Merge pull request #13 from ai-workspace-lab/fix/gateway-turn-stability-day1
fix(gateway): day-1 stability — stop infinite running & un-stoppable tasks
This commit is contained in:
GitHub
commit
da3a654ab4
189
docs/cases/06-gateway-turn-stability-and-robustness.md
Normal file
189
docs/cases/06-gateway-turn-stability-and-robustness.md
Normal file
@ -0,0 +1,189 @@
|
||||
# Gateway Turn 稳定性与健壮性｜全链路分析与编码改进规划
|
||||
|
||||
> 适用目录：`xworkmate-app/docs/cases/`
|
||||
>
|
||||
> 相关仓库（`/Users/shenlan/workspaces/`）：
|
||||
> - `ai-workspace-infra/`（Caddy 入口、OpenClaw gateway 部署、playbooks）
|
||||
> - `ai-workspace-lab/`（`xworkmate-app` Flutter 客户端、`xworkmate-bridge` Go 网桥）
|
||||
> - `ai-workspace-service/`（accounts / console / litellm 等旁路服务）
|
||||
>
|
||||
> 触发现象：任务进度条 `任务运行中...` 永不结束、`停止` 无效，**实际 OpenClaw gateway 已执行完毕**；
|
||||
> 伴随报错 `Bridge 响应读取中断，本轮结果未完成。错误码：ACP_HTTP_CONNECTION_CLOSED`。
|
||||
|
||||
---
|
||||
|
||||
## 1. The full chain (one gateway turn)
|
||||
|
||||
一次 gateway turn 的完整时序：
|
||||
|
||||
```
|
||||
App turn ──SSE POST──▶ Bridge http_handler ──▶ handleRequest ──▶ OpenClaw gateway submit
|
||||
(executeTask) (text/event-stream) (阻塞 6min/60min) (WS 127.0.0.1:18789)
|
||||
│ │ │
|
||||
│ keepalive 20s ◀────┘ ▼
|
||||
│ 返回 "running task handle"(runId + artifactScope)
|
||||
▼ │
|
||||
persist association ──▶ pollOpenClawTaskAssociationInternal (tasks.get 每 2s)
|
||||
│
|
||||
isOpenClawRunningTaskHandle? ──yes──▶ persist + continue ──┐
|
||||
│ │
|
||||
no ──▶ applyGatewayChatResult (terminal) └─ ∞ (无 deadline)
|
||||
```
|
||||
|
||||
完成信号（completion signal）只活在**两个脆弱位置**：
|
||||
|
||||
1. **带内 SSE 结果帧**（result envelope / `[DONE]`）—— 依赖那条长连接活到任务结束；
|
||||
2. **Gateway 对 `tasks.get` 的内存态应答** —— 依赖 bridge↔gateway 的 WS 连接仍持有该 run 的状态。
|
||||
|
||||
截图场景里这两条同时被打断，于是"任务跑完了，但客户端永远收不到终态"。
|
||||
|
||||
关键代码锚点：
|
||||
|
||||
| 环节 | 位置 |
|
||||
|---|---|
|
||||
| App 发起 SSE turn | `xworkmate-app/lib/app/app_controller_desktop_thread_actions.dart:644` (`executeTask`) |
|
||||
| Bridge SSE handler / keepalive 20s | `xworkmate-bridge/internal/acp/http_handler.go:198-283`、`:19` |
|
||||
| Bridge 阻塞等待 gateway（6min 默认 / 60min 上限） | `xworkmate-bridge/internal/acp/orchestrator.go:31-33` |
|
||||
| Bridge↔Gateway WebSocket（dial 18789） | `xworkmate-bridge/internal/gatewayruntime/runtime.go:372-376` |
|
||||
| App 持久化 + 轮询 run | `app_controller_desktop_thread_actions.dart:680-693`、`:747` |
|
||||
| running handle 判定 | `xworkmate-app/lib/runtime/go_task_service_client.dart:519` |
