src/ParallelContainer/SALOME_ParallelContainerNodeMpi.cxx

   1 //  SALOME ParallelContainerNodeMpi : Launch mpi PaCO++ object nodes
   2 //
   3 //  Copyright (C) 2007 OPEN CASCADE, EADS/CCR, LIP6, CEA/DEN,
   4 //  CEDRAT, EDF R&D, LEG, PRINCIPIA R&D, BUREAU VERITAS
   5 //
   6 //  This library is free software; you can redistribute it and/or
   7 //  modify it under the terms of the GNU Lesser General Public
   8 //  License as published by the Free Software Foundation; either
   9 //  version 2.1 of the License.
  10 //
  11 //  This library is distributed in the hope that it will be useful,
  12 //  but WITHOUT ANY WARRANTY; without even the implied warranty of
  13 //  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14 //  Lesser General Public License for more details.
  15 //
  16 //  You should have received a copy of the GNU Lesser General Public
  17 //  License along with this library; if not, write to the Free Software
  18 //  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
  19 //
  20 //  See http://www.opencascade.org/SALOME/ or email : webmaster.salome@opencascade.org
  21 //
  22 //
  23 //
  24 //  File   : SALOME_ParallelContainerNodeMpi.cxx
  25 //  Author : André Ribes, EDF
  26 //  Module : SALOME PARALLEL
  27
  28 #include <iostream>
  29 #include <string>
  30 #include <stdio.h>
  31
  32 #ifndef WNT
  33 #include <unistd.h>
  34 #else
  35 #include <process.h>
  36 #endif
  37
  38 #include "SALOME_ParallelContainer_i.hxx"
  39
  40 // PaCO++ include
  41 #include <paco_omni.h>
  42 #include <paco_mpi.h>
  43
  44 #include <mpi.h>
  45
  46 #include "SALOME_NamingService.hxx"
  47
  48 #include "utilities.h"
  49 #include "Utils_ORB_INIT.hxx"
  50 #include "Utils_SINGLETON.hxx"
  51 #include "SALOMETraceCollector.hxx"
  52 #include "OpUtil.hxx"
  53
  54 using namespace std;
  55
  56 #ifdef _DEBUG_
  57 #include <signal.h>
  58
  59 void test(int sigval) {
  60         cerr << "!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!" << endl;
  61         cerr << "!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!" << endl;
  62         cerr << "SIGSEGV in :" << getpid() << endl;
  63         cerr << "!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!" << endl;
  64         cerr << "!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!" << endl;
  65         while (1) {}
  66 }
  67
  68 void handler(int t) {
  69         cerr << "!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!" << endl;
  70         cerr << "!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!" << endl;
  71         cerr << "SIGSEGV in :" << getpid() << endl;
  72         cerr << "!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!" << endl;
  73         cerr << "!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!" << endl;
  74         while (1) {}
  75 }
  76 #endif
  77
  78 typedef void (*sighandler_t)(int);
  79 sighandler_t setsig(int sig, sighandler_t handler)
  80 {
  81   struct sigaction context, ocontext;
  82   context.sa_handler = handler;
  83   sigemptyset(&context.sa_mask);
  84   context.sa_flags = 0;
  85   if (sigaction(sig, &context, &ocontext) == -1)
  86     return SIG_ERR;
  87   return ocontext.sa_handler;
  88 }
  89
  90 void AttachDebugger()
  91 {
  92   if(getenv ("DEBUGGER"))
  93     {
  94       std::stringstream exec;
  95       exec << "$DEBUGGER SALOME_ParallelContainerNodeMpi " << getpid() << "&";
  96       std::cerr << exec.str() << std::endl;
  97       system(exec.str().c_str());
  98       while(1);
  99     }
 100 }
 101
 102 void Handler(int theSigId)
 103 {
 104   std::cerr << "SIGSEGV: "  << std::endl;
 105   AttachDebugger();
 106   //to exit or not to exit
 107   exit(1);
 108 }
 109
 110 void terminateHandler(void)
 111 {
 112   std::cerr << "Terminate: not managed exception !"  << std::endl;
 113   AttachDebugger();
 114 }
 115
 116 void unexpectedHandler(void)
 117 {
 118   std::cerr << "Unexpected: unexpected exception !"  << std::endl;
 119   AttachDebugger();
 120 }
 121
 122 int main(int argc, char* argv[])
 123 {
 124         INFOS("Launching a parallel Mpi container node");
 125
 126 #ifdef _DEBUG_
 127 //      struct sigaction action;
 128 //      action.sa_handler = &test;
 129 //      sigaction(SIGSEGV, &action, NULL);
 130 #endif
 131
 132         // MPI Init
 133         int provided;
 134         MPI_Init_thread(&argc, &argv, MPI_THREAD_MULTIPLE ,&provided);
 135
 136   if(getenv ("DEBUGGER"))
 137     {
 138   std::cerr << "Unexpected: unexpected exception !"  << std::endl;
 139       setsig(SIGSEGV,&Handler);
 140       set_terminate(&terminateHandler);
 141       set_unexpected(&unexpectedHandler);
 142     }
 143 #ifdef _DEBUG_
 144         cerr << "Level MPI_THREAD_SINGLE : " << MPI_THREAD_SINGLE << endl;
 145         cerr << "Level MPI_THREAD_SERIALIZED : " << MPI_THREAD_SERIALIZED << endl;
 146         cerr << "Level MPI_THREAD_FUNNELED : " << MPI_THREAD_FUNNELED << endl;
 147         cerr << "Level MPI_THREAD_MULTIPLE : " << MPI_THREAD_MULTIPLE << endl;
 148         cerr << "Level provided : " << provided << endl;
 149 #endif
 150         // Initialise the ORB.
