Skip to content

armgit / cindy

Public Imported from GitHub
0

Consider it done. The open-source AI agent that works out of the box · 想到,就能做到。开源、开箱即用的 AI Agent。

fix(orca): 拒绝重新关联已归档的远端 Worker

DavidShen05533686208e
diff --git a/apps/desktop/src/main/localDb/__tests__/sessionOpening.test.ts b/apps/desktop/src/main/localDb/__tests__/sessionOpening.test.ts
index e8cced72ca..4cc19ed1e4 100644
--- a/apps/desktop/src/main/localDb/__tests__/sessionOpening.test.ts
+++ b/apps/desktop/src/main/localDb/__tests__/sessionOpening.test.ts
@@ -190,6 +190,7 @@ it('keeps a persisted remote Worker recognizable after startup fails or the host
readSession: async () => {
const row = h.values.mock.calls[0]?.[0] as OpenedSessionRow | undefined;
return row ? { orcaRemoteLead: parseOrcaRemoteLead(row.orcaRemoteLead),
+ status: row.status ?? 'active',
workingDir: row.workingDir ?? null, model: row.model, agentKind: 'codex' as const, effort: row.effort } : null;
},
openSession: start,
diff --git a/apps/desktop/src/main/maker-ipc/__tests__/orcaRemoteWorkerHost.test.ts b/apps/desktop/src/main/maker-ipc/__tests__/orcaRemoteWorkerHost.test.ts
index 49ac539710..bcc7dc9e22 100644
--- a/apps/desktop/src/main/maker-ipc/__tests__/orcaRemoteWorkerHost.test.ts
+++ b/apps/desktop/src/main/maker-ipc/__tests__/orcaRemoteWorkerHost.test.ts
@@ -39,6 +39,7 @@ function setup(
readSession: vi.fn(async (id: string) => sessions.get(id) ?? null),
openSession: vi.fn(async (req, lead: OrcaRemoteLead) => {
sessions.set(req.sessionId, {
+ status: 'active',
orcaRemoteLead: lead,
workingDir: req.workingDir ?? '/Users/demo/Cindy/dialogues/w_remote_1',
model: 'claude-opus-5-5',
@@ -114,6 +115,18 @@ describe('orca remote worker host', () => {
expect(deps.openSession).toHaveBeenCalledTimes(1);
});
+ it.each(['archived', 'deleted'] as const)('does not reopen a %s task after rebuilding the host', async status => {
+ const { host, deps, sessions } = setup();
+ await host.open(request);
+ sessions.set(request.sessionId, { ...sessions.get(request.sessionId)!, status });
+
+ await expect(createOrcaRemoteWorkerHost(deps).open(request)).rejects.toThrow(/PRECONDITION_FAILED/);
+ expect(deps.openSession).toHaveBeenCalledOnce();
+ expect(deps.writeRemoteLead).not.toHaveBeenCalled();
+ expect(sessions.get(request.sessionId)?.status).toBe(status);
+ expect(sessions.get(request.sessionId)?.orcaRemoteLead?.releasedAt).toBeUndefined();
+ });
+
it('replays the stored effort instead of a changed request or device default', async () => {
const { host, deps } = setup();
await host.open({ ...request, effort: 'low' });
@@ -124,7 +137,7 @@ describe('orca remote worker host', () => {
it.each([false, true])('replays stored Fast %s after rebuilding the host instead of a changed request', async fastMode => {
const { host, deps, sessions } = setup();
