def test_scaling_action_allowed_without_min_max_specified(self):
self.service_dict[
self.service_name_1].service_config_details = MagicMock()
self.service_dict[
self.service_name_1].service_config_details.max_instance = 5
self.service_dict[
self.service_name_1].service_config_details.min_instance = None
self.service_dict[
self.service_name_2].service_config_details = MagicMock()
self.service_dict[
self.service_name_2].service_config_details.max_instance = None
self.service_dict[
self.service_name_2].service_config_details.min_instance = 1
registry = RunningServicesRegistry(self.service_dict)
# service2 has 3 instances initially, max is undefined
scaling_action_service2 = ScalingAction(self.service_name_2)
scaling_action_service2.command = DynamiteScalingCommand.SCALE_UP
result = registry.scaling_action_allowed(scaling_action_service2)
assert result is True
scaling_action_service2.command = DynamiteScalingCommand.SCALE_DOWN
result = registry.scaling_action_allowed(scaling_action_service2)
assert result is True
# service1 has 2 instances initially, min is undefined
scaling_action_service1 = ScalingAction(self.service_name_1)
scaling_action_service1.command = DynamiteScalingCommand.SCALE_DOWN
result = registry.scaling_action_allowed(scaling_action_service1)
assert result is True
scaling_action_service1.command = DynamiteScalingCommand.SCALE_UP
result = registry.scaling_action_allowed(scaling_action_service1)
assert result is True
def test_scaling_action_allowed(self):
self.service_dict[self.service_name_1].service_config_details = MagicMock()
self.service_dict[self.service_name_1].service_config_details.max_instance = 5
self.service_dict[self.service_name_1].service_config_details.min_instance = 2
self.service_dict[self.service_name_2].service_config_details = MagicMock()
self.service_dict[self.service_name_2].service_config_details.max_instance = 3
self.service_dict[self.service_name_2].service_config_details.min_instance = 1
registry = RunningServicesRegistry(self.service_dict)
# service2 has 3 instances initially, max is 3
scaling_action_service2 = ScalingAction(self.service_name_2)
scaling_action_service2.command = DynamiteScalingCommand.SCALE_UP
result = registry.scaling_action_allowed(scaling_action_service2)
assert result is False
scaling_action_service2.command = DynamiteScalingCommand.SCALE_DOWN
result = registry.scaling_action_allowed(scaling_action_service2)
assert result is True
# service1 has 2 instances initially, min is 2
scaling_action_service1 = ScalingAction(self.service_name_1)
scaling_action_service1.command = DynamiteScalingCommand.SCALE_DOWN
result = registry.scaling_action_allowed(scaling_action_service1)
assert result is False
scaling_action_service1.command = DynamiteScalingCommand.SCALE_UP
result = registry.scaling_action_allowed(scaling_action_service1)
assert result is True
def test_scaling_action_allowed_with_unknown_service(self):
with pytest.raises(ValueError):
registry = RunningServicesRegistry(self.service_dict)
scaling_action_service1 = ScalingAction(
self.non_existing_service_name)
scaling_action_service1.command = DynamiteScalingCommand.SCALE_DOWN
registry.scaling_action_allowed(scaling_action_service1)
def test_scaling_action_allowed_scale_below_zero(self):
self.service_dict[self.service_name_1].service_config_details = MagicMock()
self.service_dict[self.service_name_1].service_config_details.max_instance = None
self.service_dict[self.service_name_1].service_config_details.min_instance = 2
self.service_dict[self.service_name_1].fleet_service_instances = {}
registry = RunningServicesRegistry(self.service_dict)
scaling_action_service1 = ScalingAction(self.service_name_1)
scaling_action_service1.command = DynamiteScalingCommand.SCALE_DOWN
result = registry.scaling_action_allowed(scaling_action_service1)
assert result is False
def test_scaling_action_allowed_scale_below_zero(self):
