1 // Copyright (C) 2007-2015 CEA/DEN, EDF R&D, OPEN CASCADE
3 // Copyright (C) 2003-2007 OPEN CASCADE, EADS/CCR, LIP6, CEA/DEN,
4 // CEDRAT, EDF R&D, LEG, PRINCIPIA R&D, BUREAU VERITAS
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, or (at your option) any later version.
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.
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
20 // See http://www.salome-platform.org/ or email : webmaster.salome@opencascade.com
23 #include "SALOME_ContainerManager.hxx"
24 #include "SALOME_ResourcesManager.hxx"
25 #include "SALOME_LoadRateManager.hxx"
26 #include "SALOME_NamingService.hxx"
27 #include "SALOME_ResourcesManager_Client.hxx"
28 #include "SALOME_ModuleCatalog.hh"
29 #include "Basics_Utils.hxx"
30 #include "Basics_DirUtils.hxx"
31 #include <sys/types.h>
38 #include "Utils_CorbaException.hxx"
42 #include <SALOMEconfig.h>
43 #include CORBA_CLIENT_HEADER(SALOME_Session)
51 #define getpid _getpid
54 #ifdef WITH_PACO_PARALLEL
58 const int SALOME_ContainerManager::TIME_OUT_TO_LAUNCH_CONT=60;
60 const char *SALOME_ContainerManager::_ContainerManagerNameInNS =
63 omni_mutex SALOME_ContainerManager::_numInstanceMutex;
65 Utils_Mutex SALOME_ContainerManager::_getenvMutex;
67 Utils_Mutex SALOME_ContainerManager::_systemMutex;
69 //=============================================================================
73 * Define a CORBA single thread policy for the server, which avoid to deal
74 * with non thread-safe usage like Change_Directory in SALOME naming service
76 //=============================================================================
78 SALOME_ContainerManager::SALOME_ContainerManager(CORBA::ORB_ptr orb, PortableServer::POA_var poa, SALOME_NamingService *ns)
81 MESSAGE("constructor");
83 _resManager = new SALOME_ResourcesManager_Client(ns);
85 PortableServer::POAManager_var pman = poa->the_POAManager();
86 _orb = CORBA::ORB::_duplicate(orb) ;
87 CORBA::PolicyList policies;
89 PortableServer::ThreadPolicy_var threadPol(poa->create_thread_policy(PortableServer::ORB_CTRL_MODEL));
90 policies[0] = PortableServer::ThreadPolicy::_duplicate(threadPol);
92 _poa = poa->create_POA("MThreadPOA",pman,policies);
94 PortableServer::ObjectId_var id = _poa->activate_object(this);
95 CORBA::Object_var obj = _poa->id_to_reference(id);
96 Engines::ContainerManager_var refContMan =
97 Engines::ContainerManager::_narrow(obj);
99 _NS->Register(refContMan,_ContainerManagerNameInNS);
100 _isAppliSalomeDefined = (GetenvThreadSafe("APPLI") != 0);
105 // the urifile name depends on pid of the process
106 std::stringstream urifile;
107 urifile << GetenvThreadSafe("HOME") << "/.urifile_" << getpid();
108 setenv("OMPI_URI_FILE",urifile.str().c_str(),1);
109 if( GetenvThreadSafe("OMPI_URI_FILE") != NULL ){
110 // get the pid of all ompi-server
111 std::set<pid_t> thepids1 = getpidofprogram("ompi-server");
112 // launch a new ompi-server
114 command = "ompi-server -r ";
115 command += GetenvThreadSafe("OMPI_URI_FILE");
116 int status=SystemThreadSafe(command.c_str());
118 throw SALOME_Exception("Error when launching ompi-server");
119 // get the pid of all ompi-server
120 std::set<pid_t> thepids2 = getpidofprogram("ompi-server");
121 // my ompi-server is the new one
122 std::set<pid_t>::const_iterator it;
123 for(it=thepids2.begin();it!=thepids2.end();it++)
124 if(thepids1.find(*it) == thepids1.end())
125 _pid_mpiServer = *it;
126 if(_pid_mpiServer < 0)
127 throw SALOME_Exception("Error when getting ompi-server id");
131 // get the pid of all hydra_nameserver
132 std::set<pid_t> thepids1 = getpidofprogram("hydra_nameserver");
133 // launch a new hydra_nameserver
135 command = "hydra_nameserver &";
136 SystemThreadSafe(command.c_str());
137 // get the pid of all hydra_nameserver
138 std::set<pid_t> thepids2 = getpidofprogram("hydra_nameserver");
139 // my hydra_nameserver is the new one
140 std::set<pid_t>::const_iterator it;
141 for(it=thepids2.begin();it!=thepids2.end();it++)
142 if(thepids1.find(*it) == thepids1.end())
143 _pid_mpiServer = *it;
147 MESSAGE("constructor end");
150 //=============================================================================
154 //=============================================================================
156 SALOME_ContainerManager::~SALOME_ContainerManager()
158 MESSAGE("destructor");
162 if( GetenvThreadSafe("OMPI_URI_FILE") != NULL ){
163 // kill my ompi-server
164 if( kill(_pid_mpiServer,SIGTERM) != 0 )
165 throw SALOME_Exception("Error when killing ompi-server");
167 int status=SystemThreadSafe("rm -f ${OMPI_URI_FILE}");
169 throw SALOME_Exception("Error when removing urifile");
172 // kill my hydra_nameserver
173 if(_pid_mpiServer > -1)
174 if( kill(_pid_mpiServer,SIGTERM) != 0 )
175 throw SALOME_Exception("Error when killing hydra_nameserver");
180 //=============================================================================
181 //! shutdown all the containers, then the ContainerManager servant
184 //=============================================================================
186 void SALOME_ContainerManager::Shutdown()
189 ShutdownContainers();
190 _NS->Destroy_Name(_ContainerManagerNameInNS);
191 PortableServer::ObjectId_var oid = _poa->servant_to_id(this);
192 _poa->deactivate_object(oid);
195 //=============================================================================
196 //! Loop on all the containers listed in naming service, ask shutdown on each
199 //=============================================================================
201 void SALOME_ContainerManager::ShutdownContainers()
203 MESSAGE("ShutdownContainers");
205 SALOME::Session_var session = SALOME::Session::_nil();
207 CORBA::Object_var objS = _NS->Resolve("/Kernel/Session");
208 if (!CORBA::is_nil(objS))
210 session = SALOME::Session::_narrow(objS);
211 if (!CORBA::is_nil(session))
212 pid = session->getPID();
216 isOK = _NS->Change_Directory("/Containers");
218 std::vector<std::string> vec = _NS->list_directory_recurs();
219 std::list<std::string> lstCont;
220 for(std::vector<std::string>::iterator iter = vec.begin();iter!=vec.end();iter++)
223 CORBA::Object_var obj=_NS->Resolve((*iter).c_str());
226 Engines::Container_var cont=Engines::Container::_narrow(obj);
227 if(!CORBA::is_nil(cont) && pid != cont->getPID())
228 lstCont.push_back((*iter));
230 catch(const CORBA::Exception& e)
232 // ignore this entry and continue
235 MESSAGE("Container list: ");
236 for(std::list<std::string>::iterator iter=lstCont.begin();iter!=lstCont.end();iter++){
239 for(std::list<std::string>::iterator iter=lstCont.begin();iter!=lstCont.end();iter++)
244 CORBA::Object_var obj=_NS->Resolve((*iter).c_str());
245 Engines::Container_var cont=Engines::Container::_narrow(obj);
246 if(!CORBA::is_nil(cont))
248 MESSAGE("ShutdownContainers: " << (*iter));
252 MESSAGE("ShutdownContainers: no container ref for " << (*iter));
254 catch(CORBA::SystemException& e)
256 INFOS("CORBA::SystemException ignored : " << e);
258 catch(CORBA::Exception&)
260 INFOS("CORBA::Exception ignored.");
264 INFOS("Unknown exception ignored.");
270 //=============================================================================
271 //! Give a suitable Container given constraints
273 * \param params Container Parameters required for the container
274 * \return the container or nil
276 //=============================================================================
277 Engines::Container_ptr SALOME_ContainerManager::GiveContainer(const Engines::ContainerParameters& params)
279 std::string machFile;
280 Engines::Container_ptr ret(Engines::Container::_nil());
282 // Step 0: Default mode is start
283 Engines::ContainerParameters local_params(params);
284 if (std::string(local_params.mode.in()) == "")
285 local_params.mode = CORBA::string_dup("start");
286 std::string mode = local_params.mode.in();
287 MESSAGE("[GiveContainer] starting with mode: " << mode);
289 // Step 1: Find Container for find and findorstart mode
290 if (mode == "find" || mode == "findorstart")
