auto-sync: 2026-06-01 00:15:29

src/daemon/spawner.py

@@ -430,6 +430,7 @@ curl -X POST http://{self.api_host}:{self.api_port}/api/projects/{project_id}/ta
         task_db_path: Optional[Path] = None,
         reuse_session_id: Optional[str] = None,
         on_checks_passed: Optional[Any] = None,
+        skip_counter: bool = False,
     ) -> str:
         """Spawn Full Agent(异步非阻塞)
 
@@ -484,7 +485,8 @@ curl -X POST http://{self.api_host}:{self.api_port}/api/projects/{project_id}/ta
                 raise AgentBusyError(f"{agent_id}: compacting")
 
         # 3. counter acquire(per session key 粒度)
-        if self.counter:
+        # v2.8.1 Bug-4 fix: retry 时跳过 counter（counter 从原始 spawn 保持到 retry 完成）
+        if self.counter and not skip_counter:
             if not await self.counter.can_acquire(agent_id, _sid_key):
                 raise AgentBusyError(agent_id)
             await self.counter.acquire(agent_id, _sid_key)
@@ -880,11 +882,9 @@ curl -X POST http://{api_host}:{api_port}/api/projects/{project_id}/tasks/{task_
         需要手动 release counter,然后 spawn_full_agent 内部会 acquire。
         on_complete(含 counter release)置为 None,避免 double release。
         """
-        # ── 关键:手动 release counter(进程退出 = agent 空闲)──
-        if self.counter:
-            self.counter.release(agent_id, session_id or "main")
-        # 旧 wrapped_on_complete 含 counter.release,不再使用,防止 double release
-        on_complete = None
+        # v2.8.1 Bug-4 fix: 不再手动 release counter + 置 None on_complete
+        # counter 从原始 spawn 保持到 retry 完成，避免窗口期 ticker acquire 同一 agent
+        # on_complete 保留原始 wrapped_on_complete，retry 完成后自然 release counter
 
         # 续杯前检查任务状态,已终态则跳过
         if db_path and task_id:
@@ -894,7 +894,7 @@ curl -X POST http://{api_host}:{api_port}/api/projects/{project_id}/tasks/{task_
                     row = conn.execute(
                         "SELECT status FROM tasks WHERE id=?", (task_id,)
                     ).fetchone()
-                    if row and row["status"] in ("done", "failed", "cancelled", "review", "pending"):
+                    if row and row["status"] in ("done", "failed", "cancelled", "review")  # Bug-6 fix: pending 不是终态:
                         logger.info("Retry skip: task %s already %s (agent=%s)",
                                     task_id, row["status"], agent_id)
                         # on_complete = wrapped_on_complete,会 release counter
@@ -984,6 +984,7 @@ curl -X POST http://{api_host}:{api_port}/api/projects/{project_id}/tasks/{task_
                 on_complete=on_complete,
                 use_main_session=True,  # #02: 续杯走 main session
                 task_db_path=db_path,
+                skip_counter=True,  # Bug-4 fix: counter 已在原始 spawn 中持有
             )
         except AgentBusyError:
             # agent 被其他任务占用(不应发生,但防御)