Improve error checking in MPI backend

Author: marcelkrueger

nestkernel/exceptions.cpp

@@ -23,6 +23,7 @@
 #include "exceptions.h"
 
 // C++ includes:
+#include <mpi.h>
 #include <sstream>
 
 // Generated includes:
@@ -398,6 +399,31 @@ nest::MPIPortsFileUnknown::message() const
       << " which specifies the folder with files containing the MPI ports";
   return msg.str();
 }
+
+
+std::string
+nest::MPIPortsFileMissing::message() const
+{
+  std::ostringstream msg;
+  msg << "The node with ID " << node_id_ << " has no file that contains the MPI address.";
+  return msg.str();
+}
+
+std::string
+nest::MPIErrorCode::message() const
+{
+
+  char errmsg[ 256 ];
+  int len;
+
+  MPI_Error_string( error_code_, errmsg, &len );
+  std::string error;
+  error.assign( errmsg, len );
+
+  std::ostringstream msg;
+  msg << "MPI Error: " << error;
+  return msg.str();
+}
 #endif
 
 std::string

nestkernel/exceptions.h

@@ -1188,6 +1188,34 @@ class MPIPortsFileUnknown : public KernelException
 
   std::string message() const;
 
+private:
+  const size_t node_id_;
+};
+
+class MPIErrorCode : public KernelException
+{
+public:
+  explicit MPIErrorCode( const int error_code )
+    : error_code_( error_code )
+  {
+  }
+
+  std::string message() const;
+
+private:
+  int error_code_;
+};
+
+class MPIPortsFileMissing : public KernelException
+{
+public:
+  explicit MPIPortsFileMissing( const size_t node_id )
+    : node_id_( node_id )
+  {
+  }
+
+  std::string message() const;
+
 private:
   const size_t node_id_;
 };

nestkernel/recording_backend_mpi.cpp

@@ -174,11 +174,13 @@ nest::RecordingBackendMPI::prepare()
   // 2) connect the thread to the MPI process it needs to be connected to
   for ( auto& it_comm : commMap_ )
   {
-    MPI_Comm_connect( it_comm.first.data(),
-      MPI_INFO_NULL,
-      0,
-      MPI_COMM_WORLD,
-      std::get< 1 >( it_comm.second ) ); // should use the status for handle error
+    int ret =
+      MPI_Comm_connect( it_comm.first.data(), MPI_INFO_NULL, 0, MPI_COMM_WORLD, std::get< 1 >( it_comm.second ) );
+
+    if ( ret != MPI_SUCCESS )
+    {
+      throw MPIErrorCode( ret );
+    }
     std::ostringstream msg;
     msg << "Connect to " << it_comm.first.data() << "\n";
     LOG( M_INFO, "MPI Record connect", msg.str() );
@@ -383,8 +385,12 @@ nest::RecordingBackendMPI::get_port( const size_t index_node, const std::string&
   }
 
   basename << "/" << index_node << ".txt";
-  std::cout << basename.rdbuf() << std::endl;
+  std::cout << basename.str() << std::endl;
   std::ifstream file( basename.str() );
+  if ( !file.good() )
+  {
+    throw MPIPortsFileMissing( index_node );
+  }
   if ( file.is_open() )
   {
     getline( file, *port_name );

nestkernel/stimulation_backend_mpi.cpp

@@ -188,11 +188,13 @@ nest::StimulationBackendMPI::prepare()
   // 2) connect the master thread to the MPI process it needs to be connected to
   for ( auto& it_comm : commMap_ )
   {
-    MPI_Comm_connect( it_comm.first.data(),
-      MPI_INFO_NULL,
-      0,
-      MPI_COMM_WORLD,
-      std::get< 0 >( it_comm.second ) ); // should use the status for handle error
+    int ret =
+      MPI_Comm_connect( it_comm.first.data(), MPI_INFO_NULL, 0, MPI_COMM_WORLD, std::get< 0 >( it_comm.second ) );
+
+    if ( ret != MPI_SUCCESS )
+    {
+      throw MPIErrorCode( ret );
+    }
     std::ostringstream msg;
     msg << "Connect to " << it_comm.first.data() << "\n";
     LOG( M_INFO, "MPI Input connect", msg.str() );
@@ -311,8 +313,12 @@ nest::StimulationBackendMPI::get_port( const size_t index_node, const std::strin
   }
   // add the id of the device to the path
   basename << "/" << index_node << ".txt";
-  std::cout << basename.rdbuf() << std::endl;
+  std::cout << basename.str() << std::endl;
   std::ifstream file( basename.str() );
+  if ( !file.good() )
+  {
+    throw MPIPortsFileMissing( index_node );
+  }
 
   // read the file
   if ( file.is_open() )