deps.openSession = vi.fn(async (req, lead) => {
- const stored = { orcaRemoteLead: lead, workingDir: '/remote', model: 'm',
+ const stored = { status: 'active' as const, orcaRemoteLead: lead, workingDir: '/remote', model: 'm',
agentKind: req.agentKind, fastMode };
sessions.set(req.sessionId, stored);
return stored;
@@ -137,7 +150,7 @@ describe('orca remote worker host', () => {
it('refuses an id already used by another task or device', async () => {
const plain = setup({
- existing: { orcaRemoteLead: null, workingDir: '/x', model: 'm', agentKind: 'codex' },
+ existing: { status: 'active', orcaRemoteLead: null, workingDir: '/x', model: 'm', agentKind: 'codex' },
});
await expect(plain.host.open(request)).rejects.toThrow(/ALREADY_EXISTS/);
diff --git a/apps/desktop/src/main/maker-ipc/__tests__/orcaRemoteWorkers.test.ts b/apps/desktop/src/main/maker-ipc/__tests__/orcaRemoteWorkers.test.ts
index c1043db920..e998cd1fab 100644
--- a/apps/desktop/src/main/maker-ipc/__tests__/orcaRemoteWorkers.test.ts
+++ b/apps/desktop/src/main/maker-ipc/__tests__/orcaRemoteWorkers.test.ts
@@ -828,6 +828,56 @@ describe('wrapTeamDeps', () => {
});
describe('remote open recovery', () => {
+ it.each(['archived', 'deleted'] as const)('rejects a lost open ACK retry after the task becomes %s and releases only its marker', async status => {
+ let session: OrcaRemoteWorkerExistingSession | null = null;
+ const host = createOrcaRemoteWorkerHost({
+ getCaller: () => ({ controllerDeviceId: 'lead-device' }),
+ readSession: async () => session,
+ openSession: vi.fn(async (request, lead) => {
+ session = { status: 'active', orcaRemoteLead: lead, workingDir: '/remote', model: 'device-model',
+ agentKind: request.agentKind, effort: 'medium' };
+ return { workingDir: '/remote', model: 'device-model', agentKind: request.agentKind, effort: 'medium' };
+ }),
+ writeRemoteLead: async (_id, lead) => { session = { ...session!, orcaRemoteLead: lead }; },
+ withSessionLock: async <T,>(_id: string, task: () => Promise<T>) => task(),
+ now: () => 123,
+ });
+ let loseOpenReply = true;
+ const handle = async (_device: string, channel: string, args: unknown[]) => {
+ if (channel === 'maker:orca:remote-worker:open') {
+ try {
+ const result = await host.open(args[0]);
+ if (loseOpenReply) {
+ loseOpenReply = false;
+ session = { ...session!, status };
+ return fail('INVOKE_TIMEOUT', 'ACK lost');
+ }
+ return ok(result);
+ } catch (err) { return fail('IPC_ERROR', (err as Error).message); }
+ }
+ if (channel === 'local-db:sessions:get') return ok(session);
+ if (channel === 'maker:orca:remote-worker:release') return ok(await host.release(args[0]));
+ return undefined;
+ };
+ const { workers, remoteInvoke } = setup({ handle });
+ try {
+ await expect(workers.openRemoteWorker(openInput)).resolves.toMatchObject({ ok: false });