|
||||
| 进度条 phase（仅看 pending） | `xworkmate-app/lib/widgets/assistant_task_progress_bar.dart:190` |
|
||||
|
||||
---
|
||||
|
||||
## 2. 三仓库端到端拓扑（实测自代码）
|
||||
|
||||
```
|
||||
┌─ ai-workspace-lab/xworkmate-app (Flutter)
|
||||
│ executeTask → SSE POST Accept: text/event-stream
|
||||
▼
|
||||
┌─ ai-workspace-infra Caddy 入口 xworkmate-bridge.svc.plus
|
||||
│ handle /acp* : flush_interval -1 ✓ read_timeout 30m write_timeout 30m keepalive 5m
|
||||
│ handle /api* : (Caddy 默认超时, 无 flush_interval) ⚠
|
||||
│ handle / : (Caddy 默认) ⚠
|
||||
▼
|
||||
┌─ ai-workspace-lab/xworkmate-bridge (Go)
|
||||
│ SSE handler keepalive 20s → handleRequest 阻塞 6min(默认)/60min(max)
|
||||
│ gatewayruntime: WebSocket → 127.0.0.1:18789 (pending map 绑定在连接生命周期上)
|
||||
▼
|
||||
┌─ ai-workspace-infra deploy_gateway_openclaw (roles/vhosts/gateway_openclaw)
|
||||
│ OpenClaw gateway runtime, WS 18789
|
||||
▼
|
||||
OpenClaw 执行 → artifacts → tasks.get 回查
|
||||
```
|
||||
|
||||
旁路（`ai-workspace-service`）：`accounts.svc.plus`（登录 / Token）、`console.svc.plus`（自带 openclaw assistant route）、`litellm` / `AI-Relay-Kit` / `codex-relay`（模型出口）、`qmd`（记忆）。
|
||||
它们不在本次卡死主链上，但 **Token 失效 / 出口异常会以同样的"连接中断"形态**表现出来，排查时需先用 `runId` 区分是主链断还是旁路断。
|
||||
|
||||
入口配置出处：`ai-workspace-infra/playbooks/roles/vhosts/xworkmate_bridge/templates/xworkmate-bridge-site.caddy.j2`
|
||||
Gateway 部署出处：`ai-workspace-infra/playbooks/deploy_gateway_openclaw.yml` → `roles/vhosts/gateway_openclaw/`
|
||||
|
||||
---
|
||||
|
||||
## 3. 跨层失效点清单（按链路从上到下）
|
||||
|
||||
| # | 层 | 失效描述 | 证据 |
|
||||
|---|---|---|---|
|
||||
| ① | App | running 轮询分支**无 deadline / 无 maxAttempts**；只要 `tasks.get` 一直回 `running` 就永远轮询 | `app_controller_desktop_thread_actions.dart:788-794`（对比 `:831` 的 artifact-sync 分支**有**封顶 `openClawArtifactSyncLimitReachedInternal`） |
|
||||
| ② | App | 进度条 phase 仅由 `pending` 驱动，而 `pending` 被该 loop 独占持有，loop 不退出 = 永远 `任务运行中...` | `assistant_task_progress_bar.dart:190`；`pending` 源 = `aiGatewayPendingSessionKeysInternal` |
|
||||
| ③ | App | `停止`（abortRun → cancelAssistantTaskForSessionInternal）只向 gateway 发 `tasks.cancel`，**从不本地清 `pending`**；gateway 不可达 / run 已丢时为空操作 → 永远停不下 | `app_controller_desktop_thread_actions.dart:1793-1809` |
|
||||
| ④ | **入口 / Bridge 超时错配** | Caddy `/acp*` `read/write_timeout 30m`，但 bridge `openClawAgentWaitMaxTimeout = 60min`。**>30min 的 complex_chain_task，SSE 在入口被掐断**，而 gateway 仍在跑 → `ACP_HTTP_CONNECTION_CLOSED` | caddy.j2 `read_timeout 30m` vs `orchestrator.go:32` |
|
||||