 151         ORB_INIT &init = *SINGLETON_<ORB_INIT>::Instance();
 152         ASSERT(SINGLETON_<ORB_INIT>::IsAlreadyExisting());
 153         CORBA::ORB_var orb = init(0, 0);
 154         //CORBA::ORB_var orb = CORBA::ORB_init(argc, argv);
 155
 156         // Code pour choisir le reseau infiniband .....
 157 /*      string hostname_temp = GetHostname();
 158         hostent * t = gethostbyname(hostname_temp.c_str());
 159         cerr << " AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA " << t->h_addr << " " << hostname_temp << endl;
 160         cerr << t->h_addr << endl;
 161         in_addr * address=(in_addr * ) t->h_addr;
 162         cerr << inet_ntoa(* address) << endl;
 163         string ip = inet_ntoa(* address);
 164         cerr << " AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA " << endl;
 165         string com = "giop:tcp:" + ip + ":";
 166         const char* options[][2] = { { "endPoint", com.c_str() }, { 0, 0 } };
 167         CORBA::ORB_var orb = CORBA::ORB_init(argc, argv, "omniORB4", options);
 168 */
 169         char * containerName = "";
 170         if(argc > 1) {
 171                 containerName = argv[1];
 172         }
 173
 174         char * hostname = "";
 175         if(argc > 3) {
 176                 hostname = argv[3];
 177         }
 178
 179         try {
 180                 CORBA::Object_var obj = orb->resolve_initial_references("RootPOA");
 181                 PortableServer::POA_var root_poa = PortableServer::POA::_narrow(obj);
 182                 PortableServer::POAManager_var pman = root_poa->the_POAManager();
 183
 184 #ifndef WNT
 185                 // add this container to the kill list
 186                 char aCommand[100];
 187                 sprintf(aCommand, "addToKillList.py %d SALOME_ParallelContainerNodeMpi", getpid());
 188                 system(aCommand);
 189 #endif
 190
 191                 SALOME_NamingService * ns = new SALOME_NamingService(CORBA::ORB::_duplicate(orb));
 192                 // On récupère le proxy
 193                 string proxyNameInNS = ns->BuildContainerNameForNS(containerName, hostname);
 194                 obj = ns->Resolve(proxyNameInNS.c_str());
 195                 char * proxy_ior = orb->object_to_string(obj);
 196
 197                 // Node creation
 198                 string name(containerName);
 199                 string node_name = name + "Node";
 200                 Engines_Parallel_Container_i * servant = new Engines_Parallel_Container_i(CORBA::ORB::_duplicate(orb), proxy_ior,
 201                                                                                                                                                                                                                                                                                                                         root_poa,
 202                                                                                                                                                                                                                                                                                                                         (char*) node_name.c_str(),
 203                                                                                                                                                                                                                                                                                                                         argc, argv);
 204                 // PaCO++ init
 205                 paco_fabrique_manager * pfm = paco_getFabriqueManager();
 206                 pfm->register_com("mpi", new paco_mpi_fabrique());
 207                 pfm->register_thread("omni", new paco_omni_fabrique());
 208
 209                 // Global context
 210                 PaCO_operation * global_ptr = servant->getContext("global_paco_context");
 211                 MPI_Comm group = MPI_COMM_WORLD;
 212                 global_ptr->setLibCom("mpi", &group);
 213                 global_ptr->setLibThread("omni");
 214
 215                 // Activation
 216                 PortableServer::ObjectId * _id = root_poa->activate_object(servant);
 217                 servant->set_id(_id);
 218                 obj = root_poa->id_to_reference(*_id);
 219
 220                 // In the NamingService
 221                 string hostname = GetHostname();
 222
 223                 int myid;
 224                 MPI_Comm_rank(MPI_COMM_WORLD, &myid);
 225                 char buffer [5];
 226                 snprintf(buffer, 5, "%d", myid);
 227                 node_name = node_name + buffer;
 228                 string _containerName = ns->BuildContainerNameForNS((char*) node_name.c_str(),
 229                                                                                                                                                                                                                                 hostname.c_str());
 230                 cerr << "---------" << _containerName << "----------" << endl;
 231                 ns->Register(obj, _containerName.c_str());
 232                 pman->activate();
 233                 orb->run();
 234         }
 235         catch(CORBA::SystemException&)
 236         {
 237                 INFOS("Caught CORBA::SystemException.");
 238         }
 239         catch(PortableServer::POA::ServantAlreadyActive&)
 240         {
 241                 INFOS("Caught CORBA::ServantAlreadyActiveException");
 242         }
 243         catch(CORBA::Exception&)
 244         {
 245                 INFOS("Caught CORBA::Exception.");
 246         }
 247         catch(std::exception& exc)
 248         {
 249                 INFOS("Caught std::exception - "<<exc.what());
 250         }
 251         catch(...)
 252         {
 253                 INFOS("Caught unknown exception.");
 254         }
 255
 256         MPI_Finalize();
 257
 258         return 0 ;
 259 }
 260