292 ret = FindContainer(params, params.resource_params.resList);
293 if(!CORBA::is_nil(ret))
299 MESSAGE("[GiveContainer] no container found");
309 // Step 2: Get all possibleResources from the parameters
310 // Consider only resources that can run containers
311 resourceParams resource_params = resourceParameters_CORBAtoCPP(local_params.resource_params);
312 resource_params.can_run_containers = true;
313 std::vector<std::string> possibleResources = _resManager->GetFittingResources(resource_params);
314 MESSAGE("[GiveContainer] - length of possible resources " << possibleResources.size());
315 std::vector<std::string> local_resources;
317 // Step 3: if mode is "get" keep only machines with existing containers
320 for(unsigned int i=0; i < possibleResources.size(); i++)
322 Engines::Container_ptr cont = FindContainer(params, possibleResources[i]);
325 if(!cont->_non_existent())
326 local_resources.push_back(possibleResources[i]);
328 catch(CORBA::Exception&) {}
331 // if local_resources is empty, we cannot give a container
332 if (local_resources.size() == 0)
334 MESSAGE("[GiveContainer] cannot find a container for mode get");
339 local_resources = possibleResources;
341 // Step 4: select the resource where to get/start the container
342 bool resource_available = true;
343 std::string resource_selected;
344 std::vector<std::string> resources = local_resources;
345 while (resource_available)
347 if (resources.size() == 0)
348 resource_available = false;
353 resource_selected = _resManager->Find(params.resource_params.policy.in(), resources);
354 // Remove resource_selected from vector
355 std::vector<std::string>::iterator it;
356 for (it=resources.begin() ; it < resources.end(); it++ )
357 if (*it == resource_selected)
363 catch(const SALOME_Exception &ex)
365 MESSAGE("[GiveContainer] Exception in ResourceManager find !: " << ex.what());
368 MESSAGE("[GiveContainer] Resource selected is: " << resource_selected);
370 // Step 5: Create container name
371 ParserResourcesType resource_definition = _resManager->GetResourceDefinition(resource_selected);
372 std::string hostname(resource_definition.HostName);
373 std::string containerNameInNS;
376 if ( params.nb_proc <= 0 )
379 nbproc = params.nb_proc;
382 if( GetenvThreadSafe("LIBBATCH_NODEFILE") != NULL )
383 machFile = machinesFile(nbproc);
385 catch(const SALOME_Exception & ex)
387 std::string err_msg = ex.what();
388 err_msg += params.container_name;
389 INFOS(err_msg.c_str());
392 // A mpi parallel container register on zero node in NS
393 containerNameInNS = _NS->BuildContainerNameForNS(params, GetMPIZeroNode(hostname,machFile).c_str());
396 containerNameInNS = _NS->BuildContainerNameForNS(params, hostname.c_str());
397 MESSAGE("[GiveContainer] Container name in the naming service: " << containerNameInNS);
399 // Step 6: check if the name exists in naming service
400 //if params.mode == "getorstart" or "get" use the existing container
401 //if params.mode == "start" shutdown the existing container before launching a new one with that name
403 { // critical section
404 Utils_Locker lock (&_giveContainerMutex1);
405 CORBA::Object_var obj = _NS->Resolve(containerNameInNS.c_str());
406 if (!CORBA::is_nil(obj))
410 Engines::Container_var cont=Engines::Container::_narrow(obj);
411 if(!cont->_non_existent())
413 if(std::string(params.mode.in())=="getorstart" || std::string(params.mode.in())=="get"){
414 return cont._retn(); /* the container exists and params.mode is getorstart or get use it*/
418 INFOS("[GiveContainer] A container is already registered with the name: " << containerNameInNS << ", shutdown the existing container");
419 cont->Shutdown(); // shutdown the registered container if it exists
423 catch(CORBA::Exception&)
425 INFOS("[GiveContainer] CORBA::Exception ignored when trying to get the container - we start a new one");
428 } // end critical section
429 Engines::Container_var cont = LaunchContainer(params, resource_selected, hostname, machFile, containerNameInNS);
430 if (!CORBA::is_nil(cont))
432 INFOS("[GiveContainer] container " << containerNameInNS << " launched");
437 INFOS("[GiveContainer] Failed to launch container on resource " << resource_selected);
442 // We were not able to launch the container
443 INFOS("[GiveContainer] Cannot launch the container on the following selected resources:")
444 std::vector<std::string>::iterator it;
445 for (it=local_resources.begin() ; it < local_resources.end(); it++ )
446 INFOS("[GiveContainer] " << *it)
450 Engines::Container_ptr
451 SALOME_ContainerManager::LaunchContainer(const Engines::ContainerParameters& params,
452 const std::string & resource_selected,
453 const std::string & hostname,
454 const std::string & machFile,
455 const std::string & containerNameInNS)
457 std::string user,command,logFilename,tmpFileName;
459 Engines::Container_ptr ret(Engines::Container::_nil());
460 {//start of critical section
461 Utils_Locker lock (&_giveContainerMutex1);
462 // Step 1: type of container: PaCO, Exe, Mpi or Classic
463 // Mpi already tested in step 5, specific code on BuildCommandToLaunch Local/Remote Container methods
464 // TODO -> separates Mpi from Classic/Exe
466 std::string container_exe = "SALOME_Container"; // Classic container
467 Engines::ContainerParameters local_params(params);
471 CORBA::String_var container_exe_tmp;
472 CORBA::Object_var obj = _NS->Resolve("/Kernel/ModulCatalog");
473 SALOME_ModuleCatalog::ModuleCatalog_var Catalog = SALOME_ModuleCatalog::ModuleCatalog::_narrow(obj) ;
474 if (CORBA::is_nil (Catalog))
476 INFOS("[GiveContainer] Module Catalog is not found -> cannot launch a container");
479 // Loop through component list
480 for(unsigned int i=0; i < local_params.resource_params.componentList.length(); i++)
482 const char* compoi = local_params.resource_params.componentList[i];
483 SALOME_ModuleCatalog::Acomponent_var compoInfo = Catalog->GetComponent(compoi);
484 if (CORBA::is_nil (compoInfo))
488 SALOME_ModuleCatalog::ImplType impl=compoInfo->implementation_type();
489 container_exe_tmp=compoInfo->implementation_name();
490 if(impl==SALOME_ModuleCatalog::CEXE)
494 INFOS("ContainerManager Error: you can't have 2 CEXE component in the same container" );
495 return Engines::Container::_nil();
497 MESSAGE("[GiveContainer] Exe container found !: " << container_exe_tmp);
498 container_exe = container_exe_tmp.in();
503 catch (ServiceUnreachable&)
505 INFOS("Caught exception: Naming Service Unreachable");
510 INFOS("Caught unknown exception.");
514 // Step 2: test resource
515 // Only if an application directory is set
516 if(hostname != Kernel_Utils::GetHostname() && _isAppliSalomeDefined)
518 // Preparing remote command
519 std::string command = "";
520 const ParserResourcesType resInfo(_resManager->GetResourceDefinition(resource_selected));
521 command = getCommandToRunRemoteProcess(resInfo.Protocol, resInfo.HostName, resInfo.UserName);
522 if (resInfo.AppliPath != "")
523 command += resInfo.AppliPath;
526 ASSERT(GetenvThreadSafe("APPLI"));
527 command += GetenvThreadSafe("APPLI");
529 command += "/runRemote.sh ";
530 ASSERT(GetenvThreadSafe("NSHOST"));
531 command += GetenvThreadSafe("NSHOST"); // hostname of CORBA name server
533 ASSERT(GetenvThreadSafe("NSPORT"));
534 command += GetenvThreadSafe("NSPORT"); // port of CORBA name server
535 command += " \"ls /tmp >/dev/null 2>&1\"";
537 // Launch remote command
538 int status = SystemThreadSafe(command.c_str());
541 // Error on resource - cannot launch commands
542 INFOS("[LaunchContainer] Cannot launch commands on machine " << hostname);
543 INFOS("[LaunchContainer] Command was " << command);
545 INFOS("[LaunchContainer] Command status is " << WEXITSTATUS(status));
547 return Engines::Container::_nil();
551 // Step 3: start a new container
552 // Check if a PaCO container
554 if (std::string(local_params.parallelLib.in()) != "")
556 ret = StartPaCOPPContainer(params, resource_selected);
559 // Other type of containers...