| ⑤ | 入口 | 仅 `/acp*` 有 `flush_interval -1` 与长超时；`/api*`、`/` 用 Caddy 默认（短超时、无即时 flush），任何走这两条的流式 / 轮询都更脆 | caddy.j2 |
|
||||
| ⑥ | **Bridge↔Gateway WS 关联绑定连接** | WS 一抖，`onConnLost` 把**所有在途请求**立即判 `SOCKET_CLOSED`——包括 OpenClaw 仍在执行的那个 run；重连后无任何 pending 复关联 | `gatewayruntime/runtime.go:802-823`、`:935-938`（`takePendingLocked` 清空） |
|
||||
| ⑦ | **Bridge 无持久 run 仓** | 同步 SSE 路径不落 `xworkmate.jobs.*`，结果只活在内存 `responseCh`；连接一断即丢（已存在 jobs.submit/get/list 未被 gateway submit 复用） | `rpc_handler.go:73`（jobs.*）vs http_handler 同步路径 |
|
||||
| ⑧ | Gateway 重连丢 run 态 | 重连到新 WS 后 `tasks.get` 经 `ensureProductionGatewayConnected` 落到新连接，查不到 terminal → 回 stale `running` 或 `not_found` | `rpc_handler.go:143-175` |
|
||||
| ⑨ | Bridge | `gatewayRPCError` 把可重试错误统一映射为 `OPENCLAW_GATEWAY_SOCKET_CLOSED`，但缺少"run 仍在后台、稍后可查"的语义，客户端只能当硬失败 | `orchestrator.go:1678-1702` |
|
||||
|
||||
---
|
||||
|
||||
## 4. 根因链（Root cause）
|
||||
|
||||
> "实际 gateway 已执行完毕，但任务永远停不下" =
|
||||
>
|
||||
> **服务端把已完成的结果弄丢（⑥ + ⑦ + ⑧）** ：bridge↔gateway 的请求关联随 WS 连接销毁、且无持久副本，run 完成事件投递到已被放弃的 channel；重连后无法复关联。
|
||||
>
|
||||
> **+ 客户端把"无限 running"固化（① + ② + ③）** ：轮询无截止、进度条只看 pending、停止不本地生效。
|
||||
>
|
||||
> **+ 触发器（④ / ⑤）** ：长任务在入口 30min 处必断，把链路推向上述失效（尤其 complex_chain_task budget=60min）。
|
||||
|
||||
最小可复现路径：
|
||||
1. 提交一个 budget > 30min 或网络抖动概率高的 gateway 任务；
|
||||
2. SSE 在入口（30min）或 WS（瞬断）处断开 → App 收 `ACP_HTTP_CONNECTION_CLOSED`；
|
||||
3. OpenClaw 后台继续跑完，但 bridge 已丢失该 run 的 pending / 无持久记录；
|
||||
4. App 落入 running 轮询（或失败后仍 pending），`tasks.get` 拿不到 terminal；
|
||||
5. 进度条永远 `任务运行中...`，`停止` 不本地生效 → 卡死。
|
||||
|
||||
---
|
||||
|
||||
## 5. 编码改进规划（Stability / Robustness TODO）
|
||||
|
||||
> 排序原则：先**当天可上、零协议变更**的止血项（L0 配置 + L2 客户端），再**治本**的服务端持久化（L1），最后横切可观测性（L3）。
|
||||
> 每项标注：所属仓库 · 改动面 · 验收要点。
|
||||
|
||||
### L0 入口配置（ai-workspace-infra · 改配置即可 · 当天可上）
|
||||
|
||||
- [ ] **T1 对齐入口与 bridge 的超时上限**
|
||||
`caddy.j2` `/acp*` 的 `read_timeout` / `write_timeout` 提到 **≥ bridge `openClawAgentWaitMaxTimeout`（60min）+ 余量**。
|
||||
最好让入口超时与 `orchestrator.go` 的上限**由同一变量 / 同源常量**渲染，避免再次漂移。
|
||||
验收：budget=60min 的 complex_chain_task 全程 SSE 不被入口掐断。
|
||||
|
||||
- [ ] **T2 收敛 / 补齐非 /acp 路由的流式配置**
|
||||
给 `/api*`（及任何承载 `tasks.get` 轮询、流式响应的路由）补 `flush_interval -1` + 同样长超时；或显式把所有 gateway 流量收敛到 `/acp*`。
|
||||
验收：轮询 / 流式不再依赖 Caddy 默认短超时。
|
||||
|
||||
### L2 App 客户端止血（ai-workspace-lab/xworkmate-app · 纯客户端 · 零协议变更）
|
||||
|
||||
- [ ] **T3 running 轮询加硬截止**
|
||||
