def _ray_serve_call(self, request):
work_item = request
if work_item.request_context == TaskContext.Web:
serve_context.web = True
asgi_scope, body_bytes = work_item.request_args
flask_request = build_flask_request(asgi_scope, body_bytes)
args = (flask_request, )
kwargs = {}
else:
serve_context.web = False
args = (FakeFlaskQuest(), )
kwargs = work_item.request_kwargs
result_object_id = work_item.result_object_id
start_timestamp = time.time()
try:
result = self.__call__(*args, **kwargs)
ray.worker.global_worker.put_object(result, result_object_id)
except Exception as e:
wrapped_exception = wrap_to_ray_error(e)
self._serve_metric_error_counter += 1
ray.worker.global_worker.put_object(wrapped_exception,
result_object_id)
self._serve_metric_latency_list.append(time.time() - start_timestamp)
serve_context.web = False
self._ray_serve_fetch()
def _ray_serve_main_loop(self, my_handle):
assert self._ray_serve_setup_completed
self._ray_serve_self_handle = my_handle
# Only retrieve the next task if we have completed previous task.
if self._ray_serve_cached_work_token is None:
work_token = ray.get(
self._ray_serve_router_handle.dequeue_request.remote(
self._ray_serve_dequeue_requestr_name))
else:
work_token = self._ray_serve_cached_work_token
work_token_id = ray.ObjectID(work_token)
ready, not_ready = ray.wait([work_token_id],
num_returns=1,
timeout=0.5)
if len(ready) == 1:
work_item = ray.get(work_token_id)
self._ray_serve_cached_work_token = None
else:
self._ray_serve_cached_work_token = work_token
self._ray_serve_self_handle._ray_serve_main_loop.remote(my_handle)
return
if work_item.request_context == TaskContext.Web:
serve_context.web = True
asgi_scope, body_bytes = work_item.request_args
flask_request = build_flask_request(asgi_scope, body_bytes)
args = (flask_request, )
kwargs = {}
else:
serve_context.web = False
args = (FakeFlaskQuest(), )
kwargs = work_item.request_kwargs
result_object_id = work_item.result_object_id
start_timestamp = time.time()
try:
result = self.__call__(*args, **kwargs)
ray.worker.global_worker.put_object(result_object_id, result)
except Exception as e:
wrapped_exception = wrap_to_ray_error(e)
self._serve_metric_error_counter += 1
ray.worker.global_worker.put_object(result_object_id,
wrapped_exception)
self._serve_metric_latency_list.append(time.time() - start_timestamp)
serve_context.web = False
# The worker finished one unit of work.
# It will now tail recursively schedule the main_loop again.
# TODO(simon): remove tail recursion, ask router to callback instead
self._ray_serve_self_handle._ray_serve_main_loop.remote(my_handle)
def parse_request_item(request_item):
if request_item.request_context == TaskContext.Web:
is_web_context = True
asgi_scope, body_bytes = request_item.request_args
flask_request = build_flask_request(asgi_scope, io.BytesIO(body_bytes))
args = (flask_request, )
kwargs = {}
else:
is_web_context = False
args = (FakeFlaskRequest(), )
kwargs = request_item.request_kwargs
return args, kwargs, is_web_context
def parse_request_item(request_item):
if request_item.request_context == TaskContext.Web:
is_web_context = True
asgi_scope, body_bytes = request_item.request_args
# http_body_bytes enclosed in list due to
# https://github.com/ray-project/ray/issues/6944
# TODO(alind): remove list enclosing after issue is fixed
flask_request = build_flask_request(asgi_scope,
io.BytesIO(body_bytes[0]))
args = (flask_request, )
kwargs = {}
else:
is_web_context = False
args = (FakeFlaskRequest(), )
kwargs = request_item.request_kwargs
return args, kwargs, is_web_context
def _ray_serve_main_loop(self, my_handle):
assert self._ray_serve_setup_completed
self._ray_serve_self_handle = my_handle
work_token = ray.get(
self._ray_serve_router_handle.dequeue_request.remote(
self._ray_serve_dequeue_requestr_name))
work_item = ray.get(ray.ObjectID(work_token))
if work_item.request_context == TaskContext.Web:
serve_context.web = True
asgi_scope, body_bytes = work_item.request_args
flask_request = build_flask_request(asgi_scope, body_bytes)
args = (flask_request, )
kwargs = {}
else:
serve_context.web = False
args = (FakeFlaskQuest(), )
kwargs = work_item.request_kwargs
result_object_id = work_item.result_object_id
try:
result = self.__call__(*args, **kwargs)
ray.worker.global_worker.put_object(result_object_id, result)
except Exception as e:
wrapped_exception = wrap_to_ray_error(e)
ray.worker.global_worker.put_object(result_object_id,
wrapped_exception)
serve_context.web = False
# The worker finished one unit of work.
# It will now tail recursively schedule the main_loop again.
# TODO(simon): remove tail recursion, ask router to callback instead
self._ray_serve_self_handle._ray_serve_main_loop.remote(my_handle)