fix: preserve live tasks during daemon recovery

tests/unit/admin_console/test_automation.py

@@ -300,6 +300,116 @@ def test_restart_never_repeats_started_work(isolated):
     assert isolated.jobs(retry["id"])[0]["session_id"] != job["session_id"]
 
 
+@pytest.mark.parametrize("liveness_source", ["pid", "device_lock"])
+def test_recovery_preserves_live_runner_and_batch(isolated, monkeypatch, liveness_source):
+    import os
+    from types import SimpleNamespace
+    from artemis.runtime import DeviceExecutionLock
+
+    batch = service.create_batch(payload())
+    service.batch_action(batch["id"], "resume")
+    job = isolated.jobs(batch["id"])[0]
+    isolated.save_job({**job, "status": "running"})
+    if liveness_source == "pid":
+        with sqlite3.connect(isolated.path) as conn:
+            conn.execute(
+                "INSERT INTO sessions(session_id,initial_goal,start_time,status,device_info,pid) "
+                "VALUES (?,?,?,?,?,?)",
+                (job["session_id"], "goal", 1, "running", "{}", os.getpid()),
+            )
+    monkeypatch.setattr(
+        DeviceExecutionLock,
+        "get_active_owners",
+        lambda: (
+            {"device": SimpleNamespace(session_id=job["session_id"])}
+            if liveness_source == "device_lock"
+            else {}
+        ),
+    )
+    service.recover_batches()
+    assert isolated.job(job["session_id"])["status"] == "running"
+    assert not isolated.job(job["session_id"]).get("error")
+    assert isolated.get("batch", batch["id"])["status"] == "running"
+
+
+@pytest.mark.asyncio
+async def test_natural_completion_resolves_old_recovery_error_in_job_report_and_history(
+    client, isolated
+):
+    job = isolated.jobs(service.create_batch(payload())["id"])[0]
+    message = "服务中断；未完成任务需手动重跑"
+    isolated.save_job({**job, "status": "interrupted", "error": message})
+    await service.on_terminal({"session_id": job["session_id"]}, "completed")
+    saved = isolated.job(job["session_id"])
+    assert saved["status"] == "completed"
+    assert not saved.get("error")
+    assert saved["recovery_events"][0]["message"] == message
+    assert reports.session_report(job["session_id"])["error"] == ""
+    assert not client.get("/api/automation/history").json()["items"][0].get("error")
+
+
+def test_recovery_repairs_completed_legacy_warning_but_preserves_real_failure(isolated):
+    batch = service.create_batch(payload())
+    first, second = isolated.jobs(batch["id"])
+    isolated.save_job({**first, "status": "completed", "error": "服务中断，需手动重跑"})
+    isolated.save_job({**second, "status": "failed", "error": "任务执行超时（30 秒）"})
+    assert reports.session_report(first["session_id"])["error"] == ""
+    service.recover_batches()
+    assert not isolated.job(first["session_id"]).get("error")
+    assert "服务中断" in isolated.job(first["session_id"])["recovery_events"][0]["message"]
+    assert "超时" in isolated.job(second["session_id"])["error"]
+
+
+@pytest.mark.asyncio
+async def test_surviving_worker_is_finalized_when_it_exits_without_resubmission(
+    isolated, monkeypatch
+):
+    import os
+    from artemis.runtime import DeviceExecutionLock
+
+    batch = service.create_batch(payload())
+    service.batch_action(batch["id"], "resume")
+    job = isolated.jobs(batch["id"])[0]
+    isolated.save_job({**job, "status": "running"})
+    monkeypatch.setattr(DeviceExecutionLock, "get_active_owners", lambda: {})
+    with sqlite3.connect(isolated.path) as conn:
+        conn.execute(
+            "INSERT INTO sessions(session_id,initial_goal,start_time,status,device_info,pid) "
+            "VALUES (?,?,?,?,?,?)",
+            (job["session_id"], "goal", time.time(), "running", "{}", os.getpid()),
+        )
+    service.recover_batches()
+    await service.reconcile_recovered_jobs()
+    assert isolated.job(job["session_id"])["status"] == "running"
+    with sqlite3.connect(isolated.path) as conn:
+        conn.execute(
+            "UPDATE sessions SET pid=NULL,status='completed' WHERE session_id=?",
+            (job["session_id"],),
+        )
+    await service.reconcile_recovered_jobs()
+    saved = isolated.job(job["session_id"])
+    assert saved["status"] == "completed"
+    assert saved["report_ready"]
+    assert not saved.get("recovery_waiting")
+    assert not saved.get("error")
+    assert len(isolated.jobs(batch["id"])) == 2
+    assert isolated.jobs(batch["id"])[1]["status"] == "pending"
+
+
+@pytest.mark.asyncio
+async def test_surviving_worker_that_dies_without_result_is_interrupted(isolated, monkeypatch):
+    from artemis.runtime import DeviceExecutionLock
+
+    job = isolated.jobs(service.create_batch(payload())["id"])[0]
+    isolated.save_job({**job, "status": "running", "recovery_waiting": True})
+    monkeypatch.setattr(DeviceExecutionLock, "get_active_owners", lambda: {})
+    await service.reconcile_recovered_jobs()
+    saved = isolated.job(job["session_id"])
+    assert saved["status"] == "interrupted"
+    assert "服务中断" in saved["error"]
+    assert not saved.get("recovery_waiting")
+
+
 @pytest.mark.asyncio
 async def test_batch_admission_idempotent_and_bound_device(isolated, monkeypatch):
     from apps.admin_console import server

tests/unit/admin_console/test_server_lifecycle.py

@@ -15,6 +15,7 @@
 import asyncio
 from contextlib import suppress
 import signal
+import socket
 from unittest.mock import AsyncMock, MagicMock, patch
 
 import pytest
@@ -23,10 +24,55 @@ import uvicorn
 from apps.admin_console.core.state import state
 from apps.admin_console.routers.tasks import stream_events
 from apps.admin_console.server import ArtemisUvicornServer, app, on_shutdown
+from apps.admin_console import server as server_module
 