在 `pollOpenClawTaskAssociationInternal` 的 running-handle 分支（`thread_actions.dart:788`）引入 deadline / maxAttempts，复用 bridge 下发的 `deadlineAt`（`openclaw_async_tasks.go:92`）+ grace；到点落 `interrupted`（可恢复态，`isRecoverableAssistantTaskStateInternal` 已支持），退出 loop。
|
||||
参照同函数 `:831` artifact-sync 分支的封顶写法 `openClawArtifactSyncLimitReachedInternal`。
|
||||
验收：gateway 始终回 `running` 时，客户端在 deadline 后必终止，不再无限轮询。
|
||||
|
||||
- [ ] **T4 `停止` 本地权威化**
|
||||
`abortRun` / `cancelAssistantTaskForSessionInternal`（`:1793`）改为：**先**乐观清 `aiGatewayPendingSessionKeysInternal`、置 lifecycle=`aborted`、退出 loop，**再** best-effort 发 `tasks.cancel`。UI 终止不得依赖 gateway 往返。
|
||||
验收：gateway 不可达时点 `停止` 仍能立刻停下。
|
||||
|
||||
- [ ] **T5 传输中断降级为"后台续跑·重连中"**
|
||||
收到实时 `ACP_HTTP_CONNECTION_CLOSED` 时，不直接当硬失败把任务留在 running，而是降级为"已转后台 / 重连中"，触发 `resumeOpenClawTaskAssociationsInternal`（`:906`，目前仅在 thread 加载时触发）续轮询。
|
||||
验收：SSE 瞬断后，任务能自动恢复轮询并最终拿到终态或明确终止。
|
||||
|
||||
- [ ] **T6 失败路径与 pending 清理一致性**
|
||||
审计 `applyGatewayChatFailureInternal`（`:1613`，置 `ready` 但不清 pending）与调用方 `finally`（`:715` 仅 `!handedOffToBridgeTask` 才清 pending）之间的竞态，确保任一终态路径都能确定性地清 pending，杜绝"错误已渲染但仍 running"。
|
||||
验收：失败 / 中断 / 取消三类终态后，`pending` 必为 false。
|
||||
|
||||
### L1 Bridge 持久化（ai-workspace-lab/xworkmate-bridge · 根因修复 · 有协议/状态面改动）
|
||||
|
||||
- [ ] **T7 run 关联与 WS 连接解耦**
|
||||
维护 `runId → {status, result, artifacts, deadline}` 的持久 / 独立登记表，由 gateway notification 更新。`onConnLost`（`runtime.go:802`）时**不要把长任务 pending 直接判死**，而是标记 `detached`，重连后用 `runId` 复关联 / 回放完成事件。
|
||||
验收：WS 瞬断 + 重连后，已完成的 run 仍能被 `tasks.get` 查到 terminal + artifacts。
|
||||
|
||||
- [ ] **T8 OpenClaw submit 接入 `xworkmate.jobs.*` 持久存储**
|
||||
让 gateway submit 落 `xworkmate.jobs.*`（`rpc_handler.go:73` 已有 submit/get/list/stats），使结果不再只活在内存 `responseCh`。
|
||||
验收：bridge 重启 / 连接抖动后，结果与 artifacts 仍可检索。
|
||||
|
||||
- [ ] **T9 服务端 DeadlineAt 兜底终态**
|
||||
run 过期且 gateway 无法确认时，bridge 主动回 terminal `interrupted`（而非无限 `running`），给客户端确定终态。与 T3 客户端 deadline 形成双保险。
|
||||
验收：gateway 失联超过 budget 后，`tasks.get` 返回确定 terminal。
|
||||
|
||||
- [ ] **T10 错误语义细化**
|
||||
`gatewayRPCError`（`orchestrator.go:1678`）区分"连接断但 run 仍在后台可查" vs "run 确实失败"，前者携带 `runId` + `retryable/poll` 提示，供客户端走 T5 续轮询而非硬失败。
|
||||
验收：客户端能据错误语义区分"重连续跑"与"真失败"。
|
||||
|
||||
### L3 可观测性（横切 · infra/service/lab）
|
||||
|
||||
- [ ] **T11 端到端贯穿 runId**
|
||||
App 日志 → Caddy access log → bridge SSE 日志（已有 `component=acp_sse`，`http_handler.go:221`）→ gateway run，全链路带同一 `runId`，便于定位"入口断"还是"WS 断"。
|
||||
验收：任一 `runId` 可在四层日志串联。
|
||||
|
||||
- [ ] **T12 关键指标 + 告警**