self.service_dict[
self.service_name_1].service_config_details = MagicMock()
self.service_dict[
self.service_name_1].service_config_details.max_instance = None
self.service_dict[
self.service_name_1].service_config_details.min_instance = 2
self.service_dict[self.service_name_1].fleet_service_instances = {}
registry = RunningServicesRegistry(self.service_dict)
scaling_action_service1 = ScalingAction(self.service_name_1)
scaling_action_service1.command = DynamiteScalingCommand.SCALE_DOWN
result = registry.scaling_action_allowed(scaling_action_service1)
assert result is False
class ScalingEngine(object):
AGGREGATE_METRICS = False
RETRANSMIT_COUNT_OF_FAILED_MESSAGES = 5
_exit_flag = None
_running = False
def __init__(self, configuration, exit_flag):
self._exit_flag = exit_flag
self._logger = logging.getLogger(__name__)
self._logger.info("Initialized scaling engine with configuration %s",
repr(configuration))
self._services_dictionary = configuration.services_dictionary
self._metrics_receiver = configuration.metrics_receiver
self._metrics = ScalingMetrics()
self._rule_checker = RuleChecker(configuration.scaling_policies,
configuration.services_dictionary)
self._executed_tasks_receiver = configuration.executed_task_receiver
self._scaling_action_sender = configuration.scaling_action_sender
atexit.register(self._on_engine_shutdown)
self._running_services_registry = RunningServicesRegistry(
configuration.services_dictionary)
instance_name_resolver = ServiceInstanceNameResolver(
configuration.etcd_connection)
self._service_instance_name_resolver = CachingServiceInstanceNameResolver(
instance_name_resolver)
def start(self):
self._running = True
self._logger.info("Starting scaling engine")
self._start_minimal_required_services()
while self._running:
try:
metrics_message = self._retreive_metrics()
scaling_actions = self._rule_checker.check_and_return_needed_scaling_actions(
metrics_message)
self._send_filtered_scaling_actions(scaling_actions)
executed_tasks = self._executed_tasks_receiver.receive()
self._update_running_tasks_registry(executed_tasks)
self._resend_failed_messages(executed_tasks)
except Exception:
self._logger.exception(
"Unexpected error in scaling engine occurred!")
finally:
if self._exit_flag == 1:
self._running = False
def _retreive_metrics(self):
# get metrics from etcd queue
metrics_message = self._metrics_receiver.receive_metrics()
if self.AGGREGATE_METRICS:
# aggregate metrics per type
self._metrics.add_metrics(metrics_message)
return metrics_message
def _send_filtered_scaling_actions(self, scaling_actions):
for scaling_action in scaling_actions:
if self._running_services_registry.scaling_action_allowed(
scaling_action):
if scaling_action.command == DynamiteScalingCommand.SCALE_DOWN:
scaling_action.service_instance_name = self._service_instance_name_resolver.resolve(
scaling_action.uuid)
self._scaling_action_sender.send_action(
DynamiteScalingRequest.from_scaling_action(scaling_action))
else:
continue
def _update_running_tasks_registry(self, executed_tasks):
for executed_task in executed_tasks:
if executed_task.success:
self._add_remove_running_service_based_on_command(
executed_task)
executed_task.message_processed()
def _add_remove_running_service_based_on_command(self, executor_response):
if executor_response.command == DynamiteScalingCommand.SCALE_UP:
self._logger.info(
"add running service because got positive response from executor %s",
repr(executor_response))
self._running_services_registry.add_running_service(
executor_response.service_name)
elif executor_response.command == DynamiteScalingCommand.SCALE_DOWN:
self._logger.info(
"remove running service because got positive response from executor %s",
repr(executor_response))
self._running_services_registry.remove_running_service(
executor_response.service_name)
else:
raise ValueError(
"Unknown command {}, cannot parse executor response: {}".
