def test_transient_heartbeat_failure(mocker):
with instance_for_test() as instance:
mocker.patch(
"dagster.daemon.controller.get_daemon_status",
side_effect=Exception("Transient heartbeat failure"),
)
heartbeat_interval_seconds = 1
heartbeat_tolerance_seconds = 5
with daemon_controller_from_instance(
instance,
workspace_load_target=EmptyWorkspaceTarget(),
heartbeat_interval_seconds=heartbeat_interval_seconds,
heartbeat_tolerance_seconds=heartbeat_tolerance_seconds,
) as controller:
controller.check_daemon_heartbeats() # doesn't immediately fail despite transient error
time.sleep(2 * heartbeat_tolerance_seconds)
with pytest.raises(
Exception,
match="Stopping dagster-daemon process since the following threads are no longer sending heartbeats",
):
controller.check_daemon_heartbeats()
def test_multiple_error_daemon(monkeypatch):
with instance_for_test() as instance:
from dagster.daemon.daemon import SensorDaemon
def run_loop_error(_, _instance, _workspace):
# ?message stack cls_name cause"
yield SerializableErrorInfo("foobar", None, None, None)
yield SerializableErrorInfo("bizbuz", None, None, None)
while True:
yield
time.sleep(0.5)
monkeypatch.setattr(SensorDaemon, "core_loop", run_loop_error)
init_time = pendulum.now("UTC")
heartbeat_interval_seconds = 1
with daemon_controller_from_instance(
instance,
workspace_load_target=EmptyWorkspaceTarget(),
heartbeat_interval_seconds=heartbeat_interval_seconds,
) as controller:
while True:
now = pendulum.now("UTC")
if all_daemons_live(
instance,
heartbeat_interval_seconds=heartbeat_interval_seconds):
# Despite error, daemon should still be running
controller.check_daemon_threads()
controller.check_daemon_heartbeats()
status = get_daemon_statuses(
instance, [SensorDaemon.daemon_type()],
now.float_timestamp)[SensorDaemon.daemon_type()]
if status.healthy == False and len(
status.last_heartbeat.errors) == 2:
assert status.last_heartbeat.errors[0].message.strip(
) == "bizbuz"
assert status.last_heartbeat.errors[1].message.strip(
) == "foobar"
break
if (now - init_time).total_seconds() > 10:
raise Exception("timed out waiting for heartbeat error")
time.sleep(0.5)
def test_thread_die_daemon(monkeypatch):
with instance_for_test(overrides={}) as instance:
from dagster.daemon.daemon import SchedulerDaemon, SensorDaemon
iteration_ran = {"ran": False}
def run_loop_error(_, _instance, _workspace):
iteration_ran["ran"] = True
raise KeyboardInterrupt
yield # pylint: disable=unreachable
monkeypatch.setattr(SensorDaemon, "core_loop", run_loop_error)
heartbeat_interval_seconds = 1
init_time = pendulum.now("UTC")
with daemon_controller_from_instance(
instance,
workspace_load_target=EmptyWorkspaceTarget(),
heartbeat_interval_seconds=heartbeat_interval_seconds,
) as controller:
while True:
now = pendulum.now("UTC")
status = get_daemon_statuses(
instance,
[SchedulerDaemon.daemon_type()],
now.float_timestamp,
heartbeat_interval_seconds=heartbeat_interval_seconds,
)[SchedulerDaemon.daemon_type()]
if iteration_ran["ran"] and status.healthy:
try:
controller.check_daemon_threads(
) # Should eventually throw since the sensor thread is interrupted
except Exception as e:
assert (
"Stopping dagster-daemon process since the following threads are no longer running: ['SENSOR']"
in str(e))
break
if (now - init_time).total_seconds() > 20:
raise Exception(
"timed out waiting for check_daemons to fail")
time.sleep(0.5)
def test_scheduler_instance():