+ expect(remoteInvoke.mock.calls.filter(([, channel]) => channel === 'maker:orca:remote-worker:open')).toHaveLength(2);
+ expect(store.addRemoteWorker).not.toHaveBeenCalled();
+ expect(store.removeRemoteWorkerOpen).not.toHaveBeenCalled();
+ const [, remoteSessionId] = store.saveRemoteWorkerOpen.mock.calls[0] as unknown as [string, string];
+ const receipt = { deviceId: 'mac-mini', remoteSessionId, createdAt: Date.now() };
+ store.listOrphanRemoteWorkerOpens.mockResolvedValueOnce([receipt] as never[]);
+ store.listOrphanRemoteWorkerOpens.mockResolvedValueOnce([receipt] as never[]);
+ await workers.releaseEnded([]);
+ expect(session).toMatchObject({ status, orcaRemoteLead: { releasedAt: 123 } });
+ expect(store.removeRemoteWorkerOpen).toHaveBeenCalledWith(remoteSessionId);
+ expect(store.archiveSingleWorkerSession).not.toHaveBeenCalled();
+ expect(remoteInvoke.mock.calls.some(([, channel]) =>
+ channel === 'maker:abort-session' || channel === 'maker:input:enqueue')).toBe(false);
+ } finally { workers.stop(); }
+ });
+
it.each(['startup failed', '[INTERNAL] startup failed'])(
'reconciles a persisted task after a non-timeout startup error (%s)', async message => {
let session: OrcaRemoteWorkerExistingSession | null = null;
@@ -835,7 +885,7 @@ describe('remote open recovery', () => {
getCaller: () => ({ controllerDeviceId: 'lead-device' }),
readSession: async () => session,
openSession: async (_request, lead) => {
- session = { orcaRemoteLead: lead, workingDir: '/remote', model: 'device-model',
+ session = { status: 'active', orcaRemoteLead: lead, workingDir: '/remote', model: 'device-model',
agentKind: 'claude-code', effort: 'medium' };
throw new Error(message);
},
@@ -891,7 +941,7 @@ describe('remote open recovery', () => {
readSession: async id => sessions.get(id) ?? null,
openSession: async (request, lead) => {
const opened = { workingDir: '/remote', model: 'device-model', agentKind: request.agentKind, effort: 'medium' };
- sessions.set(request.sessionId, { ...opened, orcaRemoteLead: lead });
+ sessions.set(request.sessionId, { ...opened, status: 'active', orcaRemoteLead: lead });
if (controllerDeviceId === 'first-controller') throw new Error('startup failed');
return opened;
},
diff --git a/apps/desktop/src/main/maker-ipc/orcaRemoteWorkerHost.ts b/apps/desktop/src/main/maker-ipc/orcaRemoteWorkerHost.ts
index ce4bef6133..901734293e 100644
--- a/apps/desktop/src/main/maker-ipc/orcaRemoteWorkerHost.ts
+++ b/apps/desktop/src/main/maker-ipc/orcaRemoteWorkerHost.ts
@@ -64,6 +64,7 @@ export interface OrcaRemoteWorkerCaller {
}
export interface OrcaRemoteWorkerExistingSession {
+ status: 'active' | 'archived' | 'deleted';
orcaRemoteLead: OrcaRemoteLead | null;
workingDir: string | null;
model: string;