560 MESSAGE("[GiveContainer] Try to launch a new container on " << resource_selected);
561 // if a parallel container is launched in batch job, command is: "mpirun -np nbproc -machinefile nodesfile SALOME_MPIContainer"
562 if( GetenvThreadSafe("LIBBATCH_NODEFILE") != NULL && params.isMPI )
563 command = BuildCommandToLaunchLocalContainer(params, machFile, container_exe, tmpFileName);
564 // if a container is launched on localhost, command is "SALOME_Container" or "mpirun -np nbproc SALOME_MPIContainer"
565 else if(hostname == Kernel_Utils::GetHostname())
566 command = BuildCommandToLaunchLocalContainer(params, machFile, container_exe, tmpFileName);
567 // if a container is launched in remote mode, command is "ssh resource_selected SALOME_Container" or "ssh resource_selected mpirun -np nbproc SALOME_MPIContainer"
569 command = BuildCommandToLaunchRemoteContainer(resource_selected, params, container_exe);
571 //redirect stdout and stderr in a file
573 logFilename=GetenvThreadSafe("TEMP");
575 user = GetenvThreadSafe( "USERNAME" );
577 user = GetenvThreadSafe( "USER" );
579 char* val = GetenvThreadSafe("SALOME_TMP_DIR");
582 struct stat file_info;
583 stat(val, &file_info);
584 bool is_dir = S_ISDIR(file_info.st_mode);
585 if (is_dir)logFilename=val;
586 else std::cerr << "SALOME_TMP_DIR environment variable is not a directory use /tmp instead" << std::endl;
590 logFilename += _NS->ContainerName(params)+"_"+ resource_selected +"_"+user;
591 std::ostringstream tmp;
592 tmp << "_" << getpid();
593 logFilename += tmp.str();
594 logFilename += ".log" ;
595 command += " > " + logFilename + " 2>&1";
596 MakeTheCommandToBeLaunchedASync(command);
598 // launch container with a system call
599 status=SystemThreadSafe(command.c_str());
600 }//end of critical of section
604 INFOS("[LaunchContainer] command failed (system command status -1): " << command);
605 RmTmpFile(tmpFileName); // command file can be removed here
606 return Engines::Container::_nil();
608 else if (status == 217)
610 INFOS("[LaunchContainer] command failed (system command status 217): " << command);
611 RmTmpFile(tmpFileName); // command file can be removed here
612 return Engines::Container::_nil();
616 // Step 4: Wait for the container
617 int count(GetTimeOutToLoaunchServer());
618 INFOS("[GiveContainer] waiting " << count << " second steps container " << containerNameInNS);
619 while (CORBA::is_nil(ret) && count)
623 MESSAGE("[GiveContainer] step " << count << " Waiting for container on " << resource_selected);
624 CORBA::Object_var obj(_NS->Resolve(containerNameInNS.c_str()));
625 ret=Engines::Container::_narrow(obj);
627 if (CORBA::is_nil(ret))
629 INFOS("[GiveContainer] was not able to launch container " << containerNameInNS);
633 // Setting log file name
634 logFilename=":"+logFilename;
635 logFilename="@"+Kernel_Utils::GetHostname()+logFilename;//threadsafe
636 logFilename=user+logFilename;
637 ret->logfilename(logFilename.c_str());
638 RmTmpFile(tmpFileName); // command file can be removed here
644 //=============================================================================
645 //! Find a container given constraints (params) on a list of machines (possibleComputers)
646 //! agy : this method is ThreadSafe
650 //=============================================================================
652 Engines::Container_ptr SALOME_ContainerManager::FindContainer(const Engines::ContainerParameters& params, const Engines::ResourceList& possibleResources)
654 MESSAGE("[FindContainer] FindContainer on " << possibleResources.length() << " resources");
655 for(unsigned int i=0; i < possibleResources.length();i++)
657 Engines::Container_ptr cont = FindContainer(params, possibleResources[i].in());
658 if(!CORBA::is_nil(cont))
661 MESSAGE("[FindContainer] no container found");
662 return Engines::Container::_nil();
665 //=============================================================================
666 //! Find a container given constraints (params) on a machine (theMachine)
667 //! agy : this method is ThreadSafe
671 //=============================================================================
673 Engines::Container_ptr
674 SALOME_ContainerManager::FindContainer(const Engines::ContainerParameters& params, const std::string& resource)
676 ParserResourcesType resource_definition = _resManager->GetResourceDefinition(resource);
677 std::string hostname(resource_definition.HostName);
678 std::string containerNameInNS(_NS->BuildContainerNameForNS(params, hostname.c_str()));
679 MESSAGE("[FindContainer] Try to find a container " << containerNameInNS << " on resource " << resource);
680 CORBA::Object_var obj = _NS->Resolve(containerNameInNS.c_str());
683 if(obj->_non_existent())
684 return Engines::Container::_nil();
686 return Engines::Container::_narrow(obj);
688 catch(const CORBA::Exception& e)
690 return Engines::Container::_nil();
695 bool isPythonContainer(const char* ContainerName);
697 //=============================================================================
699 * This is no longer valid (C++ container are also python containers)
701 //=============================================================================
702 bool isPythonContainer(const char* ContainerName)
704 return false; // VSR 02/08/2013: Python containers are no more supported
706 int len = strlen(ContainerName);
709 if (strcmp(ContainerName + len - 2, "Py") == 0)
715 //=============================================================================
717 * Builds the script to be launched
719 * If SALOME Application not defined ($APPLI),
720 * see BuildTempFileToLaunchRemoteContainer()
722 * Else rely on distant configuration. Command is under the form (example):
723 * ssh user@machine distantPath/runRemote.sh hostNS portNS WORKINGDIR workingdir \
724 * SALOME_Container containerName &"
726 * - where user is ommited if not specified in CatalogResources,
727 * - where distant path is always relative to user@machine $HOME, and
728 * equal to $APPLI if not specified in CatalogResources,
729 * - where hostNS is the hostname of CORBA naming server (set by scripts to
730 * use to launch SALOME and servers in $APPLI: runAppli.sh, runRemote.sh)
731 * - where portNS is the port used by CORBA naming server (set by scripts to
732 * use to launch SALOME and servers in $APPLI: runAppli.sh, runRemote.sh)
733 * - where workingdir is the requested working directory for the container.