with instance_for_test(
overrides={
"scheduler": {
"module": "dagster.core.scheduler",
"class": "DagsterDaemonScheduler",
},
}) as instance:
with daemon_controller_from_instance(
instance,
workspace_load_target=EmptyWorkspaceTarget(),
) as controller:
daemons = controller.daemons
assert len(daemons) == 3
assert any(
isinstance(daemon, SchedulerDaemon) for daemon in daemons)
def test_run_coordinator_instance():
with instance_for_test(
overrides={
"run_coordinator": {
"module":
"dagster.core.run_coordinator.queued_run_coordinator",
"class": "QueuedRunCoordinator",
},
}) as instance:
with daemon_controller_from_instance(
instance,
workspace_load_target=EmptyWorkspaceTarget(),
) as controller:
daemons = controller.daemons
assert len(daemons) == 4
assert any(
isinstance(daemon, QueuedRunCoordinatorDaemon)
for daemon in daemons)
def test_healthy_with_different_daemons():
with instance_for_test() as instance:
with daemon_controller_from_instance(
instance,
workspace_load_target=EmptyWorkspaceTarget(),
):
with instance_for_test(
overrides={
"run_coordinator": {
"module": "dagster.core.run_coordinator.queued_run_coordinator",
"class": "QueuedRunCoordinator",
},
}
) as other_instance:
now = pendulum.now("UTC")
assert not all_daemons_healthy(
other_instance, curr_time_seconds=now.float_timestamp
)
assert not all_daemons_live(other_instance, curr_time_seconds=now.float_timestamp)
def workspace_fixture():
with create_test_daemon_workspace(workspace_load_target=EmptyWorkspaceTarget()) as workspace:
yield workspace
def test_warn_multiple_daemons(capsys):
from dagster.daemon.daemon import SensorDaemon
with instance_for_test() as instance:
init_time = pendulum.now("UTC")
heartbeat_interval_seconds = 1
with daemon_controller_from_instance(
instance,
workspace_load_target=EmptyWorkspaceTarget(),
heartbeat_interval_seconds=heartbeat_interval_seconds,
):
while True:
now = pendulum.now("UTC")
if all_daemons_live(
instance, heartbeat_interval_seconds=heartbeat_interval_seconds
):
captured = capsys.readouterr()
assert "Another SENSOR daemon is still sending heartbeats" not in captured.out
break
if (now - init_time).total_seconds() > 10:
raise Exception("timed out waiting for daemon status")
time.sleep(0.5)
capsys.readouterr()
init_time = pendulum.now("UTC")
status = get_daemon_status(
instance,
SensorDaemon.daemon_type(),
now.float_timestamp,
heartbeat_interval_seconds=heartbeat_interval_seconds,
)
last_heartbeat_time = status.last_heartbeat.timestamp
# No warning when a second controller starts up again
with daemon_controller_from_instance(
instance,
workspace_load_target=EmptyWorkspaceTarget(),
heartbeat_interval_seconds=heartbeat_interval_seconds,
):
while True:
now = pendulum.now("UTC")
status = get_daemon_status(
instance,
SensorDaemon.daemon_type(),
now.float_timestamp,
heartbeat_interval_seconds=heartbeat_interval_seconds,
)
if status.last_heartbeat and status.last_heartbeat.timestamp != last_heartbeat_time:
captured = capsys.readouterr()
assert "Another SENSOR daemon is still sending heartbeats" not in captured.out
break
if (now - init_time).total_seconds() > 10:
raise Exception("timed out waiting for new daemon status")
time.sleep(0.5)
status = get_daemon_status(
instance,
SensorDaemon.daemon_type(),
now.float_timestamp,
heartbeat_interval_seconds=heartbeat_interval_seconds,
)
last_heartbeat_time = status.last_heartbeat.timestamp
