Support up to 10MiB http request (microsoft#61)

* Changes minimum request size to 10MB to support all models in ONNX Model Zoo

onnxruntime/server/http/core/session.cc

@@ -15,11 +15,12 @@ HttpSession::HttpSession(const Routes& routes, tcp::socket socket)
 }
 
 void HttpSession::DoRead() {
-  // Make the request empty before reading,
-  // otherwise the operation behavior is undefined.
-  req_ = {};
+  req_.emplace();
 
-  http::async_read(socket_, buffer_, req_,
+  // TODO: make the max request size configable.
+  req_->body_limit(10 * 1024 * 1024);  // Max request size: 10 MiB
+
+  http::async_read(socket_, buffer_, *req_,
                    net::bind_executor(
                        strand_,
                        std::bind(
@@ -43,7 +44,7 @@ void HttpSession::OnRead(beast::error_code ec, std::size_t bytes_transferred) {
   }
 
   // Send the response
-  HandleRequest(std::move(req_));
+  HandleRequest(req_->release());
 }
 
 void HttpSession::OnWrite(beast::error_code ec, std::size_t bytes_transferred, bool close) {

onnxruntime/server/http/core/session.h

@@ -39,7 +39,7 @@ class HttpSession : public std::enable_shared_from_this<HttpSession> {
   tcp::socket socket_;
   net::strand<net::io_context::executor_type> strand_;
   beast::flat_buffer buffer_;
-  http::request<http::string_body> req_;
+  boost::optional<http::request_parser<http::string_body>> req_;
   std::shared_ptr<void> res_{nullptr};
 
   // Writes the message asynchronously back to the socket
@@ -76,4 +76,3 @@ class HttpSession : public std::enable_shared_from_this<HttpSession> {
 
 }  // namespace server
 }  // namespace onnxruntime
-

onnxruntime/test/server/integration_tests/model_zoo_tests.py

@@ -20,7 +20,7 @@ class ModelZooTests(unittest.TestCase):
     model_zoo_model_path = ''  # Required
     model_zoo_test_data_path = ''  # Required
     supported_opsets = ['opset_7', 'opset_8', 'opset_9']
-    skipped_models = ['tiny_yolov2']
+    skipped_models = []
 
     def test_models_from_model_zoo(self):
         json_request_headers = {

onnxruntime/test/server/integration_tests/test_util.py

@@ -55,9 +55,9 @@ def decode_base64_string(s, count_and_type):
     return r
 
 
-def compare_floats(a, b, rel_tol=0.0001):
-    if not math.isclose(a, b, rel_tol=rel_tol):
-        test_log('Not match with relative tolerance {0}: {1} and {2}'.format(rel_tol, a, b))
+def compare_floats(a, b, rel_tol=0.0001, abs_tol=0.0001):
+    if not math.isclose(a, b, rel_tol=rel_tol, abs_tol=abs_tol):
+        test_log('Not match with relative tolerance {0} and absolute tolerance {1}: {2} and {3}'.format(rel_tol, abs_tol, a, b))
         return False
 
     return True