|
||||
bridge 暴露：`SOCKET_CLOSED 在途任务数`、gateway WS 重连计数、running 轮询超 deadline 计数。
|
||||
验收：⑥ 类事件发生即在监控可见，无需靠用户截图。
|
||||
|
||||
---
|
||||
|
||||
## 6. 落地顺序建议
|
||||
|
||||
1. **当天止血**：T1 + T2（入口配置）+ T3 + T4 + T6（客户端），消除"30min 必断 / 路由漏配 / 无限 running / 停不掉"。
|
||||
2. **本周治本**：T7 + T8 + T9（bridge 持久化与解耦），让"gateway 跑完 = 结果一定拿得到"。
|
||||
3. **跟进**：T5 + T10（断连续跑语义）、T11 + T12（可观测性）。
|
||||
|
||||
> 回归对照：本目录 `00-review-env-and-matrix.md` 第 2 节"通用验收标准"中"长任务执行期间状态流 / 取消 / 重试稳定""同一任务重复执行 3 次不卡死"，即本规划的回归出口。
|
||||
@ -51,6 +51,20 @@ const Duration kOpenClawArtifactSyncMaxDuration = Duration(seconds: 90);
|
||||
const String kOpenClawArtifactSyncTimeoutCode =
|
||||
'OPENCLAW_ARTIFACT_SYNC_TIMEOUT';
|
||||
|
||||
// running 轮询（OpenClaw run handle）兜底截止（T3）：
|
||||
// 当 gateway 始终回 running（例如 bridge↔gateway socket 抖动后 run 态丢失）时，防止客户端无限轮询、
|
||||
// 进度条永远卡在「任务运行中...」。预算按 taskLoadClass 估算，与 bridge 侧任务预算对齐
|
||||
// (xworkmate-bridge/internal/acp/openclaw_async_tasks.go: short=10/long=30/complex=60 min)，再加 grace。
|
||||
const Duration kOpenClawRunningPollGrace = Duration(minutes: 5);
|
||||
const Duration kOpenClawRunningPollDefaultBudget = Duration(minutes: 30);
|
||||
const Map<String, Duration> kOpenClawRunningPollBudgets = <String, Duration>{
|
||||
'short_task': Duration(minutes: 10),
|
||||
'long_task': Duration(minutes: 30),
|
||||
'complex_chain_task': Duration(minutes: 60),
|
||||
'complex_long_chain_task': Duration(minutes: 60),
|
||||
};
|
||||
const String kOpenClawRunningPollTimeoutCode = 'OPENCLAW_RUN_POLL_TIMEOUT';
|
||||
|
||||
bool openClawArtifactPathHasRequiredExtension(String path, String extension) {
|
||||
final normalizedPath = path.trim().toLowerCase();
|
||||
final normalizedExtension = extension.trim().toLowerCase().replaceFirst(
|
||||
@ -977,6 +991,74 @@ extension AppControllerDesktopRuntimeHelpers on AppController {
|
||||
);
|
||||
}
|
||||
|
||||
// T3: running 轮询是否已越过兜底截止。
|
||||
// 预算从 run 开始时间(startedAtMs)起算；startedAtMs 缺失时退化为本轮首次轮询时间(firstPollAtMs)。
|
||||
bool openClawRunningPollDeadlineReachedInternal({
|
||||
required double? startedAtMs,
|
||||
required String taskLoadClass,
|
||||
required double? firstPollAtMs,
|
||||
double? nowMs,
|
||||
}) {
|
||||
final anchorMs = (startedAtMs != null && startedAtMs > 0)
|
||||
? startedAtMs
|
||||
: firstPollAtMs;
|
||||
if (anchorMs == null || anchorMs <= 0) {
|
||||
return false;
|
||||
}
|
||||
final budget =
|
||||
kOpenClawRunningPollBudgets[taskLoadClass.trim().toLowerCase()] ??
