完成后才返回。即使后端响应取消而 reject,调用方取消仍原样保留其 `AbortSignal.reason`后端或工具插件 reload 不会遗留会话:所有权持续存放在 `PtyService` 中,直到 agent 结束,与 [`ctx.tasks`](../../../../packages/tasks/tasks/README.md) 的服务持有记录模式一致。服务会先同步把会话预留给一次活跃发送,再返回该操作;后台发送同样会在 task id 对外可见前完成预留。第二次发送会以 `SEND_ACTIVE` 失败,因此输出与取消无法跨越操作所有权。
### 安全与进程边界
@@ -152,7 +152,7 @@ plugins:
## 验证
- 每文件覆盖率固定 owner 隔离、并发预留、沙箱模式变更拒绝、可重试的生命周期清理、就绪层级、sanitizer carry state、完整 UTF-8 结果上限、task 集成、schema 和精确 render intent。
- 每文件覆盖率固定 owner 隔离、并发预留、未发布 spawn 的取消与等待式 teardown、沙箱模式变更拒绝、可重试的生命周期清理、就绪层级、sanitizer carry state、完整 UTF-8 结果上限、task 集成、schema 和精确 render intent。
- Linux 进程 fixture 覆盖非 leader 与非主线程的 stdin 等待、僵尸进程静止性、不可读进程状态、受支持的 syscall 表、不支持的架构和误报拒绝;同一单元测试套件通过注入覆盖 macOS 检查器逻辑。
- 真实 `node-pty` 测试在受支持宿主上覆盖 shell 状态、共享沙箱策略、环境清洗、raw mode 下的前台 `SIGINT`、忽略 `SIGTERM` 的子进程,以及 dispose 返回后立即静默。
- Loader 驱动的 `cordis.yml` 测试挂载真实三包组合;ACP 与 headless 快照通过 opt-in overlay 固定 6 个 schema、有界结果、错误渲染和 terminal/generic card。
+1
View File
@@ -6,6 +6,7 @@ Owner-scoped persistent PTY seam. `PtyService` registers as `ctx.pty`, mints opa
- Backends register one stable `type` and return an unpublished `PtyBackendSession`; failed or cancelled setup must clean partial resources.
- Spawn cancellation preserves the caller's exact abort reason. Service disposal and owner loss remain distinct machine-routable failures after backend setup.
- Owner and service disposal abort unpublished setup through a service-owned signal and await backend settlement plus rollback before returning.
- `hasOwnerActivity(owner)` spans unpublished setup through final close, so lifecycle policy can fence the exact owner without a publication race.
- A successful spawn publishes one `PtySessionId`. The optional `name` is owner-local display metadata, never authority.
- One session accepts at most one live send operation. Reads and signals may observe it; another send fails until the operation settles.
+56 -14
View File
@@ -87,12 +87,22 @@ interface SessionRecord {
closing: Promise<void> | undefined
}
interface PendingSpawn {
readonly controller: AbortController
readonly settled: Promise<void>
}
interface SpawnReservation {
readonly signal: AbortSignal
release(): void
}
/** In-process registry for replaceable PTY backends and exact-Agent sessions. */
export class PtyService extends Service {
private readonly backends = new Map<string, PtyBackend>()
private readonly sessions = new Map<PtySessionId, SessionRecord>()
private readonly reservedNames = new Map<Agent, Set<string>>()
private readonly pendingSpawns = new Map<Agent, number>()
private readonly pendingSpawns = new Map<Agent, Set<PendingSpawn>>()
private readonly ownerCleanups = new Map<Agent, () => Promise<void> | void>()
private readonly disposedOwners = new WeakSet<Agent>()
private nextId = 0
@@ -145,7 +155,10 @@ export class PtyService extends Service {
if (backend === undefined) throw new PtyError(`no PTY backend registered for "${request.type}"`, 'NO_BACKEND')
if (request.name !== undefined && request.name.length === 0) throw new Error('PTY session name must be non-empty')
const releaseName = this.reserveName(owner, request.name)
const releaseSpawn = this.reserveSpawn(owner)
const spawnReservation = this.reserveSpawn(owner)
const backendSignal = signal === undefined
? spawnReservation.signal
: AbortSignal.any([signal, spawnReservation.signal])
const sessionId = PtySessionId(`pty-${++this.nextId}`)
let session: PtyBackendSession | undefined
try {
@@ -155,7 +168,7 @@ export class PtyService extends Service {
type: request.type,
...request.name !== undefined ? { name: request.name } : {},
...request.cwd !== undefined ? { cwd: request.cwd } : {},
...signal !== undefined ? { signal } : {},
signal: backendSignal,
})
signal?.throwIfAborted()
if (this.disposing) {
@@ -176,16 +189,27 @@ export class PtyService extends Service {
this.sessions.set(sessionId, record)
return this.snapshot(record, session.motd)
} catch (error) {
let rollbackFailure: { error: unknown } | undefined
if (session !== undefined && !this.sessions.has(sessionId)) {
try {
await session.close('PTY spawn rolled back')
} catch (closeError: unknown) {
throw new AggregateError([error, closeError], 'PTY spawn and rollback both failed')
rollbackFailure = { error: closeError }
}
}
throw error
let failure: unknown = error
try {
signal?.throwIfAborted()
spawnReservation.signal.throwIfAborted()
} catch (cancellation: unknown) {
failure = cancellation
}
if (rollbackFailure !== undefined) {
throw new AggregateError([failure, rollbackFailure.error], 'PTY spawn and rollback both failed')
}
throw failure
} finally {
releaseSpawn()
spawnReservation.release()
releaseName()
}
}
@@ -196,7 +220,7 @@ export class PtyService extends Service {
* @returns true across the entire spawn-to-close interval, with no publication gap.