734 * If WORKINGDIR (and workingdir) is not present the working dir will be $HOME
736 //=============================================================================
739 SALOME_ContainerManager::BuildCommandToLaunchRemoteContainer(const std::string& resource_name, const Engines::ContainerParameters& params, const std::string& container_exe) const
741 std::string command,tmpFileName;
742 if (!_isAppliSalomeDefined)
743 command = BuildTempFileToLaunchRemoteContainer(resource_name, params, tmpFileName);
747 const ParserResourcesType resInfo(_resManager->GetResourceDefinition(resource_name));
751 if ( params.nb_proc <= 0 )
754 nbproc = params.nb_proc;
757 // "ssh -l user machine distantPath/runRemote.sh hostNS portNS WORKINGDIR workingdir \
758 // SALOME_Container containerName &"
759 command = getCommandToRunRemoteProcess(resInfo.Protocol, resInfo.HostName, resInfo.UserName);
761 if (resInfo.AppliPath != "")
762 command += resInfo.AppliPath; // path relative to user@machine $HOME
765 ASSERT(GetenvThreadSafe("APPLI"));
766 command += GetenvThreadSafe("APPLI"); // path relative to user@machine $HOME
769 command += "/runRemote.sh ";
771 ASSERT(GetenvThreadSafe("NSHOST"));
772 command += GetenvThreadSafe("NSHOST"); // hostname of CORBA name server
775 ASSERT(GetenvThreadSafe("NSPORT"));
776 command += GetenvThreadSafe("NSPORT"); // port of CORBA name server
778 std::string wdir = params.workingdir.in();
781 command += " WORKINGDIR ";
783 if(wdir == "$TEMPDIR")
785 command += wdir; // requested working directory
791 command += " mpirun -np ";
792 std::ostringstream o;
796 command += "-x PATH,LD_LIBRARY_PATH,OMNIORB_CONFIG,SALOME_trace ";
797 #elif defined(OPEN_MPI)
798 if( GetenvThreadSafe("OMPI_URI_FILE") == NULL )
799 command += "-x PATH -x LD_LIBRARY_PATH -x OMNIORB_CONFIG -x SALOME_trace";
801 command += "-x PATH -x LD_LIBRARY_PATH -x OMNIORB_CONFIG -x SALOME_trace -ompi-server file:";
802 command += GetenvThreadSafe("OMPI_URI_FILE");
805 command += "-nameserver " + Kernel_Utils::GetHostname();
807 command += " SALOME_MPIContainer ";
810 command += " " +container_exe+ " ";
812 command += _NS->ContainerName(params);
814 AddOmninamesParams(command);
816 MESSAGE("command =" << command);
822 //=============================================================================
824 * builds the command to be launched.
826 //=============================================================================
827 std::string SALOME_ContainerManager::BuildCommandToLaunchLocalContainer(const Engines::ContainerParameters& params, const std::string& machinesFile, const std::string& container_exe, std::string& tmpFileName) const
829 tmpFileName = BuildTemporaryFileName();
833 std::ostringstream o;
839 if ( params.nb_proc <= 0 )
842 nbproc = params.nb_proc;
846 if( GetenvThreadSafe("LIBBATCH_NODEFILE") != NULL )
847 o << "-machinefile " << machinesFile << " ";
850 o << "-x PATH,LD_LIBRARY_PATH,OMNIORB_CONFIG,SALOME_trace ";
851 #elif defined(OPEN_MPI)
852 if( GetenvThreadSafe("OMPI_URI_FILE") == NULL )
853 o << "-x PATH -x LD_LIBRARY_PATH -x OMNIORB_CONFIG -x SALOME_trace";
856 o << "-x PATH -x LD_LIBRARY_PATH -x OMNIORB_CONFIG -x SALOME_trace -ompi-server file:";
857 o << GetenvThreadSafe("OMPI_URI_FILE");
860 o << "-nameserver " + Kernel_Utils::GetHostname();
863 if (isPythonContainer(params.container_name))
864 o << " pyMPI SALOME_ContainerPy.py ";
866 o << " SALOME_MPIContainer ";
871 std::string wdir=params.workingdir.in();
874 // a working directory is requested
875 if(wdir == "$TEMPDIR")
877 // a new temporary directory is requested
878 std::string dir = Kernel_Utils::GetTmpDir();
880 o << "cd /d " << dir << std::endl;
882 o << "cd " << dir << ";";
888 // a permanent directory is requested use it or create it
890 o << "mkdir " + wdir << std::endl;
891 o << "cd /D " + wdir << std::endl;
893 o << "mkdir -p " << wdir << " && cd " << wdir + ";";
898 if (isPythonContainer(params.container_name))
899 o << "SALOME_ContainerPy.py ";
901 o << container_exe + " ";
905 o << _NS->ContainerName(params);
907 AddOmninamesParams(o);
909 std::ofstream command_file( tmpFileName.c_str() );
910 command_file << o.str();
911 command_file.close();
914 chmod(tmpFileName.c_str(), 0x1ED);
916 command = tmpFileName;
918 MESSAGE("Command is file ... " << command);
919 MESSAGE("Command is ... " << o.str());
924 //=============================================================================
926 * removes the generated temporary file in case of a remote launch.
927 * This method is thread safe
929 //=============================================================================
931 void SALOME_ContainerManager::RmTmpFile(std::string& tmpFileName)
933 int lenght = tmpFileName.size();
937 std::string command = "del /F ";
939 std::string command = "rm ";
942 command += tmpFileName.substr(0, lenght - 3 );
944 command += tmpFileName;
946 SystemThreadSafe(command.c_str());
947 //if dir is empty - remove it
948 std::string tmp_dir = Kernel_Utils::GetDirByPath( tmpFileName );
949 if ( Kernel_Utils::IsEmptyDir( tmp_dir ) )
952 command = "del /F " + tmp_dir;
954 command = "rmdir " + tmp_dir;
956 SystemThreadSafe(command.c_str());
961 //=============================================================================
963 * add to command all options relative to naming service.
965 //=============================================================================
967 void SALOME_ContainerManager::AddOmninamesParams(std::string& command) const
969 std::ostringstream oss;
970 AddOmninamesParams(oss);
974 //=============================================================================
976 * add to command all options relative to naming service.
978 //=============================================================================
980 void SALOME_ContainerManager::AddOmninamesParams(std::ostream& fileStream) const
982 AddOmninamesParams(fileStream,_NS);
985 //=============================================================================
987 * add to command all options relative to naming service.
989 //=============================================================================
991 void SALOME_ContainerManager::AddOmninamesParams(std::ostream& fileStream, SALOME_NamingService *ns)
993 CORBA::String_var iorstr(ns->getIORaddr());
994 fileStream << "ORBInitRef NameService=";
995 fileStream << iorstr;
998 void SALOME_ContainerManager::MakeTheCommandToBeLaunchedASync(std::string& command)
1001 command = "%PYTHONBIN% -c \"import win32pm ; win32pm.spawnpid(r'" + command + "', '')\"";
1007 int SALOME_ContainerManager::GetTimeOutToLoaunchServer()
1009 int count(TIME_OUT_TO_LAUNCH_CONT);
1010 if (GetenvThreadSafe("TIMEOUT_TO_LAUNCH_CONTAINER") != 0)
1012 std::string new_count_str(GetenvThreadSafe("TIMEOUT_TO_LAUNCH_CONTAINER"));
1014 std::istringstream ss(new_count_str);
1015 if (!(ss >> new_count))
1017 INFOS("[LaunchContainer] TIMEOUT_TO_LAUNCH_CONTAINER should be an int");
1025 void SALOME_ContainerManager::SleepInSecond(int ellapseTimeInSecond)
1028 sleep( ellapseTimeInSecond ) ;
1030 int timeInMS(1000*ellapseTimeInSecond);
1035 //=============================================================================
1037 * generate a file name in /tmp directory
1039 //=============================================================================
1041 std::string SALOME_ContainerManager::BuildTemporaryFileName()
1043 //build more complex file name to support multiple salome session
1044 std::string aFileName = Kernel_Utils::GetTmpFileName();
1048 aFileName += ".bat";
1053 //=============================================================================
1055 * Builds in a temporary file the script to be launched.
1057 * Used if SALOME Application ($APPLI) is not defined.