|
||||
kOpenClawRunningPollDefaultBudget;
|
||||
final limitMs =
|
||||
(budget + kOpenClawRunningPollGrace).inMilliseconds.toDouble();
|
||||
final currentMs = nowMs ?? DateTime.now().millisecondsSinceEpoch.toDouble();
|
||||
return currentMs - anchorMs >= limitMs;
|
||||
}
|
||||
|
||||
// T3: running 轮询越过兜底截止时，落到「可恢复的中断」终态并退出轮询。
|
||||
// 注意：服务端可能其实已经跑完，只是结果回传链路断了；因此提示用户可重发以拿回结果。
|
||||
void markOpenClawRunningPollTimeoutInternal({
|
||||
required String sessionKey,
|
||||
required OpenClawTaskAssociation association,
|
||||
}) {
|
||||
final normalizedSessionKey = normalizedAssistantSessionKeyInternal(
|
||||
sessionKey,
|
||||
);
|
||||
final nowMs = DateTime.now().millisecondsSinceEpoch.toDouble();
|
||||
aiGatewayPendingSessionKeysInternal.remove(normalizedSessionKey);
|
||||
clearAiGatewayStreamingTextInternal(normalizedSessionKey);
|
||||
upsertTaskThreadInternal(
|
||||
normalizedSessionKey,
|
||||
lifecycleStatus: 'interrupted',
|
||||
lastResultCode: kOpenClawRunningPollTimeoutCode,
|
||||
lastRunAtMs: nowMs,
|
||||
lastArtifactSyncAtMs: nowMs,
|
||||
lastArtifactSyncStatus: 'interrupted',
|
||||
clearOpenClawTaskAssociation: true,
|
||||
updatedAtMs: nowMs,
|
||||
);
|
||||
appendLocalSessionMessageInternal(
|
||||
normalizedSessionKey,
|
||||
GatewayChatMessage(
|
||||
id: nextLocalMessageIdInternal(),
|
||||
role: 'assistant',
|
||||
text: appText(
|
||||
'任务等待已超过预算上限，已结束本轮等待。任务可能已在后台完成但结果回传中断，请重新发送请求以拿回结果。错误码：OPENCLAW_RUN_POLL_TIMEOUT',
|
||||
'Waiting for the task exceeded its budget, so this round was ended. The task may have finished in the background but its result could not be delivered. Send the request again to retrieve the result. Error code: OPENCLAW_RUN_POLL_TIMEOUT',
|
||||
),
|
||||
timestampMs: nowMs,
|
||||
toolCallId: null,
|
||||
toolName: null,
|
||||
stopReason: null,
|
||||
pending: false,
|
||||
error: true,
|
||||
),
|
||||
persistInThreadContext: true,
|
||||
);
|
||||
recomputeTasksInternal();
|
||||
notifyIfActiveInternal();
|
||||
unawaited(flushAssistantThreadPersistenceInternal());
|
||||
}
|
||||
|
||||
bool openClawArtifactSyncLimitReachedInternal({
|
||||
required int attemptCount,
|
||||
required double? firstSyncAtMs,
|
||||
|
||||
@ -753,6 +753,8 @@ extension AppControllerDesktopThreadActions on AppController {
|
||||
var firstAttempt = true;
|
||||
var artifactSyncAttempts = 0;
|
||||
double? artifactSyncStartedAtMs;
|
||||
// T3: running 轮询兜底截止的起算锚点（首次进入 running 轮询的时间）。
|
||||
double? runningPollFirstAtMs;
|
||||
final existingThread = taskThreadForSessionInternal(sessionKey);
|
||||
if (association.status.trim().toLowerCase() == 'syncing-artifacts') {
|
||||
artifactSyncStartedAtMs = existingThread?.lastArtifactSyncAtMs;
|
||||
@ -786,6 +788,21 @@ extension AppControllerDesktopThreadActions on AppController {
|
||||
);
|
||||
current = nextAssociation;
|
||||
if (result.isOpenClawRunningTaskHandle) {
|
||||
// T3: 给 running 轮询加兜底截止，避免 gateway 始终回 running 时无限轮询、永远卡「任务运行中...」。
|
||||
final nowMs = DateTime.now().millisecondsSinceEpoch.toDouble();
|
||||
runningPollFirstAtMs ??= nowMs;
|
||||
if (openClawRunningPollDeadlineReachedInternal(
|
||||
startedAtMs: current.startedAtMs,
|
||||
taskLoadClass: current.taskLoadClass,
|
||||
firstPollAtMs: runningPollFirstAtMs,
|
||||
nowMs: nowMs,
|
||||
)) {
|
||||
markOpenClawRunningPollTimeoutInternal(
|
||||
sessionKey: sessionKey,
|
||||
association: current,
|