*/
hasOwnerActivity(owner: Agent): boolean {
return (this.pendingSpawns.get(owner) ?? 0) > 0
return (this.pendingSpawns.get(owner)?.size ?? 0) > 0
|| [...this.sessions.values()].some(record => record.owner === owner)
}
@@ -314,15 +338,31 @@ export class PtyService extends Service {
}
}
private reserveSpawn(owner: Agent): () => void {
this.pendingSpawns.set(owner, (this.pendingSpawns.get(owner) ?? 0) + 1)
return () => {
const remaining = (this.pendingSpawns.get(owner) ?? 1) - 1
if (remaining === 0) this.pendingSpawns.delete(owner)
else this.pendingSpawns.set(owner, remaining)
private reserveSpawn(owner: Agent): SpawnReservation {
const controller = new AbortController()
const settlement = Promise.withResolvers<void>()
const pending: PendingSpawn = { controller, settled: settlement.promise }
const owned = this.pendingSpawns.get(owner) ?? new Set<PendingSpawn>()
owned.add(pending)
this.pendingSpawns.set(owner, owned)
return {
signal: controller.signal,
release: () => {
owned.delete(pending)
if (owned.size === 0) this.pendingSpawns.delete(owner)
settlement.resolve()
},
}
}
private async abortPendingSpawns(owner: Agent | undefined, reason: PtyError): Promise<void> {
const pending = owner === undefined
? [...this.pendingSpawns.values()].flatMap(owned => [...owned])
: [...(this.pendingSpawns.get(owner) ?? [])]
for (const spawn of pending) spawn.controller.abort(reason)
await Promise.all(pending.map(spawn => spawn.settled))
}
private expectOwned(owner: Agent, id: PtySessionId): SessionRecord {
const record = this.sessions.get(id)
if (record === undefined) throw new PtyError(`unknown PTY session ${id}`, 'NO_SESSION')
@@ -344,6 +384,7 @@ export class PtyService extends Service {
}
private async disposeOwned(owner: Agent): Promise<void> {
await this.abortPendingSpawns(owner, new PtyError('PTY owner is no longer live', 'OWNER_NOT_LIVE'))
const owned = [...this.sessions.values()].filter(record => record.owner === owner)
await this.closeRecords(owned, 'PTY owner disposed')
this.reservedNames.delete(owner)
@@ -351,11 +392,12 @@ export class PtyService extends Service {
private async disposeAll(): Promise<void> {
this.disposing = true
const records = [...this.sessions.values()]
// Teardown is best-effort: a close failure still clears registries and runs
// owner cleanups before the aggregated error propagates, so one stuck
// session cannot orphan backends, reservations, or owner detachers.