1058 * The command is build with data from CatalogResources, in which every path
1059 * used on remote computer must be defined.
1061 //=============================================================================
1063 std::string SALOME_ContainerManager::BuildTempFileToLaunchRemoteContainer (const std::string& resource_name, const Engines::ContainerParameters& params, std::string& tmpFileName) const
1067 tmpFileName = BuildTemporaryFileName();
1068 std::ofstream tempOutputFile;
1069 tempOutputFile.open(tmpFileName.c_str(), std::ofstream::out );
1070 const ParserResourcesType resInfo(_resManager->GetResourceDefinition(resource_name));
1071 tempOutputFile << "#! /bin/sh" << std::endl;
1075 tempOutputFile << "export SALOME_trace=local" << std::endl; // mkr : 27.11.2006 : PAL13967 - Distributed supervision graphs - Problem with "SALOME_trace"
1076 //tempOutputFile << "source " << resInfo.PreReqFilePath << endl;
1082 tempOutputFile << "mpirun -np ";
1085 if ( params.nb_proc <= 0 )
1088 nbproc = params.nb_proc;
1090 std::ostringstream o;
1092 tempOutputFile << nbproc << " ";
1094 tempOutputFile << "-x PATH,LD_LIBRARY_PATH,OMNIORB_CONFIG,SALOME_trace ";
1095 #elif defined(OPEN_MPI)
1096 if( GetenvThreadSafe("OMPI_URI_FILE") == NULL )
1097 tempOutputFile << "-x PATH -x LD_LIBRARY_PATH -x OMNIORB_CONFIG -x SALOME_trace";
1099 tempOutputFile << "-x PATH -x LD_LIBRARY_PATH -x OMNIORB_CONFIG -x SALOME_trace -ompi-server file:";
1100 tempOutputFile << GetenvThreadSafe("OMPI_URI_FILE");
1102 #elif defined(MPICH)
1103 tempOutputFile << "-nameserver " + Kernel_Utils::GetHostname();
1107 tempOutputFile << GetenvThreadSafe("KERNEL_ROOT_DIR") << "/bin/salome/";
1111 if (isPythonContainer(params.container_name))
1112 tempOutputFile << " pyMPI SALOME_ContainerPy.py ";
1114 tempOutputFile << " SALOME_MPIContainer ";
1119 if (isPythonContainer(params.container_name))
1120 tempOutputFile << "SALOME_ContainerPy.py ";
1122 tempOutputFile << "SALOME_Container ";
1125 tempOutputFile << _NS->ContainerName(params) << " -";
1126 AddOmninamesParams(tempOutputFile);
1127 tempOutputFile << " &" << std::endl;
1128 tempOutputFile.flush();
1129 tempOutputFile.close();
1131 chmod(tmpFileName.c_str(), 0x1ED);
1134 // --- Build command
1136 std::string command;
1138 if (resInfo.Protocol == rsh)
1141 std::string commandRcp = "rcp ";
1142 commandRcp += tmpFileName;
1144 commandRcp += resInfo.HostName;
1146 commandRcp += tmpFileName;
1147 status = SystemThreadSafe(commandRcp.c_str());
1150 else if (resInfo.Protocol == ssh)
1153 std::string commandRcp = "scp ";
1154 commandRcp += tmpFileName;
1156 commandRcp += resInfo.HostName;
1158 commandRcp += tmpFileName;
1159 status = SystemThreadSafe(commandRcp.c_str());
1162 else if (resInfo.Protocol == srun)
1164 command = "srun -n 1 -N 1 --share --nodelist=";
1165 std::string commandRcp = "rcp ";
1166 commandRcp += tmpFileName;
1168 commandRcp += resInfo.HostName;
1170 commandRcp += tmpFileName;
1171 status = SystemThreadSafe(commandRcp.c_str());
1174 throw SALOME_Exception("Unknown protocol");
1177 throw SALOME_Exception("Error of connection on remote host");
1179 command += resInfo.HostName;
1181 command += tmpFileName;
1189 std::string SALOME_ContainerManager::GetMPIZeroNode(const std::string machine, const std::string machinesFile) const
1192 std::string zeronode;
1193 std::string command;
1194 std::string tmpFile = BuildTemporaryFileName();
1196 if( GetenvThreadSafe("LIBBATCH_NODEFILE") == NULL )
1198 if (_isAppliSalomeDefined)
1200 const ParserResourcesType resInfo(_resManager->GetResourceDefinition(machine));
1202 if (resInfo.Protocol == rsh)
1204 else if (resInfo.Protocol == ssh)
1206 else if (resInfo.Protocol == srun)
1207 command = "srun -n 1 -N 1 --share --nodelist=";
1209 throw SALOME_Exception("Unknown protocol");
1211 if (resInfo.UserName != "")
1214 command += resInfo.UserName;
1218 command += resInfo.HostName;
1221 if (resInfo.AppliPath != "")
1222 command += resInfo.AppliPath; // path relative to user@machine $HOME
1225 ASSERT(GetenvThreadSafe("APPLI"));
1226 command += GetenvThreadSafe("APPLI"); // path relative to user@machine $HOME
1229 command += "/runRemote.sh ";
1231 ASSERT(GetenvThreadSafe("NSHOST"));
1232 command += GetenvThreadSafe("NSHOST"); // hostname of CORBA name server
1235 ASSERT(GetenvThreadSafe("NSPORT"));
1236 command += GetenvThreadSafe("NSPORT"); // port of CORBA name server
1238 command += " mpirun -np 1 hostname -s > " + tmpFile;
1241 command = "mpirun -np 1 hostname -s > " + tmpFile;
1244 command = "mpirun -np 1 -machinefile " + machinesFile + " hostname -s > " + tmpFile;
1246 status = SystemThreadSafe(command.c_str());
1248 std::ifstream fp(tmpFile.c_str(),std::ios::in);
1249 while(fp >> zeronode);
1257 std::string SALOME_ContainerManager::machinesFile(const int nbproc)
1260 std::string nodesFile = GetenvThreadSafe("LIBBATCH_NODEFILE");
1261 std::string machinesFile = Kernel_Utils::GetTmpFileName();
1262 std::ifstream fpi(nodesFile.c_str(),std::ios::in);
1263 std::ofstream fpo(machinesFile.c_str(),std::ios::out);
1265 _numInstanceMutex.lock();
1267 for(int i=0;i<_nbprocUsed;i++)
1270 for(int i=0;i<nbproc;i++)
1272 fpo << tmp << std::endl;
1274 throw SALOME_Exception("You need more processors than batch session have allocated for you! Unable to launch the mpi container: ");
1276 _nbprocUsed += nbproc;
1280 _numInstanceMutex.unlock();
1282 return machinesFile;
1286 std::set<pid_t> SALOME_ContainerManager::getpidofprogram(const std::string program)
1288 std::set<pid_t> thepids;
1289 std::string tmpFile = Kernel_Utils::GetTmpFileName();
1292 cmd = "pidof " + program + " > " + tmpFile;
1293 SystemThreadSafe(cmd.c_str());
1294 std::ifstream fpi(tmpFile.c_str(),std::ios::in);
1295 while(fpi >> thepid){
1296 thepids.insert(atoi(thepid.c_str()));
1301 std::string SALOME_ContainerManager::getCommandToRunRemoteProcess(AccessProtocolType protocol,
1302 const std::string & hostname,
1303 const std::string & username)
1305 std::ostringstream command;
1312 command << "-l " << username << " ";
1314 command << hostname << " ";
1320 command << "-l " << username << " ";
1322 command << hostname << " ";
1325 // no need to redefine the user with srun, the job user is taken by default
1326 // (note: for srun, user id can be specified with " --uid=<user>")
1327 command << "srun -n 1 -N 1 --share --nodelist=" << hostname << " ";
1330 command << "pbsdsh -o -h " << hostname << " ";
1333 command << "blaunch -no-shell " << hostname << " ";
1336 throw SALOME_Exception("Unknown protocol");
1339 return command.str();
1343 SALOME_ContainerManager::checkPaCOParameters(Engines::ContainerParameters & params, std::string resource_selected)
1347 // Step 1 : check ContainerParameters
1348 // Check container_name, has to be defined
1349 if (std::string(params.container_name.in()) == "")
1351 INFOS("[checkPaCOParameters] You must define a container_name to launch a PaCO++ container");
1354 // Check parallelLib
1355 std::string parallelLib = params.parallelLib.in();
1356 if (parallelLib != "Mpi" && parallelLib != "Dummy")
1358 INFOS("[checkPaCOParameters] parallelLib is not correctly defined");
1359 INFOS("[checkPaCOParameters] you can chosse between: Mpi and Dummy");
1360 INFOS("[checkPaCOParameters] you entered: " << parallelLib);
1364 if (params.nb_proc <= 0)
1366 INFOS("[checkPaCOParameters] You must define a nb_proc > 0");
1370 // Step 2 : check resource_selected
1371 const ParserResourcesType resource_definition = _resManager->GetResourceDefinition(resource_selected);
1372 //std::string protocol = resource_definition->protocol.in();
1373 std::string username = resource_definition.UserName;
1374 std::string applipath = resource_definition.AppliPath;
1376 //if (protocol == "" || username == "" || applipath == "")
1377 if (username == "" || applipath == "")
1379 INFOS("[checkPaCOParameters] resource selected is not well defined");
1380 INFOS("[checkPaCOParameters] resource name: " << resource_definition.Name);
1381 INFOS("[checkPaCOParameters] resource hostname: " << resource_definition.HostName);
1382 INFOS("[checkPaCOParameters] resource protocol: " << resource_definition.getAccessProtocolTypeStr());
1383 INFOS("[checkPaCOParameters] resource username: " << username);
1384 INFOS("[checkPaCOParameters] resource applipath: " << applipath);
1391 char *SALOME_ContainerManager::GetenvThreadSafe(const char *name)
1392 {// getenv is not thread safe. See man 7 pthread.
