def describe(self, app_id: str) -> Optional[DescribeAppResponse]:
if app_id not in self._apps:
return None
local_app = self._apps[app_id]
# check if the app has been known to have finished
if is_terminal(local_app.state):
state = local_app.state
else:
running = False
failed = False
for (_, procs) in local_app.role_procs.items():
for p in procs:
running |= self._is_alive(p)
failed |= self._failed(p)
if running:
state = AppState.RUNNING
elif failed:
state = AppState.FAILED
self._terminate(local_app) # terminate danglers
else:
state = AppState.SUCCEEDED
local_app.set_state(state)
resp = DescribeAppResponse()
resp.app_id = app_id
resp.state = state
resp.num_restarts = 0
return resp
def test_log_lines(self):
app_id = "mock_app"
scheduler_mock = MagicMock()
scheduler_mock.describe.return_value = DescribeAppResponse(
app_id, AppState.RUNNING)
scheduler_mock.log_iter.return_value = iter(["hello", "world"])
session = StandaloneSession(name=SESSION_NAME,
schedulers={"default": scheduler_mock},
wait_interval=1)
role_name = "trainer"
replica_id = 2
regex = "QPS.*"
since = datetime.datetime.now()
until = datetime.datetime.now()
lines = list(
session.log_lines(
f"default://test_session/{app_id}",
role_name,
replica_id,
regex,
since,
until,
))
self.assertEqual(["hello", "world"], lines)
scheduler_mock.log_iter.assert_called_once_with(
app_id, role_name, replica_id, regex, since, until)
def describe(self, app_id: str) -> Optional[DescribeAppResponse]:
if app_id not in self._apps:
return None
local_app = self._apps[app_id]
structured_error_msg = local_app.get_structured_error_msg()
# check if the app is known to have finished
if is_terminal(local_app.state):
state = local_app.state
else:
running = False
failed = False
for replicas in local_app.role_replicas.values():
for r in replicas:
running |= r.is_alive()
failed |= r.failed()
if running:
state = AppState.RUNNING
elif failed:
state = AppState.FAILED
else:
state = AppState.SUCCEEDED
local_app.set_state(state)
if is_terminal(local_app.state):
local_app.close()
resp = DescribeAppResponse()
resp.app_id = app_id
resp.structured_error_msg = structured_error_msg
resp.state = state
resp.num_restarts = 0
return resp
def test_status_ui_url(self):
app_id = "test_app"
mock_scheduler = MagicMock()
resp = DescribeAppResponse()
resp.ui_url = "https://foobar"
mock_scheduler.submit.return_value = app_id
mock_scheduler.describe.return_value = resp
session = StandaloneSession(name="test_ui_url_session",
scheduler=mock_scheduler)
role = Role("ignored").runs("/bin/echo").on(self.test_container)
session.run(Application(app_id).of(role))
status = session.status(app_id)
self.assertEquals(resp.ui_url, status.ui_url)
def test_status_structured_msg(self):
app_id = "test_app"
mock_scheduler = MagicMock()
resp = DescribeAppResponse()
resp.scheduler_error_msg = '{"message": "test error"}'
mock_scheduler.submit.return_value = app_id
mock_scheduler.describe.return_value = resp
session = StandaloneSession(name="test_structured_msg",
schedulers={"default": mock_scheduler})
role = Role("ignored").runs("/bin/echo").on(self.test_container)
app_handle = session.run(Application(app_id).of(role))
status = session.status(app_handle)
self.assertEquals(resp.scheduler_error_msg, status.scheduler_error_msg)