# Starting up a controller while one is running produces the warning though
with daemon_controller_from_instance(
instance,
workspace_load_target=EmptyWorkspaceTarget(),
heartbeat_interval_seconds=heartbeat_interval_seconds,
):
# Wait for heartbeats while two controllers are running at once and there will
# be a warning
init_time = pendulum.now("UTC")
while True:
now = pendulum.now("UTC")
captured = capsys.readouterr()
if "Another SENSOR daemon is still sending heartbeats" in captured.out:
break
if (now - init_time).total_seconds() > 60:
raise Exception("timed out waiting for heartbeats")
time.sleep(5)
def test_error_daemon(monkeypatch):
with instance_for_test() as instance:
from dagster.daemon.daemon import SensorDaemon
should_raise_errors = True
error_count = {"count": 0}
def run_loop_error(_, _instance, _workspace):
if should_raise_errors:
time.sleep(0.5)
error_count["count"] = error_count["count"] + 1
raise DagsterInvariantViolationError("foobar:" + str(error_count["count"]))
while True:
yield
time.sleep(0.5)
def _get_error_number(error):
error_message = error.message.strip()
return int(error_message.split("foobar:")[1])
monkeypatch.setattr(SensorDaemon, "core_loop", run_loop_error)
heartbeat_interval_seconds = 1
gen_daemons = lambda instance: [SensorDaemon()]
init_time = pendulum.now("UTC")
with daemon_controller_from_instance(
instance,
workspace_load_target=EmptyWorkspaceTarget(),
heartbeat_interval_seconds=heartbeat_interval_seconds,
gen_daemons=gen_daemons,
error_interval_seconds=10,
) as controller:
while True:
now = pendulum.now("UTC")
if get_daemon_status(
instance,
SensorDaemon.daemon_type(),
heartbeat_interval_seconds=heartbeat_interval_seconds,
ignore_errors=True,
).healthy:
# Despite error, daemon should still be running
controller.check_daemon_threads()
controller.check_daemon_heartbeats()
status = get_daemon_status(
instance,
SensorDaemon.daemon_type(),
now.float_timestamp,
heartbeat_interval_seconds=heartbeat_interval_seconds,
)
assert status.healthy == False
# Errors build up until there are > 5, then pull off the last
if len(status.last_heartbeat.errors) >= 5:
first_error_number = _get_error_number(status.last_heartbeat.errors[0])
if first_error_number > 5:
# Verify error numbers decrease consecutively
assert [
_get_error_number(error) for error in status.last_heartbeat.errors
] == list(range(first_error_number, first_error_number - 5, -1))
assert not get_daemon_status(
instance,
SensorDaemon.daemon_type(),
curr_time_seconds=now.float_timestamp,
heartbeat_interval_seconds=heartbeat_interval_seconds,
).healthy
assert get_daemon_status(
instance,
SensorDaemon.daemon_type(),
curr_time_seconds=now.float_timestamp,
heartbeat_interval_seconds=heartbeat_interval_seconds,
ignore_errors=True,
).healthy
time.sleep(3)
status = get_daemon_status(
instance,
SensorDaemon.daemon_type(),
now.float_timestamp,
heartbeat_interval_seconds=heartbeat_interval_seconds,
)
# Error count does not rise above 5, continues to increase
assert len(status.last_heartbeat.errors) == 5
new_first_error_number = _get_error_number(
status.last_heartbeat.errors[0]
)
assert new_first_error_number > first_error_number
break
if (now - init_time).total_seconds() > 15:
raise Exception("timed out waiting for heartbeat error")
time.sleep(0.5)
# Once the sensor no longer raises errors, they should return to 0 once
# enough time passes
should_raise_errors = False
