def handle_connection(self, cl_socket):
"""
Handle socket connection.
:param cl_socket:
:return:
"""
service = None
while True:
cmd, msg = retrieve_msg(cl_socket)
if cmd == b'I':
resp = service.predict(msg)
cl_socket.send(resp)
elif cmd == b'L':
service, result, code = self.load_model(msg)
resp = bytearray()
resp += create_load_model_response(code, result)
cl_socket.send(resp)
if code != 200:
raise RuntimeError("{} - {}".format(code, result))
else:
raise ValueError("Received unknown command: {}".format(cmd))
if service is not None and service.context is not None and service.context.metrics is not None:
emit_metrics(service.context.metrics.store)
def handle_connection(self, cl_socket):
"""
Handle socket connection.
:param cl_socket:
:return:
"""
logging.basicConfig(stream=sys.stdout,
format="%(message)s",
level=logging.INFO)
cl_socket.setblocking(True)
while True:
cmd, msg = retrieve_msg(cl_socket)
if cmd == b'I':
resp = self.service.predict(msg)
cl_socket.send(resp)
elif cmd == b'L':
result, code = self.load_model(msg)
resp = bytearray()
resp += create_load_model_response(code, result)
cl_socket.send(resp)
self._remap_io()
if code != 200:
raise RuntimeError("{} - {}".format(code, result))
else:
raise ValueError("Received unknown command: {}".format(cmd))
if self.service is not None and self.service.context is not None \
and self.service.context.metrics is not None:
emit_metrics(self.service.context.metrics.store)
def load(self, model_name, model_dir, handler, gpu_id, batch_size):
"""
Load MMS 1.0 model from file.
:param model_name:
:param model_dir:
:param handler:
:param gpu_id:
:param batch_size:
:return:
"""
logging.debug("Loading model - working dir: %s", os.getcwd())
# TODO: Request ID is not given. UUID is a temp UUID.
metrics = MetricsStore(uuid.uuid4(), model_name)
manifest_file = os.path.join(model_dir, "MAR-INF/MANIFEST.json")
manifest = None
if os.path.exists(manifest_file):
with open(manifest_file) as f:
manifest = json.load(f)
temp = handler.split(":", 1)
module_name = temp[0]
function_name = None if len(temp) == 1 else temp[1]
if module_name.endswith(".py"):
module_name = module_name[:-3]
module = importlib.import_module(module_name)
if module is None:
raise ValueError("Unable to load module {}, make sure it is added to python path".format(module_name))
if function_name is None:
function_name = "handle"
if hasattr(module, function_name):
entry_point = getattr(module, function_name)
service = Service(model_name, model_dir, manifest, entry_point, gpu_id, batch_size)
service.context.metrics = metrics
# initialize model at load time
entry_point(None, service.context)
else:
model_class_definitions = ModelLoader.list_model_services(module)
if len(model_class_definitions) != 1:
raise ValueError("Expected only one class in custom service code or a function entry point")
model_class = model_class_definitions[0]
model_service = model_class()
handle = getattr(model_service, "handle")
if handle is None:
raise ValueError("Expect handle method in class {}".format(str(model_class)),)
service = Service(model_name, model_dir, manifest, model_service.handle, gpu_id, batch_size)
initialize = getattr(model_service, "initialize")
if initialize is not None:
# noinspection PyBroadException
try:
model_service.initialize(service.context)
# pylint: disable=broad-except
except Exception:
sys.exc_clear()
emit_metrics(metrics.store)
return service
def test_emit_metrics(self, caplog):
caplog.set_level(logging.INFO)
metrics = {'test_emit_metrics': True}
emit_metrics(metrics)
assert "[METRICS]" in caplog.text
def test_metrics(caplog):
"""
Test if metric classes methods behave as expected
Also checks global metric service methods
"""
caplog.set_level(logging.INFO)
# Create a batch of request ids
request_ids = {0: 'abcd', 1: "xyz", 2: "qwerty", 3: "hjshfj"}
all_req_ids = ','.join(request_ids.values())
model_name = "dummy model"
# Create a metrics objects
metrics = MetricsStore(request_ids, model_name)
# Counter tests
metrics.add_counter('CorrectCounter', 1, 1)
test_metric = metrics.cache[get_model_key('CorrectCounter', 'count', 'xyz',
model_name)]
assert 'CorrectCounter' == test_metric.name
metrics.add_counter('CorrectCounter', 1, 1)
metrics.add_counter('CorrectCounter', 1, 3)
metrics.add_counter('CorrectCounter', 1)
test_metric = metrics.cache[get_model_key('CorrectCounter', 'count',
all_req_ids, model_name)]
assert 'CorrectCounter' == test_metric.name
metrics.add_counter('CorrectCounter', 3)
test_metric = metrics.cache[get_model_key('CorrectCounter', 'count', 'xyz',
model_name)]
assert test_metric.value == 2
test_metric = metrics.cache[get_model_key('CorrectCounter', 'count',
'hjshfj', model_name)]
assert test_metric.value == 1
test_metric = metrics.cache[get_model_key('CorrectCounter', 'count',
all_req_ids, model_name)]
assert test_metric.value == 4
# Check what is emitted is correct
emit_metrics(metrics.store)
assert "hjshfj" in caplog.text
assert "ModelName:dummy model" in caplog.text
# Adding other types of metrics
# Check for time metric
with pytest.raises(Exception) as e_info:
metrics.add_time('WrongTime', 20, 1, 'ns')
assert "the unit for a timed metric should be one of ['ms', 's']" == e_info.value.args[
0]
metrics.add_time('CorrectTime', 20, 2, 's')
metrics.add_time('CorrectTime', 20, 0)
test_metric = metrics.cache[get_model_key('CorrectTime', 'ms', 'abcd',
model_name)]
assert test_metric.value == 20
assert test_metric.unit == 'Milliseconds'
test_metric = metrics.cache[get_model_key('CorrectTime', 's', 'qwerty',
model_name)]
assert test_metric.value == 20
assert test_metric.unit == 'Seconds'
# Size based metrics
with pytest.raises(Exception) as e_info:
metrics.add_size('WrongSize', 20, 1, 'TB')
assert "The unit for size based metric is one of ['MB','kB', 'GB', 'B']" == e_info.value.args[
0]
metrics.add_size('CorrectSize', 200, 0, 'GB')
metrics.add_size('CorrectSize', 10, 2)
test_metric = metrics.cache[get_model_key('CorrectSize', 'GB', 'abcd',
model_name)]
assert test_metric.value == 200
assert test_metric.unit == 'Gigabytes'
test_metric = metrics.cache[get_model_key('CorrectSize', 'MB', 'qwerty',
model_name)]
assert test_metric.value == 10
assert test_metric.unit == 'Megabytes'
# Check a percentage metric
metrics.add_percent('CorrectPercent', 20.0, 3)
test_metric = metrics.cache[get_model_key('CorrectPercent', 'percent',
'hjshfj', model_name)]
assert test_metric.value == 20.0
assert test_metric.unit == 'Percent'
# Check a error metric
metrics.add_error('CorrectError', 'Wrong values')
test_metric = metrics.cache[get_error_key('CorrectError', '')]
assert test_metric.value == 'Wrong values'
def test_emit_metrics(self, caplog):
metrics = {'test_emit_metrics': True}
emit_metrics(metrics)
assert "[METRICS]" in caplog.text