try {
await this.abortPendingSpawns(undefined, new PtyError('PTY service is disposing', 'SERVICE_DISPOSING'))
const records = [...this.sessions.values()]
await this.closeRecords(records, 'PTY service disposed')
} finally {
this.backends.clear()
+67 -7
View File
@@ -206,9 +206,10 @@ describe('PtyService ownership and lifecycle', () => {
ctx.agents.register(owner)
const pending = ctx.pty.spawn(owner, { type: 'slow', name: 'main' })
await expect(ctx.pty.spawn(owner, { type: 'slow', name: 'main' })).rejects.toMatchObject({ code: 'DUPLICATE_NAME' })
await disposeAgentScope(owner)
const disposal = disposeAgentScope(owner)
gate.resolve(session)
await expect(pending).rejects.toMatchObject({ code: 'OWNER_NOT_LIVE' })
await disposal
expect(session.closed).toEqual(['PTY spawn rolled back'])
})
@@ -231,19 +232,70 @@ describe('PtyService ownership and lifecycle', () => {
expect(ctx.agents.get(owner.id)).toBe(owner)
})
it('rolls back an unpublished backend session when service disposal wins', async () => {
it('preserves caller cancellation when a backend rejects in response to it', async () => {
const ctx = await harness()
const started = Promise.withResolvers<undefined>()
const backendFailure = new Error('backend observed cancellation')
ctx.pty.registerBackend({
type: 'abortable',
spawn: ({ signal }) => new Promise((_resolve, reject) => {
if (signal === undefined) throw new Error('missing spawn signal')
started.resolve(undefined)
signal.addEventListener('abort', () => { reject(backendFailure) }, { once: true })
}),
})
const owner = stubAgent(ctx, 'owner')
ctx.agents.register(owner)
const controller = new AbortController()
const reason = new Error('cancelled by caller')
const pending = ctx.pty.spawn(owner, { type: 'abortable' }, controller.signal)
await started.promise
controller.abort(reason)
await expect(pending).rejects.toBe(reason)
})
it.each([
{ scope: 'owner', code: 'OWNER_NOT_LIVE' },
{ scope: 'service', code: 'SERVICE_DISPOSING' },
] as const)('$scope disposal aborts and awaits unpublished backend setup', async ({ scope, code }) => {
const ctx = await harness()
const gate = Promise.withResolvers<PtyBackendSession>()
const started = Promise.withResolvers<undefined>()
const session = new StubSession()
ctx.pty.registerBackend({ type: 'slow', spawn: () => gate.promise })
let backendSignal: AbortSignal | undefined
ctx.pty.registerBackend({
type: 'slow',
spawn: (spec) => {
backendSignal = spec.signal
started.resolve(undefined)
return gate.promise
},
})
const owner = stubAgent(ctx, 'owner')
ctx.agents.register(owner)
const pending = ctx.pty.spawn(owner, { type: 'slow' })
await disposePtyService(ctx)
const pendingFailure = pending.then(
() => { throw new Error('pending spawn unexpectedly succeeded') },
(error: unknown) => error,
)
await started.promise
let disposalSettled = false
const disposal = (scope === 'owner' ? disposeAgentScope(owner) : disposePtyService(ctx))
.then(() => { disposalSettled = true })
await new Promise(resolve => setTimeout(resolve, 0))
const signalAbortedBeforeRelease = backendSignal?.aborted ?? false
const signalReasonBeforeRelease = backendSignal?.reason as unknown
const disposalSettledBeforeRelease = disposalSettled
gate.resolve(session)
await expect(pending).rejects.toMatchObject({ code: 'SERVICE_DISPOSING' })
expect(await pendingFailure).toMatchObject({ code })
await disposal
expect(signalAbortedBeforeRelease).toBe(true)
expect(signalReasonBeforeRelease).toMatchObject({ code })
expect(disposalSettledBeforeRelease).toBe(false)
expect(session.closed).toEqual(['PTY spawn rolled back'])
})
@@ -290,14 +342,22 @@ describe('PtyService ownership and lifecycle', () => {
ctx.agents.register(owner)
const failedSpawn = new StubSession()
failedSpawn.rejectClose = true
let ownerDisposal = Promise.resolve()
ctx.pty.registerBackend({
type: 'bad-spawn',
async spawn() {
await disposeAgentScope(owner)
async spawn({ signal }) {
if (signal === undefined) throw new Error('missing spawn signal')
ownerDisposal = disposeAgentScope(owner)
if (!signal.aborted) {
await new Promise<undefined>((resolve) => {
signal.addEventListener('abort', () => { resolve(undefined) }, { once: true })
})
}
return failedSpawn
},
})
await expect(ctx.pty.spawn(owner, { type: 'bad-spawn' })).rejects.toThrow('spawn and rollback both failed')
await ownerDisposal
const nextOwner = stubAgent(ctx, 'next')
ctx.agents.register(nextOwner)