||||
);
|
||||
return;
|
||||
}
|
||||
persistOpenClawTaskAssociationInternal(
|
||||
sessionKey: sessionKey,
|
||||
association: nextAssociation,
|
||||
@ -1615,6 +1632,14 @@ extension AppControllerDesktopThreadActions on AppController {
|
||||
required AssistantExecutionTarget target,
|
||||
required Object error,
|
||||
}) async {
|
||||
// T6: 终态一致性。失败处理器是本轮的终态出口，必须自身权威地清 pending，
|
||||
// 不能依赖各调用方的 finally（runOpenClawGatewayQueuedTurnInternal 的 finally 此前并不清 pending，
|
||||
// 导致「错误已渲染但进度条仍卡在任务运行中」）。对其它已显式 remove 的调用方是幂等的。
|
||||
// pending 集合在不同路径下分别用原始 / 归一化 key，两种都移除以确保彻底清空。
|
||||
aiGatewayPendingSessionKeysInternal.remove(sessionKey);
|
||||
aiGatewayPendingSessionKeysInternal.remove(
|
||||
normalizedAssistantSessionKeyInternal(sessionKey),
|
||||
);
|
||||
clearAiGatewayStreamingTextInternal(sessionKey);
|
||||
clearPendingToolCallsForGatewaySessionInternal(sessionKey, hasError: true);
|
||||
final completedAtMs = DateTime.now().millisecondsSinceEpoch.toDouble();
|
||||
@ -1781,27 +1806,47 @@ extension AppControllerDesktopThreadActions on AppController {
|
||||
return;
|
||||
}
|
||||
if (aiGatewayPendingSessionKeysInternal.contains(sessionKey)) {
|
||||
await cancelAssistantTaskForSessionInternal(sessionKey);
|
||||
// T4: 停止本地权威化。先捕获 association，再立刻把本轮标记为 aborted（清 pending / 置终态 /
|
||||
// 让轮询 loop 在下一次迭代退出），UI 立即停止——不依赖 gateway 往返。
|
||||
// 随后 best-effort 下发 tasks.cancel；其失败/挂起都不得阻塞或回滚本地终止。
|
||||
final association = taskThreadForSessionInternal(
|
||||
sessionKey,
|
||||
)?.openClawTaskAssociation;
|
||||
removeQueuedOpenClawGatewayTurnsForSessionInternal(sessionKey);
|
||||
removeActiveOpenClawGatewayTurnsForSessionInternal(sessionKey);
|
||||
markOpenClawGatewayTurnAbortedInternal(sessionKey);
|
||||
drainOpenClawGatewayQueueInternal();
|
||||
unawaited(
|
||||
cancelAssistantTaskForSessionInternal(
|
||||
sessionKey,
|
||||
association: association,
|
||||
),
|
||||
);
|
||||
return;
|
||||
}
|
||||
}
|
||||
|
||||
Future<void> cancelAssistantTaskForSessionInternal(String sessionKey) async {
|
||||
Future<void> cancelAssistantTaskForSessionInternal(
|
||||
String sessionKey, {
|
||||
OpenClawTaskAssociation? association,
|
||||
}) async {
|
||||
final normalized = normalizedAssistantSessionKeyInternal(sessionKey);
|
||||
final association = taskThreadForSessionInternal(
|
||||
normalized,
|
||||
)?.openClawTaskAssociation;
|
||||
await goTaskServiceClientInternal.cancelTask(
|
||||
route: GoTaskServiceRoute.externalAcpSingle,
|
||||
target: assistantExecutionTargetForSession(normalized),
|
||||
sessionId: normalized,
|
||||
threadId: normalized,
|
||||
association: association,
|
||||
);
|
||||
// association 可能已被本地终止清空，故允许调用方预先捕获后传入。
|
||||
final effectiveAssociation =
|
||||
association ??
|
||||
taskThreadForSessionInternal(normalized)?.openClawTaskAssociation;
|
||||
// best-effort：gateway 不可达 / run 已不存在时不得抛出，避免阻塞停止流程。
|
||||
try {
|
||||
await goTaskServiceClientInternal.cancelTask(
|
||||
route: GoTaskServiceRoute.externalAcpSingle,
|
||||
target: assistantExecutionTargetForSession(normalized),
|
||||
sessionId: normalized,
|
||||
threadId: normalized,
|
||||
association: effectiveAssociation,
|
||||
);
|
||||
} catch (error) {
|
||||
debugPrint('cancelAssistantTaskForSession best-effort failed: $error');
|
||||
}
|
||||
}
|
||||
|
||||
Future<void> prepareForExit() async {
|
||||
|
||||
Loading…
Reference in New Issue
Block a user