format(executor_response.command, executor_response))
def _resend_failed_messages(self, executor_response_messages):
for executor_response in executor_response_messages:
if executor_response.success:
continue
if executor_response.failure_counter <= self.RETRANSMIT_COUNT_OF_FAILED_MESSAGES:
self._logger.warn(
"got failed scaling request notification from executor %s",
repr(executor_response))
executor_response.failure_counter += 1
self._scaling_action_sender.send_action(executor_response)
executor_response.message_processed()
else:
self._report_failed_scaling_message(executor_response)
executor_response.message_processed()
def _report_failed_scaling_message(self, executor_response):
self._logger.error(
"executor could not handle scaling request %s after retransmitting %d times",
repr(executor_response), self.RETRANSMIT_COUNT_OF_FAILED_MESSAGES)
def _on_engine_shutdown(self):
self._logger.info("Shutting down scaling engine")
self._scaling_action_sender.close()
self._executed_tasks_receiver.close()
def _start_minimal_required_services(self):
self._logger.debug("Starting minimal required services")
for service_name, service in self._services_dictionary.items():
running_services = self._running_services_registry.number_of_running_instances_of_service(
service_name)
self._logger.debug("%d services are running of the service %s",
running_services, service_name)
if service.service_config_details.min_instance is None:
service.service_config_details.min_instance = 1
self._logger.debug("%d services have to be running at least of %s",
service.service_config_details.min_instance,
service_name)
services_to_start = service.service_config_details.min_instance - running_services
services_to_start = 0 if services_to_start < 0 else services_to_start
self._logger.debug("need to start %d services of %s",
services_to_start, service_name)
for x in range(0, services_to_start):
request = DynamiteScalingRequest.from_service(
service, DynamiteScalingCommand.SCALE_UP)
self._scaling_action_sender.send_action(request)
def test_scaling_action_allowed_with_invalid_command(self):
with pytest.raises(ValueError):
registry = RunningServicesRegistry(self.service_dict)
scaling_action_service1 = ScalingAction(self.service_name_1)
scaling_action_service1.command = "scale_UP"
registry.scaling_action_allowed(scaling_action_service1)
def test_scaling_action_allowed_with_unknown_service(self):
with pytest.raises(ValueError):
registry = RunningServicesRegistry(self.service_dict)
scaling_action_service1 = ScalingAction(self.non_existing_service_name)
scaling_action_service1.command = DynamiteScalingCommand.SCALE_DOWN
registry.scaling_action_allowed(scaling_action_service1)
class ScalingEngine(object):
AGGREGATE_METRICS = False
RETRANSMIT_COUNT_OF_FAILED_MESSAGES = 5
_exit_flag = None
_running = False
def __init__(self, configuration, exit_flag):
self._exit_flag = exit_flag
self._logger = logging.getLogger(__name__)
self._logger.info("Initialized scaling engine with configuration %s", repr(configuration))
self._services_dictionary = configuration.services_dictionary
self._metrics_receiver = configuration.metrics_receiver
self._metrics = ScalingMetrics()
self._rule_checker = RuleChecker(configuration.scaling_policies, configuration.services_dictionary)
self._executed_tasks_receiver = configuration.executed_task_receiver
self._scaling_action_sender = configuration.scaling_action_sender
atexit.register(self._on_engine_shutdown)
self._running_services_registry = RunningServicesRegistry(configuration.services_dictionary)
instance_name_resolver = ServiceInstanceNameResolver(configuration.etcd_connection)
self._service_instance_name_resolver = CachingServiceInstanceNameResolver(instance_name_resolver)
def start(self):
self._running = True
self._logger.info("Starting scaling engine")
self._start_minimal_required_services()
while self._running:
try:
metrics_message = self._retreive_metrics()
scaling_actions = self._rule_checker.check_and_return_needed_scaling_actions(metrics_message)
self._send_filtered_scaling_actions(scaling_actions)
executed_tasks = self._executed_tasks_receiver.receive()
self._update_running_tasks_registry(executed_tasks)
self._resend_failed_messages(executed_tasks)
except Exception:
self._logger.exception("Unexpected error in scaling engine occurred!")
