Make tcp_server retry binding the socket (#176)

It sometimes happens that binding the socket fails since it is already
in use (e.g. when restarting the driver too fast). This commit should
retry binding instead of simply throwing an exception

Author: Felix Exner (fexner)

include/ur_client_library/comm/tcp_server.h

@@ -35,6 +35,7 @@
 #include <unistd.h>
 
 #include <atomic>
+#include <chrono>
 #include <functional>
 #include <thread>
 
@@ -60,7 +61,17 @@ class TCPServer
 {
 public:
   TCPServer() = delete;
-  TCPServer(const int port);
+  /*!
+   * \brief Create a TCPServer object
+   *
+   * \param port Port on which to operate. The port will be bound to the process creating the
+   * object.
+   * \param max_num_tries If binding the socket fails, it will be retried this many times. If 0 is
+   * specified, binding the socket will be tried indefinitely.
+   * \param reconnection_time Wait time in between binding attempts.
+   */
+  explicit TCPServer(const int port, const size_t max_num_tries = 0,
+                     const std::chrono::milliseconds reconnection_time = std::chrono::seconds(1));
   virtual ~TCPServer();
 
   /*!
@@ -147,7 +158,7 @@ class TCPServer
 
 private:
   void init();
-  void bind();
+  void bind(const size_t max_num_tries, const std::chrono::milliseconds reconnection_time);
   void startListen();
 
   //! Handles connection events

src/comm/tcp_server.cpp

@@ -42,10 +42,11 @@ namespace urcl
 {
 namespace comm
 {
-TCPServer::TCPServer(const int port) : port_(port), maxfd_(0), max_clients_allowed_(0)
+TCPServer::TCPServer(const int port, const size_t max_num_tries, const std::chrono::milliseconds reconnection_time)
+  : port_(port), maxfd_(0), max_clients_allowed_(0)
 {
   init();
-  bind();
+  bind(max_num_tries, reconnection_time);
   startListen();
 }
 
@@ -124,21 +125,38 @@ void TCPServer::shutdown()
   }
 }
 
-void TCPServer::bind()
+void TCPServer::bind(const size_t max_num_tries, const std::chrono::milliseconds reconnection_time)
 {
   struct sockaddr_in server_addr;
   server_addr.sin_family = AF_INET;
 
   // INADDR_ANY is a special constant that signalizes "ANY IFACE",
   server_addr.sin_addr.s_addr = htonl(INADDR_ANY);
   server_addr.sin_port = htons(port_);
-  int err = ::bind(listen_fd_, (struct sockaddr*)&server_addr, sizeof(server_addr));
-  if (err == -1)
+  int err = -1;
+  size_t connection_counter = 0;
+  do
   {
-    std::ostringstream ss;
-    ss << "Failed to bind socket for port " << port_ << " to address. Reason: " << strerror(errno);
-    throw std::system_error(std::error_code(errno, std::generic_category()), ss.str());
-  }
+    err = ::bind(listen_fd_, (struct sockaddr*)&server_addr, sizeof(server_addr));
+    if (err == -1)
+    {
+      std::ostringstream ss;
+      ss << "Failed to bind socket for port " << port_ << " to address. Reason: " << strerror(errno);
+
+      if (connection_counter++ < max_num_tries || max_num_tries == 0)
+      {
+        std::this_thread::sleep_for(reconnection_time);
+        ss << "Retrying in " << std::chrono::duration_cast<std::chrono::duration<float>>(reconnection_time).count()
+           << " seconds";
+        URCL_LOG_WARN("%s", ss.str().c_str());
+      }
+      else
+      {
+        throw std::system_error(std::error_code(errno, std::generic_category()), ss.str());
+      }
+    }
+  } while (err == -1 && (connection_counter <= max_num_tries || max_num_tries == 0));
+
   URCL_LOG_DEBUG("Bound %d:%d to FD %d", server_addr.sin_addr.s_addr, port_, (int)listen_fd_);
 
   FD_SET(listen_fd_, &masterfds_);

tests/test_tcp_server.cpp

@@ -170,7 +170,7 @@ TEST_F(TCPServerTest, socket_creation)
   comm::TCPServer server(port_);
 
   // Shouldn't be able to create antoher server on same port
-  EXPECT_THROW(comm::TCPServer server2(port_), std::system_error);
+  EXPECT_THROW(comm::TCPServer server2(port_, 1, std::chrono::milliseconds(1)), std::system_error);
 
   server.start();
 
@@ -326,6 +326,12 @@ TEST_F(TCPServerTest, client_connections)
   EXPECT_FALSE(server.write(client2_fd, data, len, written));
   EXPECT_FALSE(server.write(client3_fd, data, len, written));
 }
+TEST_F(TCPServerTest, check_address_already_in_use)
+{
+  comm::TCPServer blocking_server(12321);
+
+  EXPECT_THROW(comm::TCPServer test_server(12321, 2, std::chrono::milliseconds(500)), std::system_error);
+}
 
 int main(int argc, char* argv[])
 {