|
| 1 | +import json |
| 2 | +import os |
| 3 | +import socket |
| 4 | +import subprocess |
| 5 | +import sys |
| 6 | +import tempfile |
| 7 | +import time |
| 8 | +import unittest |
| 9 | +import urllib.error |
| 10 | +import urllib.request |
| 11 | +from pathlib import Path |
| 12 | + |
| 13 | +from event_store import EventStore |
| 14 | +from task_engine import TaskManager |
| 15 | + |
| 16 | + |
| 17 | +class DurableTaskGatewayTests(unittest.TestCase): |
| 18 | + @staticmethod |
| 19 | + def _free_port() -> int: |
| 20 | + with socket.socket() as sock: |
| 21 | + sock.bind(("127.0.0.1", 0)) |
| 22 | + return int(sock.getsockname()[1]) |
| 23 | + |
| 24 | + @staticmethod |
| 25 | + def _request(base_url: str, path: str, *, method: str = "GET", body: dict | None = None) -> dict: |
| 26 | + data = json.dumps(body).encode("utf-8") if body is not None else None |
| 27 | + request = urllib.request.Request( |
| 28 | + base_url + path, |
| 29 | + data=data, |
| 30 | + method=method, |
| 31 | + headers={"Content-Type": "application/json", "Authorization": "Bearer durable-test-token"}, |
| 32 | + ) |
| 33 | + with urllib.request.urlopen(request, timeout=5) as response: |
| 34 | + return json.loads(response.read().decode("utf-8")) |
| 35 | + |
| 36 | + def _start_server(self, workspace: Path, db_path: Path, skills_path: Path) -> tuple[subprocess.Popen, str]: |
| 37 | + repository = Path(__file__).parents[1] |
| 38 | + port = self._free_port() |
| 39 | + env = os.environ.copy() |
| 40 | + env.update({ |
| 41 | + "HOME": str(workspace / "home"), |
| 42 | + "KYROZEN_DB_PATH": str(db_path), |
| 43 | + "KYROZEN_DISABLE_VECTOR_INDEX": "1", |
| 44 | + "KYROZEN_PROVIDER": "ollama", |
| 45 | + "KYROZEN_BASE_URL": "http://127.0.0.1:11434/v1", |
| 46 | + "KYROZEN_SERVER_TOKEN": "durable-test-token", |
| 47 | + "KYROZEN_SKILLS_DIR": str(skills_path), |
| 48 | + }) |
| 49 | + for variable in ("KYROZEN_API_KEY", "DEEPSEEK_API_KEY", "OPENAI_API_KEY", "ANTHROPIC_API_KEY"): |
| 50 | + env.pop(variable, None) |
| 51 | + process = subprocess.Popen( |
| 52 | + [sys.executable, str(repository / "server.py"), "--host", "127.0.0.1", "--port", str(port)], |
| 53 | + cwd=workspace, |
| 54 | + env=env, |
| 55 | + stdout=subprocess.PIPE, |
| 56 | + stderr=subprocess.STDOUT, |
| 57 | + text=True, |
| 58 | + ) |
| 59 | + base_url = f"http://127.0.0.1:{port}" |
| 60 | + deadline = time.monotonic() + 20 |
| 61 | + while time.monotonic() < deadline: |
| 62 | + if process.poll() is not None: |
| 63 | + output = process.stdout.read() if process.stdout else "" |
| 64 | + self.fail(f"Gateway exited before becoming ready: {output}") |
| 65 | + try: |
| 66 | + health = self._request(base_url, "/api/health") |
| 67 | + if health.get("status") in {"ok", "degraded"}: |
| 68 | + return process, base_url |
| 69 | + except (OSError, urllib.error.URLError): |
| 70 | + time.sleep(0.1) |
| 71 | + process.terminate() |
| 72 | + process.wait(timeout=5) |
| 73 | + self.fail("Gateway did not become ready") |
| 74 | + |
| 75 | + @staticmethod |
| 76 | + def _stop_server(process: subprocess.Popen) -> None: |
| 77 | + if process.poll() is None: |
| 78 | + process.terminate() |
| 79 | + process.wait(timeout=5) |
| 80 | + if process.stdout: |
| 81 | + process.stdout.close() |
| 82 | + |
| 83 | + def test_gateway_recovers_scoped_running_task_and_executes_api_task(self): |
| 84 | + with tempfile.TemporaryDirectory(prefix="openkyrozen-task-gateway-") as directory: |
| 85 | + root = Path(directory) |
| 86 | + workspace = root / "workspace" |
| 87 | + workspace.mkdir() |