finally:
if self._exit_flag == 1:
self._running = False
def _retreive_metrics(self):
# get metrics from etcd queue
metrics_message = self._metrics_receiver.receive_metrics()
if self.AGGREGATE_METRICS:
# aggregate metrics per type
self._metrics.add_metrics(metrics_message)
return metrics_message
def _send_filtered_scaling_actions(self, scaling_actions):
for scaling_action in scaling_actions:
if self._running_services_registry.scaling_action_allowed(scaling_action):
if scaling_action.command == DynamiteScalingCommand.SCALE_DOWN:
scaling_action.service_instance_name = self._service_instance_name_resolver.resolve(
scaling_action.uuid
)
self._scaling_action_sender.send_action(DynamiteScalingRequest.from_scaling_action(scaling_action))
else:
continue
def _update_running_tasks_registry(self, executed_tasks):
for executed_task in executed_tasks:
if executed_task.success:
self._add_remove_running_service_based_on_command(executed_task)
executed_task.message_processed()
def _add_remove_running_service_based_on_command(self, executor_response):
if executor_response.command == DynamiteScalingCommand.SCALE_UP:
self._logger.info("add running service because got positive response from executor %s", repr(executor_response))
self._running_services_registry.add_running_service(executor_response.service_name)
elif executor_response.command == DynamiteScalingCommand.SCALE_DOWN:
self._logger.info("remove running service because got positive response from executor %s", repr(executor_response))
self._running_services_registry.remove_running_service(executor_response.service_name)
else:
raise ValueError("Unknown command {}, cannot parse executor response: {}".format(
executor_response.command,
executor_response)
)
def _resend_failed_messages(self, executor_response_messages):
for executor_response in executor_response_messages:
if executor_response.success:
continue
if executor_response.failure_counter <= self.RETRANSMIT_COUNT_OF_FAILED_MESSAGES:
self._logger.warn("got failed scaling request notification from executor %s", repr(executor_response))
executor_response.failure_counter += 1
self._scaling_action_sender.send_action(executor_response)
executor_response.message_processed()
else:
self._report_failed_scaling_message(executor_response)
executor_response.message_processed()
def _report_failed_scaling_message(self, executor_response):
self._logger.error("executor could not handle scaling request %s after retransmitting %d times",
repr(executor_response),
self.RETRANSMIT_COUNT_OF_FAILED_MESSAGES
)
def _on_engine_shutdown(self):
self._logger.info("Shutting down scaling engine")
self._scaling_action_sender.close()
self._executed_tasks_receiver.close()
def _start_minimal_required_services(self):
self._logger.debug("Starting minimal required services")
for service_name, service in self._services_dictionary.items():
running_services = self._running_services_registry.number_of_running_instances_of_service(service_name)
self._logger.debug("%d services are running of the service %s", running_services, service_name)
if service.service_config_details.min_instance is None:
service.service_config_details.min_instance = 1
self._logger.debug("%d services have to be running at least of %s", service.service_config_details.min_instance, service_name)
services_to_start = service.service_config_details.min_instance - running_services
services_to_start = 0 if services_to_start < 0 else services_to_start
self._logger.debug("need to start %d services of %s", services_to_start, service_name)
for x in range(0, services_to_start):
request = DynamiteScalingRequest.from_service(service, DynamiteScalingCommand.SCALE_UP)
self._scaling_action_sender.send_action(request)
def test_scaling_action_allowed_with_invalid_command(self):
with pytest.raises(ValueError):
registry = RunningServicesRegistry(self.service_dict)
scaling_action_service1 = ScalingAction(self.service_name_1)
scaling_action_service1.command = "scale_UP"
registry.scaling_action_allowed(scaling_action_service1)