xinyin025/LangBot
1
0
Fork 0
mirror of https://github.com/langbot-app/LangBot.git synced 2026-11-05 23:33:03 +08:00
Code Issues Packages Projects Releases Wiki Activity

test(regression): preserve cancellation evidence across Box retirement

Browse Source
This commit is contained in:
huanghuoguoguo
2026-10-28 20:01:13 +08:00
parent f684d7f3f1
commit 1f842381df
3 changed files with 41 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tools/test_regression_execution_cases.py
+34 -4
View File
@@ -47,7 +47,15 @@ class ExecutionCasesTests(unittest.TestCase):
)
def run_fixture(
self, scenario, *, tool_ok=True, fault=True, primary_status=None, recovery_status='completed', delete_ok=True
self,
scenario,
*,
tool_ok=True,
fault=True,
primary_status=None,
recovery_status='completed',
delete_ok=True,
recycled=False,
):
expected = {
'execution.cancel_cleanup': 'cancelled',
@@ -92,6 +100,8 @@ class ExecutionCasesTests(unittest.TestCase):
if method == 'GET' and path == '/api/v1/agents/agent':
return 404, {'code': -1}
if path == '/api/v1/box/sessions':
if recycled and len(runs) == 1:
return 200, {'code': 0, 'data': []}
return 200, {'code': 0, 'data': [{'session_id': box_session_id, 'status': 'idle'}]}
if path.endswith('/events'):
run_id = path.split('/')[-2]
@@ -195,10 +205,27 @@ class ExecutionCasesTests(unittest.TestCase):
self.assertEqual(self.row['recovery_exec_results'][0]['result']['exit_code'], 42)
self.assertTrue(self.deleted)
def test_recycled_box_requires_the_same_persistent_file_check(self):
for scenario in ('execution.cancel_cleanup', 'execution.timeout_cleanup'):
with self.subTest(scenario=scenario):
with sqlite3.connect(self.ledger) as db:
db.execute('DELETE FROM agent_run')
result = self.run_fixture(scenario, recycled=True)
self.assertEqual(result['result'], 'FAULT_CASE_OK')
self.assertIsNone(self.row['box_after_interrupt'])
self.assertEqual(self.row['assertions']['no_late_side_effect'], 'passed')
with sqlite3.connect(self.ledger) as db:
db.execute('DELETE FROM agent_run')
with self.assertRaisesRegex(ValueError, 'INTERRUPT_LATE_SIDE_EFFECT'):
self.run_fixture(scenario, recycled=True, tool_ok=False, recovery_status='failed')
self.assertEqual(self.row['assertions']['no_late_side_effect'], 'failed')
def test_recovery_command_distinguishes_missing_setup_and_late_side_effect(self):
scenarios = json.loads(Path(cases.__file__).with_name('mock_llm_scenarios.json').read_text())
command = scenarios['mock-local-interrupt-check']['steps'][0]['tool_calls'][0]['arguments']['command']
command = command.removeprefix('sleep 9; ').replace('/tmp/langbot-{mock_scope}', str(self.root / 'fixture'))
command = command.removeprefix('sleep 9; ').replace(
'/workspace/langbot-{mock_scope}', str(self.root / 'fixture')
)
ready, leaked = self.root / 'fixture-ready', self.root / 'fixture-leak'
for setup, leak, status, output in (
(False, False, 41, 'INTERRUPT_SETUP_MISSING'),
@@ -253,7 +280,10 @@ class ExecutionCasesTests(unittest.TestCase):
try:
self.assertEqual(cases._stream(self.suite, 'agent', 'scope', {})[-1]['kind'], 'completed')
row = {}
with patch.object(cases, '_running_box', return_value='box'):
with (
patch.object(cases, '_running_box', return_value='box'),
patch.object(cases, '_sessions', return_value={'box': {'status': 'running'}}),
):
cancelled = cases._stream(self.suite, 'agent', 'scope', row, interrupt='cancel')
self.assertEqual(len(cancelled), 1)
self.assertEqual(row['box_session_id'], 'box')
@@ -292,7 +322,7 @@ class ExecutionCasesTests(unittest.TestCase):
for scope in ('run-a', 'run-b'):
call = request(scope)['choices'][0]['message']['tool_calls'][0]
command = json.loads(call['function']['arguments'])['command']
self.assertIn(f'/tmp/langbot-{scope}-ready', command)
self.assertIn(f'/workspace/langbot-{scope}-ready', command)
self.assertNotIn('{mock_scope}', command)
commands.append(command)
self.assertNotEqual(commands[0], commands[1])