1393 Utils_Locker lock (&_getenvMutex);
1394 return getenv(name);
1397 int SALOME_ContainerManager::SystemThreadSafe(const char *command)
1399 Utils_Locker lock (&_systemMutex);
1400 return system(command);
1403 #ifdef WITH_PACO_PARALLEL
1405 //=============================================================================
1407 * Start a suitable PaCO++ Parallel Container in a list of machines.
1408 * \param params Container Parameters required for the container
1409 * \return CORBA container reference.
1411 //=============================================================================
1412 Engines::Container_ptr
1413 SALOME_ContainerManager::StartPaCOPPContainer(const Engines::ContainerParameters& params_const,
1414 std::string resource_selected)
1416 CORBA::Object_var obj;
1417 PaCO::InterfaceManager_var container_proxy;
1418 Engines::Container_ptr ret = Engines::Container::_nil();
1419 Engines::ContainerParameters params(params_const);
1420 params.resource_params.name = CORBA::string_dup(resource_selected.c_str());
1422 // Step 0 : Check parameters
1423 if (!checkPaCOParameters(params, resource_selected))
1425 INFOS("[StartPaCOPPContainer] check parameters failed ! see logs...");
1429 // Step 1 : Starting a new parallel container !
1430 INFOS("[StartPaCOPPContainer] Starting a PaCO++ parallel container");
1431 INFOS("[StartPaCOPPContainer] on resource : " << resource_selected);
1433 // Step 2 : Get a MachineFile for the parallel container
1434 std::string machine_file_name = _resManager->getMachineFile(resource_selected,
1436 params.parallelLib.in());
1438 if (machine_file_name == "")
1440 INFOS("[StartPaCOPPContainer] Machine file generation failed");
1444 // Step 3 : starting parallel container proxy
1445 std::string command_proxy("");
1446 std::string proxy_machine;
1449 command_proxy = BuildCommandToLaunchPaCOProxyContainer(params, machine_file_name, proxy_machine);
1451 catch(const SALOME_Exception & ex)
1453 INFOS("[StartPaCOPPContainer] Exception in BuildCommandToLaunchPaCOContainer");
1457 obj = LaunchPaCOProxyContainer(command_proxy, params, proxy_machine);
1458 if (CORBA::is_nil(obj))
1460 INFOS("[StartPaCOPPContainer] LaunchPaCOContainer for proxy returns NIL !");
1463 container_proxy = PaCO::InterfaceManager::_narrow(obj);
1464 MESSAGE("[StartPaCOPPContainer] PaCO container proxy is launched");
1466 // Step 4 : starting parallel container nodes
1467 std::string command_nodes("");
1468 SALOME_ContainerManager::actual_launch_machine_t nodes_machines;
1471 command_nodes = BuildCommandToLaunchPaCONodeContainer(params, machine_file_name, nodes_machines, proxy_machine);
1473 catch(const SALOME_Exception & ex)
1475 INFOS("[StarPaCOPPContainer] Exception in BuildCommandToLaunchPaCONodeContainer");
1480 std::string container_generic_node_name = std::string(params.container_name.in()) + std::string("Node");
1481 bool result = LaunchPaCONodeContainer(command_nodes, params, container_generic_node_name, nodes_machines);
1484 INFOS("[StarPaCOPPContainer] LaunchPaCONodeContainer failed !");
1485 // Il faut tuer le proxy
1488 Engines::Container_var proxy = Engines::Container::_narrow(container_proxy);
1493 INFOS("[StarPaCOPPContainer] Exception catched from proxy Shutdown...");
1498 // Step 4 : connecting nodes and the proxy to actually create a parallel container
1499 for (int i = 0; i < params.nb_proc; i++)
1501 std::ostringstream tmp;
1503 std::string proc_number = tmp.str();
1504 std::string container_node_name = container_generic_node_name + proc_number;
1506 std::string theNodeMachine(nodes_machines[i]);
1507 std::string containerNameInNS = _NS->BuildContainerNameForNS(container_node_name.c_str(), theNodeMachine.c_str());
1508 obj = _NS->Resolve(containerNameInNS.c_str());
1509 if (CORBA::is_nil(obj))
1511 INFOS("[StarPaCOPPContainer] CONNECTION FAILED From Naming Service !");
1512 INFOS("[StarPaCOPPContainer] Container name is " << containerNameInNS);
1517 MESSAGE("[StarPaCOPPContainer] Deploying node : " << container_node_name);
1518 PaCO::InterfaceParallel_var node = PaCO::InterfaceParallel::_narrow(obj);
1520 MESSAGE("[StarPaCOPPContainer] node " << container_node_name << " is deployed");
1522 catch(CORBA::SystemException& e)
1524 INFOS("[StarPaCOPPContainer] Exception in deploying node : " << containerNameInNS);
1525 INFOS("CORBA::SystemException : " << e);
1528 catch(CORBA::Exception& e)
1530 INFOS("[StarPaCOPPContainer] Exception in deploying node : " << containerNameInNS);
1531 INFOS("CORBA::Exception" << e);
1536 INFOS("[StarPaCOPPContainer] Exception in deploying node : " << containerNameInNS);
1537 INFOS("Unknown exception !");
1542 // Step 5 : starting parallel container
1545 MESSAGE ("[StarPaCOPPContainer] Starting parallel object");
1546 container_proxy->start();
1547 MESSAGE ("[StarPaCOPPContainer] Parallel object is started");
1548 ret = Engines::Container::_narrow(container_proxy);
1550 catch(CORBA::SystemException& e)
1552 INFOS("Caught CORBA::SystemException. : " << e);
1554 catch(PortableServer::POA::ServantAlreadyActive&)
1556 INFOS("Caught CORBA::ServantAlreadyActiveException");
1558 catch(CORBA::Exception&)
1560 INFOS("Caught CORBA::Exception.");
1562 catch(std::exception& exc)
1564 INFOS("Caught std::exception - "<<exc.what());
1568 INFOS("Caught unknown exception.");
1574 SALOME_ContainerManager::BuildCommandToLaunchPaCOProxyContainer(const Engines::ContainerParameters& params,
1575 std::string machine_file_name,
1576 std::string & proxy_hostname)
1578 // In the proxy case, we always launch a Dummy Proxy
1579 std::string exe_name = "SALOME_ParallelContainerProxyDummy";
1580 std::string container_name = params.container_name.in();
1582 // Convert nb_proc in string
1583 std::ostringstream tmp_string;
1584 tmp_string << params.nb_proc;
1585 std::string nb_proc_str = tmp_string.str();
1587 // Get resource definition
1588 ParserResourcesType resource_definition =
1589 _resManager->GetResourceDefinition(params.resource_params.name.in());
1592 std::string hostname;
1593 std::ifstream machine_file(machine_file_name.c_str());
1594 std::getline(machine_file, hostname, ' ');
1595 size_t found = hostname.find('\n');
1596 if (found!=std::string::npos)
1597 hostname.erase(found, 1); // Remove \n
1598 proxy_hostname = hostname;
1599 MESSAGE("[BuildCommandToLaunchPaCOProxyContainer] machine file name extracted is " << hostname);
1602 bool remote_execution = false;
1603 if (hostname != std::string(Kernel_Utils::GetHostname()))
1605 MESSAGE("[BuildCommandToLaunchPaCOProxyContainer] remote machine case detected !");
1606 remote_execution = true;
1609 // Log environnement
1610 std::string log_type("");
1611 char * get_val = GetenvThreadSafe("PARALLEL_LOG");
1615 // Generating the command
1616 std::string command_begin("");
1617 std::string command_end("");
1618 std::ostringstream command;
1620 LogConfiguration(log_type, "proxy", container_name, hostname, command_begin, command_end);
1621 command << command_begin;
1623 // Adding connection command
1624 // We can only have a remote execution with
1625 // a SALOME application
1626 if (remote_execution)
1628 ASSERT(GetenvThreadSafe("NSHOST"));
1629 ASSERT(GetenvThreadSafe("NSPORT"));
1631 command << resource_definition.getAccessProtocolTypeStr();