@@ -122,13 +123,16 @@ export function createOrcaRemoteWorkerHost(deps: OrcaRemoteWorkerHostDeps) {
const existing = await deps.readSession(request.sessionId);
if (existing) {
const lead = existing.orcaRemoteLead;
- // 幂等:同一来源电脑、同一 Lead 重复 open(例如超时后重试)返回同一任务。
+ // 幂等:同一来源电脑、同一 Lead 重复 open(例如超时后重试)只返回仍活跃的任务。
if (
lead &&
lead.leadDeviceId === caller.controllerDeviceId &&
lead.leadSessionId === request.lead.leadSessionId &&
lead.releasedAt === undefined
) {
+ if (existing.status !== 'active') {
+ fail('PRECONDITION_FAILED', 'remote worker task is no longer active');
+ }
return {
sessionId: request.sessionId,
workingDir: existing.workingDir ?? '',
diff --git a/apps/desktop/src/main/maker-ipc/register.ts b/apps/desktop/src/main/maker-ipc/register.ts
index 49f800eb78..92194191d2 100644
--- a/apps/desktop/src/main/maker-ipc/register.ts
+++ b/apps/desktop/src/main/maker-ipc/register.ts
@@ -8497,6 +8497,7 @@ export function registerMakerIpc(maker: Maker, options: RegisterMakerIpcOptions)
const [row] = await getDbClient()
.drizzle.select({
orcaRemoteLead: sessions.orcaRemoteLead,
+ status: sessions.status,
workingDir: sessions.workingDir,
model: sessions.model,
agentKind: sessions.agentKind,
@@ -8509,6 +8510,7 @@ export function registerMakerIpc(maker: Maker, options: RegisterMakerIpcOptions)
if (!row) return null;
return {
orcaRemoteLead: parseOrcaRemoteLead(row.orcaRemoteLead),
+ status: row.status,
workingDir: row.workingDir,
model: row.model,
agentKind: row.agentKind === 'cc' ? 'claude-code' : (row.agentKind as 'codex' | 'pi'),
diff --git a/docs/dev-rules/orca-team-architecture.md b/docs/dev-rules/orca-team-architecture.md
index 459dd10863..0466f616f5 100644
--- a/docs/dev-rules/orca-team-architecture.md
+++ b/docs/dev-rules/orca-team-architecture.md
@@ -398,6 +398,9 @@ Worker turn 被 vendor 报终止型 error,但 interrupted-turn auto-resume 仍
Worker 关联共用任务锁,锁内复查后仅解除未关联标记,保留任务、文件及用户发起的工作。
本机代理任务行在 Worker 关联时才创建,与 Worker、远端路由及 open 收据在同一事务提交;
open 成功后未关联即退出时,本机没有半成品代理,仅按持久 open 身份解除远端标记。
+同 ID open 仅对来源匹配、尚未 release 且 `status = active` 的任务返回成功;用户已归档或
+软删除的任务返回 `PRECONDITION_FAILED`,不复活、不关联或派活。创建收据保留供孤儿清理
+仅解除来源标记,清理不停止或删除任务。回归覆盖丢回执后归档/删除再重试及 Host 重建。
运行设备的真实任务与 `orcaRemoteLead` 在同一 INSERT 写入,再启动 Agent;启动失败或
进程在两者之间退出时,同一来源与 Lead 仍能按任务 ID 对账,不留下无来源标记的普通任务。
实现见 `orcaRemoteWorkerHost.ts` 的 `createOrcaRemoteWorkerSessionOpener`,回归见
@@ -419,6 +422,13 @@ Mobile 的入口读取和残留表单提交都检查该身份,元数据补齐
代替解决;后续应在 Orca 专用边界设计持久幂等停止、目标轮次条件与旧端能力协商,覆盖
丢回执后新轮次、重启、并发重试及账号切换。当前实现尚未消除此风险。
+当前结束协同的另一已知限制:abort 只停止当前轮次,release 只写结束标记,尚未执行的
+远端 Lead 派活可能仍留在持久队列并随后执行。当前远控入队会剥离 Orca 语义来源,来源
+设备也不能区分同一设备发出的用户消息与自动派活;仅用最后一个 clientId 无法覆盖多条
+未消费输入。不直接清空整队列,以免丢失用户消息。完整修复须先建立可持久核对的 Lead
+投递归属及释放边界,再持久化取消回执和队列快照;覆盖多条派活、用户混排、消费竞态、
+丢回执、重启与其它控制端。该问题本轮未修复,不能视为结束协同已保证清除待执行派活。
+
当前文档要求保留以下回归方向:
- Service 边界:`orcaLifecycleService`、`orcaWorkerCreationService`、`orcaTeamService` 的单测覆盖 start/enable/create/dispatch/idle/archive/auto-bridge 关键路径。
diff --git a/docs/dev-rules/protocol-compatibility.md b/docs/dev-rules/protocol-compatibility.md
index b72cd855b5..cb6c2811e2 100644
--- a/docs/dev-rules/protocol-compatibility.md
+++ b/docs/dev-rules/protocol-compatibility.md
@@ -631,6 +631,10 @@ B 上的 Worker 是一条普通任务,`sessions.orca_remote_lead`(migration 01
`maker:orca:remote-worker:caps`(能力探测,回 `{ version }`)、`…:open`(按 A 给定的任务 id 新建 Worker 任务,
幂等;超时 60s)、`…:release`(结束协同,任务与文件保留,幂等)。B 侧实现见
`apps/desktop/src/main/maker-ipc/orcaRemoteWorkerHost.ts`。
+- **幂等 open 的终态检查**:同来源、同 Lead 且未 release 的已有任务也必须仍为 `active`;
+ 已归档/软删除时拒绝并返回既有 `PRECONDITION_FAILED`,不新建、不复活。A 保留未确认创建
+ 收据,恢复仅解除未关联标记。该检查使用 B 的现有 status 列,不新增 wire 字段或改变版本;
+ 旧 B 仍保留原行为,B 更新后生效。
- **实际档位**:`open` 回包可选 `effort` 是 B 已保存的解析结果,同 ID 重试也从已有任务读取。
空字符串保留无档位状态;旧 B 不返回该字段时,A 保留原来的请求值降级规则。字段增量不改协议版本,
旧 A 忽略它,新 A 不向旧 B 要求新增请求字段。
@@ -656,6 +660,9 @@ B 上的 Worker 是一条普通任务,`sessions.orca_remote_lead`(migration 01
- **恢复暂停**:B 的 `INPUT_ENQUEUE` 按已有自动消息来源判据处理 `origin.kind`,Orca 派活不解除
崩溃恢复后的队列暂停;普通手机/桌面用户输入仍可解除恢复暂停,用户 Stop 暂停不受影响。
不新增 wire 字段或改变版本,A 的既有派活载荷不变;旧 B 保持其原有行为,需更新 B 才有此修复。
+- **收尾限制**:当前 abort/release 不取消未消费的远端派活,结束协同后队列仍可能继续执行。
+ 远控入队后的来源不足以可靠区分 Lead 派活与用户输入,本轮不通过清空队列规避;待专门设计
+ 持久投递归属、取消回执与释放边界后修复,详见 `orca-team-architecture.md` 的已知限制。
- **B 侧约束**:带标记的任务不能再开启协同(`assertLeadCollabProjectEnabled` 统一拒绝 Worker 与远端 Worker,
覆盖 IPC、远程与 Agent 工具入口),不能复制到其他电脑(`task-migration/service.ts`);侧栏照常显示,
任务头标注「来自 X 的协同」,结束后显示「协同已结束」。
diff --git a/docs/dev-rules/remote-and-mobile-adaptation.md b/docs/dev-rules/remote-and-mobile-adaptation.md
index f9da7e5278..fb5ea7282d 100644
--- a/docs/dev-rules/remote-and-mobile-adaptation.md
+++ b/docs/dev-rules/remote-and-mobile-adaptation.md
@@ -96,6 +96,8 @@ relay 以 close 1013 `inbound backpressure` 主动断连,此时任何「立即
结束时先确认停止并落盘,再发 release;重连或 5 分钟补发只继续未完成的阶段。
清理共用单 Worker 入口并读最新阶段,不能因迟到快照重复停止用户后续的普通任务。
open 丢回执时用相同 sessionId 核对一次;无法确认的创建身份落盘,恢复后只解除未关联任务的标记。
+ 同 ID 核对要求来源匹配且任务仍为 active;归档或软删除返回 PRECONDITION_FAILED,不复活或
+ 关联任务,保留创建收据供只解除标记的孤儿清理。该检查不改变其它设备与普通任务的生命周期。
普通启动错误也可能发生在任务落库之后,不能据此删除 open 收据;只有通道拒绝或明确 ID 冲突
且没有先前在途 open 时直接清理。其余错误沿用单任务核对,不停止任务、不拆链路。
清理与 Worker 关联共用任务锁,锁内复查归属;公开写操作和迟到回调均核对数据库 owner。
@@ -120,6 +122,11 @@ relay 以 close 1013 `inbound backpressure` 主动断连,此时任何「立即
**已知未修复缺口**:远端 queued 派活仍提前执行 accepted/commit,并把旧、新输入合并为同一待回报身份,可能阻塞或覆盖旧轮次回报,也会提前清除手动中断标记。
修复需要将 active report 与未消费投递分别持久化,并覆盖连续轮次、撤回、未知回执、accepted 回滚与重启恢复;仅延迟两个回调不能补齐这个缺口。
+ **收尾未修复缺口**:abort 不清除持久队列,release 后的未消费 Lead 派活仍可能执行。
+ 当前远控入队剥离 Orca 语义来源,来源设备不足以区分自动派活与同设备用户输入;本轮不清空
+ 整队列或只删最后一个 clientId。后续须补可持久核对的投递归属及释放边界,取消只作用于该
+ Lead 的未消费派活,保留用户消息;覆盖混排、消费竞态、丢回执、重启及其它控制端。
+
## 共享恢复与请求策略
普通出站 invoke 在共享 `InvokeScheduler` 中按目标设备排队:每台最多 12 个在途,