init_time = pendulum.now("UTC")
while True:
now = pendulum.now("UTC")
status = get_daemon_status(
instance,
SensorDaemon.daemon_type(),
now.float_timestamp,
heartbeat_interval_seconds=heartbeat_interval_seconds,
)
# Error count does not rise above 5
if len(status.last_heartbeat.errors) == 0:
break
if (now - init_time).total_seconds() > 15:
raise Exception("timed out waiting for hearrteat errors to return to 0")
time.sleep(0.5)
def test_healthy():
with instance_for_test(
overrides={
"run_coordinator": {
"module": "dagster.core.run_coordinator.queued_run_coordinator",
"class": "QueuedRunCoordinator",
},
}
) as instance:
init_time = pendulum.now("UTC")
heartbeat_interval_seconds = 1
assert not all_daemons_healthy(
instance,
curr_time_seconds=init_time.float_timestamp,
heartbeat_interval_seconds=heartbeat_interval_seconds,
)
assert not all_daemons_live(
instance,
curr_time_seconds=init_time.float_timestamp,
heartbeat_interval_seconds=heartbeat_interval_seconds,
)
with daemon_controller_from_instance(
instance,
workspace_load_target=EmptyWorkspaceTarget(),
heartbeat_interval_seconds=heartbeat_interval_seconds,
) as controller:
while True:
now = pendulum.now("UTC")
if all_daemons_healthy(
instance,
curr_time_seconds=now.float_timestamp,
heartbeat_interval_seconds=heartbeat_interval_seconds,
) and all_daemons_live(
instance,
curr_time_seconds=now.float_timestamp,
heartbeat_interval_seconds=heartbeat_interval_seconds,
):
controller.check_daemon_threads()
controller.check_daemon_heartbeats()
beyond_tolerated_time = (
now.float_timestamp + DEFAULT_DAEMON_HEARTBEAT_TOLERANCE_SECONDS + 1
)
assert not all_daemons_healthy(
instance,
curr_time_seconds=beyond_tolerated_time,
heartbeat_interval_seconds=heartbeat_interval_seconds,
)
assert not all_daemons_live(
instance,
curr_time_seconds=beyond_tolerated_time,
heartbeat_interval_seconds=heartbeat_interval_seconds,
)
break
if (now - init_time).total_seconds() > 10:
raise Exception("timed out waiting for instance to become healthy")
time.sleep(0.5)
def workspace_fixture(instance):
with create_test_daemon_workspace(
workspace_load_target=EmptyWorkspaceTarget(), instance=instance
) as workspace:
yield workspace
def created_workspace_load_target(kwargs):
check.dict_param(kwargs, "kwargs")
if are_all_keys_empty(kwargs, WORKSPACE_CLI_ARGS):
if kwargs.get("empty_workspace"):
return EmptyWorkspaceTarget()
if os.path.exists("workspace.yaml"):
return WorkspaceFileTarget(paths=["workspace.yaml"])
raise click.UsageError("No arguments given and workspace.yaml not found.")
if kwargs.get("workspace"):
_check_cli_arguments_none(
kwargs,
"python_file",
"working_directory",
"empty_working_directory",
"module_name",
"package_name",
"attribute",
"grpc_host",
"grpc_port",
"grpc_socket",
)
return WorkspaceFileTarget(paths=list(kwargs["workspace"]))
if kwargs.get("python_file"):
_check_cli_arguments_none(
kwargs,
"module_name",
"package_name",
"grpc_host",
"grpc_port",
"grpc_socket",
)
working_directory = get_working_directory_from_kwargs(kwargs)
return PythonFileTarget(
python_file=kwargs.get("python_file"),
attribute=kwargs.get("attribute"),
working_directory=working_directory,
location_name=None,
)
if kwargs.get("module_name"):
_check_cli_arguments_none(
kwargs,
"package_name",
"working_directory",
"empty_working_directory",
"grpc_host",
"grpc_port",
"grpc_socket",
)
return ModuleTarget(
module_name=kwargs.get("module_name"),