1633 command << resource_definition.UserName;
1634 command << " " << hostname;
1635 command << " " << resource_definition.AppliPath;
1636 command << "/runRemote.sh ";
1637 command << GetenvThreadSafe("NSHOST") << " "; // hostname of CORBA name server
1638 command << GetenvThreadSafe("NSPORT") << " "; // port of CORBA name server
1641 command << exe_name;
1642 command << " " << container_name;
1643 command << " Dummy";
1644 command << " " << hostname;
1645 command << " " << nb_proc_str;
1647 AddOmninamesParams(command);
1650 command << command_end;
1651 MESSAGE("[BuildCommandToLaunchPaCOProxyContainer] Command is: " << command.str());
1653 return command.str();
1657 SALOME_ContainerManager::BuildCommandToLaunchPaCONodeContainer(const Engines::ContainerParameters& params,
1658 const std::string & machine_file_name,
1659 SALOME_ContainerManager::actual_launch_machine_t & vect_machine,
1660 const std::string & proxy_hostname)
1663 std::string exe_name = "SALOME_ParallelContainerNode";
1664 exe_name += params.parallelLib.in();
1665 std::string container_name = params.container_name.in();
1667 // Convert nb_proc in string
1668 std::ostringstream nb_proc_stream;
1669 nb_proc_stream << params.nb_proc;
1671 // Get resource definition
1672 ParserResourcesType resource_definition =
1673 _resManager->GetResourceDefinition(params.resource_params.name.in());
1675 // Log environnement
1676 std::string log_type("");
1677 char * get_val = GetenvThreadSafe("PARALLEL_LOG");
1681 // Now the command is different according to paralleLib
1682 std::ostringstream command_nodes;
1683 std::ifstream machine_file(machine_file_name.c_str());
1684 if (std::string(params.parallelLib.in()) == "Dummy")
1686 for (int i= 0; i < params.nb_proc; i++)
1689 std::string hostname;
1690 std::getline(machine_file, hostname);
1691 MESSAGE("[BuildCommandToLaunchPaCONodeContainer] machine file name extracted is " << hostname);
1694 bool remote_execution = false;
1695 if (hostname != std::string(Kernel_Utils::GetHostname()))
1697 MESSAGE("[BuildCommandToLaunchPaCONodeContainer] remote machine case detected !");
1698 remote_execution = true;
1701 // For each node we have a new command
1702 // Generating the command
1703 std::ostringstream command_node_stream;
1704 std::string command_node_begin("");
1705 std::string command_node_end("");
1706 std::ostringstream node_number;
1708 std::string container_node_name = container_name + node_number.str();
1709 LogConfiguration(log_type, "node", container_node_name, hostname, command_node_begin, command_node_end);
1711 // Adding connection command
1712 // We can only have a remote execution with
1713 // a SALOME application
1714 if (remote_execution)
1716 ASSERT(GetenvThreadSafe("NSHOST"));
1717 ASSERT(GetenvThreadSafe("NSPORT"));
1719 command_node_stream << resource_definition.getAccessProtocolTypeStr();
1720 command_node_stream << " -l ";
1721 command_node_stream << resource_definition.UserName;
1722 command_node_stream << " " << hostname;
1723 command_node_stream << " " << resource_definition.AppliPath;
1724 command_node_stream << "/runRemote.sh ";
1725 command_node_stream << GetenvThreadSafe("NSHOST") << " "; // hostname of CORBA name server
1726 command_node_stream << GetenvThreadSafe("NSPORT") << " "; // port of CORBA name server
1729 command_node_stream << exe_name;
1730 command_node_stream << " " << container_name;
1731 command_node_stream << " " << params.parallelLib.in();
1732 command_node_stream << " " << proxy_hostname;
1733 command_node_stream << " " << node_number.str();
1734 command_node_stream << " -";
1735 AddOmninamesParams(command_node_stream);
1737 command_nodes << command_node_begin << command_node_stream.str() << command_node_end;
1738 vect_machine.push_back(hostname);
1742 else if (std::string(params.parallelLib.in()) == "Mpi")
1745 std::string hostname;
1746 std::getline(machine_file, hostname, ' ');
1747 MESSAGE("[BuildCommandToLaunchPaCONodeContainer] machine file name extracted is " << hostname);
1750 bool remote_execution = false;
1751 if (hostname != std::string(Kernel_Utils::GetHostname()))
1753 MESSAGE("[BuildCommandToLaunchPaCONodeContainer] remote machine case detected !");
1754 remote_execution = true;
1757 // In case of Mpi and Remote, we copy machine_file in the applipath
1758 // scp mpi_machine_file user@machine:Path
1759 std::ostringstream command_remote_stream;
1760 std::string::size_type last = machine_file_name.find_last_of("/");
1761 if (last == std::string::npos)
1764 if (resource_definition.Protocol == rsh)
1765 command_remote_stream << "rcp ";
1767 command_remote_stream << "scp ";
1768 command_remote_stream << machine_file_name << " ";
1769 command_remote_stream << resource_definition.UserName << "@";
1770 command_remote_stream << hostname << ":" << resource_definition.AppliPath;
1771 command_remote_stream << "/" << machine_file_name.substr(last+1);
1773 int status = SystemThreadSafe(command_remote_stream.str().c_str());
1776 INFOS("copy of the MPI machine file failed ! - sorry !");
1780 // Generating the command
1781 std::string command_begin("");
1782 std::string command_end("");
1784 LogConfiguration(log_type, "nodes", container_name, hostname, command_begin, command_end);
1785 command_nodes << command_begin;
1787 // Adding connection command
1788 // We can only have a remote execution with
1789 // a SALOME application
1790 if (remote_execution)
1792 ASSERT(GetenvThreadSafe("NSHOST"));
1793 ASSERT(GetenvThreadSafe("NSPORT"));
1795 command_nodes << resource_definition.getAccessProtocolTypeStr();
1796 command_nodes << " -l ";
1797 command_nodes << resource_definition.UserName;
1798 command_nodes << " " << hostname;
1799 command_nodes << " " << resource_definition.AppliPath;
1800 command_nodes << "/runRemote.sh ";
1801 command_nodes << GetenvThreadSafe("NSHOST") << " "; // hostname of CORBA name server
1802 command_nodes << GetenvThreadSafe("NSPORT") << " "; // port of CORBA name server
1805 if (resource_definition.mpi == lam)
1807 command_nodes << "mpiexec -ssi boot ";
1808 command_nodes << "-machinefile " << machine_file_name << " ";
1809 command_nodes << "-n " << params.nb_proc;
1813 command_nodes << "mpirun -np " << params.nb_proc;
1815 command_nodes << " " << exe_name;
1816 command_nodes << " " << container_name;
1817 command_nodes << " " << params.parallelLib.in();
1818 command_nodes << " " << proxy_hostname;
1819 command_nodes << " -";
1820 AddOmninamesParams(command_nodes);
1822 // We don't put hostname, because nodes are registered in the resource of the proxy
1823 for (int i= 0; i < params.nb_proc; i++)
1824 vect_machine.push_back(proxy_hostname);
1826 command_nodes << command_end;
1828 return command_nodes.str();
1832 SALOME_ContainerManager::LogConfiguration(const std::string & log_type,
1833 const std::string & exe_type,
1834 const std::string & container_name,
1835 const std::string & hostname,
1836 std::string & begin,
1839 if(log_type == "xterm")
1841 begin = "xterm -e \"export LD_LIBRARY_PATH=$LD_LIBRARY_PATH; export PATH=$PATH;";
1844 else if(log_type == "xterm_debug")
1846 begin = "xterm -e \"export LD_LIBRARY_PATH=$LD_LIBRARY_PATH; export PATH=$PATH;";
1851 // default into a file...