| 88 | + (workspace / "home").mkdir() |
| 89 | + db_path = root / "state.sqlite3" |
| 90 | + skills_path = root / "skills" |
| 91 | + store = EventStore(db_path) |
| 92 | + recovery_manager = TaskManager( |
| 93 | + store, user_id="local", workspace_id="default", session_id="restart-session" |
| 94 | + ) |
| 95 | + recovery_index = recovery_manager.add_task( |
| 96 | + "recover a running file task", |
| 97 | + checkpoint={"action": "write_file", "args": "recovered.txt|after restart"}, |
| 98 | + ) |
| 99 | + recovery_manager.set_status(recovery_index, "running") |
| 100 | + failed_manager = TaskManager( |
| 101 | + store, user_id="local", workspace_id="default", session_id="resume-session" |
| 102 | + ) |
| 103 | + failed_index = failed_manager.add_task("resume only when requested") |
| 104 | + failed_manager.set_status(failed_index, "failed") |
| 105 | + failed_id = failed_manager.tasks[failed_index]["id"] |
| 106 | + |
| 107 | + process, base_url = self._start_server(workspace, db_path, skills_path) |
| 108 | + try: |
| 109 | + deadline = time.monotonic() + 10 |
| 110 | + recovered = self._request(base_url, "/api/v2/tasks?session_id=restart-session") |
| 111 | + while time.monotonic() < deadline: |
| 112 | + recovered = self._request(base_url, "/api/v2/tasks?session_id=restart-session") |
| 113 | + if recovered["tasks"][0]["status"] == "succeeded": |
| 114 | + break |
| 115 | + time.sleep(0.1) |
| 116 | + self.assertEqual(recovered["tasks"][0]["status"], "succeeded") |
| 117 | + self.assertEqual((workspace / "recovered.txt").read_text(encoding="utf-8"), "after restart") |
| 118 | + |
| 119 | + created = self._request( |
| 120 | + base_url, |
| 121 | + "/api/v2/tasks", |
| 122 | + method="POST", |
| 123 | + body={ |
| 124 | + "description": "write a live task marker", |
| 125 | + "session_id": "task-live", |
| 126 | + "action": "write_file", |
| 127 | + "args": "task-runtime.txt|created by durable worker", |
| 128 | + "acceptance": ["marker exists"], |
| 129 | + }, |
| 130 | + ) |
| 131 | + live_id = created["task"]["id"] |
| 132 | + deadline = time.monotonic() + 10 |
| 133 | + live = created["task"] |
| 134 | + while time.monotonic() < deadline: |
| 135 | + live = self._request(base_url, "/api/v2/tasks?session_id=task-live")["tasks"][0] |
| 136 | + if live["status"] == "succeeded": |
| 137 | + break |
| 138 | + time.sleep(0.1) |
| 139 | + self.assertEqual(live["id"], live_id) |
| 140 | + self.assertEqual(live["status"], "succeeded") |
| 141 | + self.assertEqual( |
| 142 | + (workspace / "task-runtime.txt").read_text(encoding="utf-8"), |
| 143 | + "created by durable worker", |
| 144 | + ) |
| 145 | + |
| 146 | + before_resume = self._request(base_url, "/api/v2/tasks?session_id=resume-session") |
| 147 | + self.assertEqual(before_resume["tasks"][0]["status"], "failed") |
| 148 | + resumed = self._request( |
| 149 | + base_url, f"/api/v2/tasks/{failed_id}/resume", method="POST", body={} |
| 150 | + ) |
| 151 | + self.assertEqual(resumed["task"]["id"], failed_id) |
| 152 | + self.assertEqual(resumed["task"]["status"], "pending") |
| 153 | + |
| 154 | + events = self._request(base_url, "/api/v2/events?session_id=restart-session")["events"] |
| 155 | + event_types = {event["event_type"] for event in events} |
| 156 | + self.assertIn("task.recovered", event_types) |
| 157 | + self.assertIn("task.execution_completed", event_types) |
| 158 | + finally: |
| 159 | + self._stop_server(process) |
| 160 | + |
| 161 | + |
| 162 | +if __name__ == "__main__": |
| 163 | + unittest.main() |
0 commit comments