attribute=kwargs.get("attribute"),
location_name=None,
)
if kwargs.get("package_name"):
_check_cli_arguments_none(
kwargs,
"working_directory",
"empty_working_directory",
"grpc_host",
"grpc_port",
"grpc_socket",
)
return PackageTarget(
package_name=kwargs.get("package_name"),
attribute=kwargs.get("attribute"),
location_name=None,
)
if kwargs.get("grpc_port"):
_check_cli_arguments_none(
kwargs,
"attribute",
"working_directory",
"empty_working_directory",
"grpc_socket",
)
return GrpcServerTarget(
port=kwargs.get("grpc_port"),
socket=None,
host=(kwargs.get("grpc_host") if kwargs.get("grpc_host") else "localhost"),
location_name=None,
)
elif kwargs.get("grpc_socket"):
_check_cli_arguments_none(
kwargs,
"attribute",
"working_directory",
"empty_working_directory",
)
return GrpcServerTarget(
port=None,
socket=kwargs.get("grpc_socket"),
host=(kwargs.get("grpc_host") if kwargs.get("grpc_host") else "localhost"),
location_name=None,
)
else:
_cli_load_invariant(False)
def get_workspace_load_target(kwargs: Dict[str, str]):
check.dict_param(kwargs, "kwargs")
if are_all_keys_empty(kwargs, WORKSPACE_CLI_ARGS):
if kwargs.get("empty_workspace"):
return EmptyWorkspaceTarget()
if os.path.exists("workspace.yaml"):
return WorkspaceFileTarget(paths=["workspace.yaml"])
raise click.UsageError(
"No arguments given and workspace.yaml not found.")
if kwargs.get("workspace"):
_check_cli_arguments_none(
kwargs,
"python_file",
"working_directory",
"module_name",
"package_name",
"attribute",
"grpc_host",
"grpc_port",
"grpc_socket",
)
return WorkspaceFileTarget(
paths=list(cast(Union[List, Tuple], kwargs.get("workspace"))))
if kwargs.get("python_file"):
_check_cli_arguments_none(
kwargs,
"module_name",
"package_name",
"grpc_host",
"grpc_port",
"grpc_socket",
)
working_directory = get_working_directory_from_kwargs(kwargs)
return PythonFileTarget(
python_file=check.str_elem(kwargs, "python_file"),
attribute=check.opt_str_elem(kwargs, "attribute"),
working_directory=working_directory,
location_name=None,
)
if kwargs.get("module_name"):
_check_cli_arguments_none(
kwargs,
"package_name",
"grpc_host",
"grpc_port",
"grpc_socket",
)
working_directory = get_working_directory_from_kwargs(kwargs)
return ModuleTarget(
module_name=check.str_elem(kwargs, "module_name"),
attribute=check.opt_str_elem(kwargs, "attribute"),
working_directory=working_directory,
location_name=None,
)
if kwargs.get("package_name"):
_check_cli_arguments_none(
kwargs,
"grpc_host",
"grpc_port",
"grpc_socket",
)
working_directory = get_working_directory_from_kwargs(kwargs)
return PackageTarget(
package_name=check.str_elem(kwargs, "package_name"),
attribute=check.opt_str_elem(kwargs, "attribute"),
working_directory=working_directory,
location_name=None,
)
if kwargs.get("grpc_port"):
_check_cli_arguments_none(
kwargs,
"attribute",
"working_directory",
"grpc_socket",
)
return GrpcServerTarget(
port=check.int_elem(kwargs, "grpc_port"),
socket=None,
host=check.opt_str_elem(kwargs, "grpc_host") or "localhost",
location_name=None,
)
elif kwargs.get("grpc_socket"):
_check_cli_arguments_none(
kwargs,
"attribute",
"working_directory",
)
return GrpcServerTarget(
port=None,
socket=check.str_elem(kwargs, "grpc_socket"),
host=check.opt_str_elem(kwargs, "grpc_host") or "localhost",
location_name=None,
)
else:
_cli_load_invariant(False)
# necessary for pyright, does not understand _cli_load_invariant(False) never returns
assert False