1852 std::string logFilename = "/tmp/" + container_name + "_" + hostname + "_" + exe_type + "_";
1853 logFilename += std::string(GetenvThreadSafe("USER")) + ".log";
1854 end = " > " + logFilename + " 2>&1 & ";
1859 SALOME_ContainerManager::LaunchPaCOProxyContainer(const std::string& command,
1860 const Engines::ContainerParameters& params,
1861 const std::string & hostname)
1863 PaCO::InterfaceManager_ptr container_proxy = PaCO::InterfaceManager::_nil();
1865 MESSAGE("[LaunchPaCOProxyContainer] Launch command");
1866 int status = SystemThreadSafe(command.c_str());
1868 INFOS("[LaunchPaCOProxyContainer] failed : system command status -1");
1869 return container_proxy;
1871 else if (status == 217) {
1872 INFOS("[LaunchPaCOProxyContainer] failed : system command status 217");
1873 return container_proxy;
1876 int count(GetTimeOutToLoaunchServer());
1877 CORBA::Object_var obj = CORBA::Object::_nil();
1878 std::string containerNameInNS = _NS->BuildContainerNameForNS(params.container_name.in(),
1880 MESSAGE("[LaunchParallelContainer] Waiting for Parallel Container proxy : " << containerNameInNS);
1882 while (CORBA::is_nil(obj) && count)
1886 obj = _NS->Resolve(containerNameInNS.c_str());
1891 container_proxy = PaCO::InterfaceManager::_narrow(obj);
1893 catch(CORBA::SystemException& e)
1895 INFOS("[StarPaCOPPContainer] Exception in _narrow after LaunchParallelContainer for proxy !");
1896 INFOS("CORBA::SystemException : " << e);
1897 return container_proxy;
1899 catch(CORBA::Exception& e)
1901 INFOS("[StarPaCOPPContainer] Exception in _narrow after LaunchParallelContainer for proxy !");
1902 INFOS("CORBA::Exception" << e);
1903 return container_proxy;
1907 INFOS("[StarPaCOPPContainer] Exception in _narrow after LaunchParallelContainer for proxy !");
1908 INFOS("Unknown exception !");
1909 return container_proxy;
1911 if (CORBA::is_nil(container_proxy))
1913 INFOS("[StarPaCOPPContainer] PaCO::InterfaceManager::_narrow returns NIL !");
1914 return container_proxy;
1919 //=============================================================================
1920 /*! This method launches the parallel container.
1921 * It will may be placed on the ressources manager.
1923 * \param command to launch
1924 * \param container's parameters
1925 * \param name of the container
1927 * \return CORBA container reference
1929 //=============================================================================
1931 SALOME_ContainerManager::LaunchPaCONodeContainer(const std::string& command,
1932 const Engines::ContainerParameters& params,
1933 const std::string& name,
1934 SALOME_ContainerManager::actual_launch_machine_t & vect_machine)
1936 INFOS("[LaunchPaCONodeContainer] Launch command");
1937 int status = SystemThreadSafe(command.c_str());
1939 INFOS("[LaunchPaCONodeContainer] failed : system command status -1");
1942 else if (status == 217) {
1943 INFOS("[LaunchPaCONodeContainer] failed : system command status 217");
1947 INFOS("[LaunchPaCONodeContainer] Waiting for the nodes of the parallel container");
1948 // We are waiting all the nodes
1949 for (int i = 0; i < params.nb_proc; i++)
1951 CORBA::Object_var obj = CORBA::Object::_nil();
1952 std::string theMachine(vect_machine[i]);
1954 std::ostringstream tmp;
1956 std::string proc_number = tmp.str();
1957 std::string container_node_name = name + proc_number;
1958 std::string containerNameInNS = _NS->BuildContainerNameForNS((char*) container_node_name.c_str(), theMachine.c_str());
1959 INFOS("[LaunchPaCONodeContainer] Waiting for Parallel Container node " << containerNameInNS << " on " << theMachine);
1960 int count(GetTimeOutToLoaunchServer());
1961 while (CORBA::is_nil(obj) && count) {
1964 obj = _NS->Resolve(containerNameInNS.c_str());
1966 if (CORBA::is_nil(obj))
1968 INFOS("[LaunchPaCONodeContainer] Launch of node failed (or not found) !");
1977 Engines::Container_ptr
1978 SALOME_ContainerManager::StartPaCOPPContainer(const Engines::ContainerParameters& params,
1979 std::string resource_selected)
1981 Engines::Container_ptr ret = Engines::Container::_nil();
1982 INFOS("[StarPaCOPPContainer] is disabled !");
1983 INFOS("[StarPaCOPPContainer] recompile SALOME Kernel to enable PaCO++ parallel extension");
1988 SALOME_ContainerManager::BuildCommandToLaunchPaCOProxyContainer(const Engines::ContainerParameters& params,
1989 std::string machine_file_name,
1990 std::string & proxy_hostname)
1996 SALOME_ContainerManager::BuildCommandToLaunchPaCONodeContainer(const Engines::ContainerParameters& params,
1997 const std::string & machine_file_name,
1998 SALOME_ContainerManager::actual_launch_machine_t & vect_machine,
1999 const std::string & proxy_hostname)
2004 SALOME_ContainerManager::LogConfiguration(const std::string & log_type,
2005 const std::string & exe_type,
2006 const std::string & container_name,
2007 const std::string & hostname,
2008 std::string & begin,
2014 SALOME_ContainerManager::LaunchPaCOProxyContainer(const std::string& command,
2015 const Engines::ContainerParameters& params,
2016 const std::string& hostname)
2018 CORBA::Object_ptr ret = CORBA::Object::_nil();
2023 SALOME_ContainerManager::LaunchPaCONodeContainer(const std::string& command,
2024 const Engines::ContainerParameters& params,
2025 const std::string& name,
2026 SALOME_ContainerManager::actual_launch_machine_t & vect_machine)