 WINDOWS_FORCE_SIGNAL = getattr(signal, "SIGBREAK", signal.SIGTERM)
 
 
+@pytest.mark.parametrize("reload", [False, True])
+def test_occupied_port_never_initializes_or_overwrites_server_metadata(
+    tmp_path, monkeypatch, reload
+):
+    from artemis.runtime import server_lifecycle
+
+    metadata = tmp_path / "server.json"
+    metadata.write_text('{"pid":123,"lifecycle_token":"original"}', encoding="utf-8")
+    monkeypatch.setattr(server_lifecycle, "get_server_info_file", lambda: metadata)
+    before = metadata.read_bytes()
+    with (
+        socket.create_server(("127.0.0.1", 0)) as original,
+        patch.object(server_module.ArtemisUvicornServer, "run") as run,
+        patch.object(server_module, "write_server_info") as publish,
+    ):
+        with pytest.raises(SystemExit, match="1"):
+            server_module.run_ui_server("127.0.0.1", original.getsockname()[1], reload=reload)
+        run.assert_not_called()
+        publish.assert_not_called()
+        assert metadata.read_bytes() == before
+
+
+def test_server_holds_port_before_running_lifespan(monkeypatch):
+    monkeypatch.setattr(state, "port", 0, raising=False)
+    monkeypatch.setattr(state, "host", "127.0.0.1", raising=False)
+    observed = []
+
+    def run(_server, sockets):
+        listener = sockets[0]
+        port = listener.getsockname()[1]
+        observed.append(listener)
+        assert state.port == port
+        with pytest.raises(OSError), socket.create_server(("127.0.0.1", port)):
+            pass
+
+    monkeypatch.setattr(server_module.ArtemisUvicornServer, "run", run)
+    monkeypatch.setattr(server_module, "clear_server_info", lambda **kwargs: None)
+    with patch.object(server_module, "write_server_info") as publish:
+        server_module.run_ui_server("127.0.0.1", 0)
+        publish.assert_not_called()  # Publication belongs to the later lifespan.
+    assert len(observed) == 1
+    assert observed[0].fileno() == -1
+
+
 def _pending_queue_get_tasks():
     return {
         task

tests/unit/runtime/test_daemon_client.py

@@ -93,6 +93,7 @@ def test_ensure_daemon_running_standalone_override(monkeypatch):
 def test_ensure_daemon_running_auto_spawns():
     with (
         patch("artemis.runtime.daemon_client.is_daemon_running", side_effect=[False, True]),
+        patch("artemis.runtime.daemon_client.is_port_in_use", return_value=False),
         patch("artemis.runtime.daemon_client.spawn_daemon") as mock_spawn,
     ):
         ok, base_url = ensure_daemon_running(timeout=1.0, wait_ready=True)
@@ -101,6 +102,42 @@ def test_ensure_daemon_running_auto_spawns():
         mock_spawn.assert_called_once()
 
 
+def test_busy_live_daemon_is_reused_without_spawning():
+    with (
+        patch("artemis.runtime.daemon_client.is_daemon_running", return_value=False),
+        patch("artemis.runtime.daemon_client.is_port_in_use", return_value=True),
+        patch(
+            "artemis.runtime.daemon_client.read_server_info",
+            return_value={"host": "127.0.0.1", "port": 8000, "pid": 12345},
+        ),
+        patch("artemis.runtime.daemon_client.pid_is_alive", return_value=True),
+        patch("artemis.runtime.daemon_client.spawn_daemon") as spawn,
+    ):
+        assert ensure_daemon_running() == (True, "http://127.0.0.1:8000")
+        spawn.assert_not_called()
+
+
+def test_unknown_port_owner_is_not_overwritten():
+    with (
+        patch("artemis.runtime.daemon_client.is_daemon_running", return_value=False),
+        patch("artemis.runtime.daemon_client.is_port_in_use", return_value=True),
+        patch("artemis.runtime.daemon_client.read_server_info", return_value=None),
+        patch("artemis.runtime.daemon_client.spawn_daemon") as spawn,
+    ):
+        assert ensure_daemon_running() == (False, None)
+        spawn.assert_not_called()
+
+
+def test_liveness_probe_uses_lightweight_endpoint_and_configured_timeout(monkeypatch):
+    from artemis.config.settings import settings
+
+    monkeypatch.setattr(settings, "ARTEMIS_DAEMON_PROBE_TIMEOUT_SECONDS", 7.0)
+    with patch("urllib.request.urlopen", return_value=_daemon_ok_response(b"{}")) as probe:
+        assert is_daemon_running()
+        assert probe.call_args.args[0].full_url.endswith("/api/system/server-status")
+        assert probe.call_args.kwargs["timeout"] == 7.0
+
+
 def test_submit_task_to_daemon():
     mock_resp = MagicMock()
     mock_resp.status = 200