fix: bound task failures and release stalled workers

artemis/runtime/daemon_client.py

@@ -221,6 +221,7 @@ def submit_task_to_daemon(
     conversation_id: str | None = None,
     verification_level: str | None = None,
     explorer_mode: str | None = None,
+    runtime_overrides: dict[str, Any] | None = None,
     base_url: str | None = None,
     timeout: float = 15.0,
 ) -> dict[str, Any] | None:
@@ -246,6 +247,7 @@ def submit_task_to_daemon(
         "session_id": session_id,
         "ingress": ingress,
         "conversation_id": conversation_id,
+        "runtime_overrides": runtime_overrides,
     }
 
     try:

artemis/runtime/task_limits.py

@@ -0,0 +1,30 @@
+"""Execution budgets shared by supervised web and MCP tasks."""
+
+from dataclasses import dataclass, field
+import time
+
+from artemis.config.settings import settings
+
+
+def task_timeout(item: dict, job: dict | None = None, batch: dict | None = None) -> int:
+    options = item.get("runtime_overrides") or (job or {}).get("case", {}).get("task_config") or {}
+    return int(
+        options.get("timeout_seconds")
+        or (batch or {}).get("timeout_seconds")
+        or settings.ARTEMIS_TASK_TIMEOUT_SECONDS
+    )
+
+
+@dataclass
+class ExecutionBudget:
+    limit: float
+    elapsed: float = 0
+    _last: float = field(default_factory=time.monotonic)
+    _excluded: bool = False
+
+    def update(self, *, excluded: bool = False) -> float:
+        now = time.monotonic()
+        if not self._excluded:
+            self.elapsed += max(0, now - self._last)
+        self._last, self._excluded = now, excluded
+        return max(0, self.limit - self.elapsed)

artemis/runtime/task_result.py

@@ -0,0 +1,64 @@
+"""Persist useful failure evidence even when a worker must be terminated."""
+
+import json
+import os
+from pathlib import Path
+import uuid
+
+from artemis.config import TRACES_PATH
+
+
+def read_worker_outcome(session_id: str) -> dict:
+    """Read a worker's terminal signal without waiting for hung SDK threads."""
+    try:
+        data = json.loads(
+            (TRACES_PATH / session_id / "run_outcome.json").read_text(encoding="utf-8")
+        )
+        if (
+            isinstance(data, dict)
+            and data.get("worker_finalized") is True
+            and data.get("task_status")
+            in {
+                "failed",
+                "completed",
+                "cancelled",
+                "interrupted",
+            }
+        ):
+            return data
+    except (OSError, ValueError):
+        pass  # Atomic outcome is absent until the task finalizes.
+    return {}
+
+
+def save_failure_result(session_id: str, error: str, *, code: str = "execution_failed") -> None:
+    if not session_id or not all(c.isalnum() or c in "-_" for c in session_id):
+        raise ValueError("无效的任务编号")
+    directory = TRACES_PATH / session_id
+    directory.mkdir(parents=True, exist_ok=True)
+    payload: dict = {}
+    try:
+        previous = json.loads((directory / "result.json").read_text(encoding="utf-8"))
+        if isinstance(previous, dict):
+            payload.update(previous)
+    except (OSError, ValueError):
+        pass  # Missing/truncated results are expected after a killed worker.
+    payload.update(
+        {
+            "status": "failed",
+            "task_status": "failed",
+            "explanation": error,
+            "error": error,
+            "code": code,
+            "worker_finalized": True,
+            "tests": payload.get("tests") or {},
+        }
+    )
+    for name in ("result.json", "run_outcome.json"):
+        target = directory / name
+        temporary = target.with_name(f".{name}.{uuid.uuid4().hex}.tmp")
+        try:
+            temporary.write_text(json.dumps(payload, ensure_ascii=False), encoding="utf-8")
+            os.replace(temporary, target)
+        finally:
+            temporary.unlink(missing_ok=True)