worker: prestart stage logging for dispatch stalls

Claim-to-dispatch stalls (cf. 2026-10-29 task A: 61s to interrupted
with zero detail) were invisible because docker run blocks with
captured output and watchdog kills were silent. Add three layers:
- worker_exec_start/done/error audit rows in _run_attempt (PG,
  survives --rm), start/done pairing shows whether an attempt died
  inside the executor call;
- docker_run_start/done/error log lines in DockerExecutor (container
  + image only, never the inner command/prompt);
- stdout tail (last 2000 chars) to the worker log on failed or
  interrupted results, and a log line on both watchdog kill paths.
Tests: cloud/tests/test_prestart_stages.py, 6 cases incl. a
no-prompt-in-logs assertion.

cloud/worker.py

@@ -1983,6 +1983,11 @@ class Worker:
                 kill = getattr(self.executor, "kill_container", None)
                 if kill is not None and container_name:
                     kill(container_name)
+                    log_event(self.logger, "worker_docker_killed",
+                              task_id=task_id,
+                              message="cancel watchdog killed container "
+                                      "%s (task interrupted/cancelled)"
+                                      % container_name)
             except Exception:
                 pass
             return
@@ -2018,6 +2023,10 @@ class Worker:
                 kill = getattr(self.executor, "kill_container", None)
                 if kill is not None and container_name:
                     kill(container_name)
+                    log_event(self.logger, "worker_docker_killed",
+                              task_id=task_id,
+                              message="quota watchdog killed container %s "
+                                      "(quota exceeded)" % container_name)
             except Exception:
                 pass
             return
@@ -3534,6 +3543,33 @@ class Worker:
                               _watchdog_stop, quota_check),
                         daemon=True)
                     _quota_watchdog_thread.start()
+        # 启动前分段日志（2026-10-29 任务 A）：认领后 61 秒中断，
+        # 但阻塞的 docker run 中途没有任何标记、容器 --rm 后输出
+        # 也没了，事后完全无法定位卡在哪一段。派发开始/结束进
+        # 审计（PG 落盘，容器没了也丢不了），下次卡死时最后一个
+        # 标记直接说明卡在哪一段；kind/容器名/超时不含 prompt/key。
+        _exec_kind = ("goal-resume-breakpoint"
+                      if task.get("_goal_resume_breakpoint")
+                      else "goal-resume" if task.get("_resume")
+                      else "goal" if goal_mode else "dispatch")
+        _exec_container = None
+        if hasattr(self.executor, "container_name"):
+            try:
+                _exec_container = self.executor.container_name(task)
+            except Exception:
+                _exec_container = None
+        _exec_t0 = time.time()
+        _exec_error = None
+        _exec_detail = "kind=%s executor=%s timeout=%s" % (
+            _exec_kind, self._executor_name(),
+            "None" if _effective_task_timeout is None
+            else _effective_task_timeout)
+        if _exec_container:
+            _exec_detail += " container=%s" % _exec_container
+        self._audit(task.get("project_id"), "worker_exec_start",
+                    _exec_detail)
+        log_event(self.logger, "worker_exec_start", task_id=task_id,
+                  message="worker exec start: %s" % _exec_detail)
         try:
             gateway_provider, gateway_env = self._ensure_gateway(task)
             if self._use_popen():
@@ -3726,6 +3762,7 @@ class Worker:
                       "status": "failed", "stdout": "", "returncode": None,
                       "host_status": None, "stderr": ""}
             error = str(exc)
+            _exec_error = exc
         finally:
             if _watchdog_stop is not None:
                 _watchdog_stop.set()
@@ -3753,6 +3790,34 @@ class Worker:
             entry = self._running_snapshot().get(task_id) or {}
             if entry.get("proc") is None:
                 self._running_clear(task_id)
+        # 启动前分段日志（续）：派发结束标记与失败输出尾巴。结束
+        # 标记与开始标记配对：只有开始没有结束 = 卡在执行器调用
+        # 里；失败/中断时的容器输出尾巴记进 worker 日志——任务 A
+        # 当时这部分输出随 --rm 消失，是事后无法定位的另一半。
+        _exec_elapsed_ms = int((time.time() - _exec_t0) * 1000)
+        if _exec_error is not None:
+            self._audit(task.get("project_id"), "worker_exec_error",
+                        "elapsed_ms=%d error=%s"
+                        % (_exec_elapsed_ms,
+                           _exec_error.__class__.__name__))
+            log_event(self.logger, "worker_exec_error", task_id=task_id,
+                      message="worker exec error: elapsed_ms=%d error=%s"
+                              % (_exec_elapsed_ms,
+                                 _exec_error.__class__.__name__))
+        else:
+            self._audit(task.get("project_id"), "worker_exec_done",
+                        "elapsed_ms=%d status=%s returncode=%s"
+                        % (_exec_elapsed_ms, result.get("status"),
+                           result.get("returncode")))
+            log_event(self.logger, "worker_exec_done", task_id=task_id,
+                      status=result.get("status"),
+                      message="worker exec done: elapsed_ms=%d status=%s"
+                              % (_exec_elapsed_ms, result.get("status")))
+        _exec_out = result.get("stdout", "") or ""
+        if result.get("status") in ("failed", "interrupted") and _exec_out:
+            self.logger.info(
+                "worker_exec_stdout_tail task_id=%s tail=%s",
+                task_id, _exec_out[-2000:])
         # Docker 超时标记：执行器已在超时时 kill+rm 容器并返回
         # failed（stdout 含 timeout），这里把状态统一成 timeout，
         # 调用方据此置 failed 并记 worker_timeout 审计。