X-Git-Url: http://git.salome-platform.org/gitweb/?a=blobdiff_plain;f=src%2FContainer%2FSALOME_ContainerManager.cxx;h=a22a39c2fc7a866980f68da5e639bb35c9fbe0e3;hb=6154bba6020465244a40f05f034ba93aabf304a2;hp=3c7fdf021649b1be5d1454d72a60648294a30bb0;hpb=7d2fe213bdf5bf962ce11e253020c9d3e0bc1cce;p=modules%2Fkernel.git diff --git a/src/Container/SALOME_ContainerManager.cxx b/src/Container/SALOME_ContainerManager.cxx index 3c7fdf021..a22a39c2f 100644 --- a/src/Container/SALOME_ContainerManager.cxx +++ b/src/Container/SALOME_ContainerManager.cxx @@ -1,4 +1,4 @@ -// Copyright (C) 2007-2008 CEA/DEN, EDF R&D, OPEN CASCADE +// Copyright (C) 2007-2010 CEA/DEN, EDF R&D, OPEN CASCADE // // Copyright (C) 2003-2007 OPEN CASCADE, EADS/CCR, LIP6, CEA/DEN, // CEDRAT, EDF R&D, LEG, PRINCIPIA R&D, BUREAU VERITAS @@ -19,6 +19,7 @@ // // See http://www.salome-platform.org/ or email : webmaster.salome@opencascade.com // + #include "SALOME_ContainerManager.hxx" #include "SALOME_NamingService.hxx" #include "SALOME_ModuleCatalog.hh" @@ -31,23 +32,25 @@ #endif #include #include "Utils_CorbaException.hxx" -#include "Batch_Date.hxx" +#include + +#ifdef WNT +#include +#define getpid _getpid +#endif #ifdef WITH_PACO_PARALLEL -#include "PaCO++.h" +#include "PaCOPP.hxx" #endif #define TIME_OUT_TO_LAUNCH_CONT 61 -using namespace std; - -vector SALOME_ContainerManager::_batchLaunchedContainers; - -vector::iterator SALOME_ContainerManager::_batchLaunchedContainersIter; - const char *SALOME_ContainerManager::_ContainerManagerNameInNS = "/ContainerManager"; +omni_mutex SALOME_ContainerManager::_numInstanceMutex; + + //============================================================================= /*! * Constructor @@ -57,12 +60,11 @@ const char *SALOME_ContainerManager::_ContainerManagerNameInNS = */ //============================================================================= -SALOME_ContainerManager::SALOME_ContainerManager(CORBA::ORB_ptr orb, PortableServer::POA_var poa, SALOME_ResourcesManager *rm, SALOME_NamingService *ns) +SALOME_ContainerManager::SALOME_ContainerManager(CORBA::ORB_ptr orb, PortableServer::POA_var poa, SALOME_ResourcesManager *rm, SALOME_NamingService *ns):_nbprocUsed(0) { MESSAGE("constructor"); _NS = ns; _ResManager = rm; - _id=0; PortableServer::POAManager_var pman = poa->the_POAManager(); _orb = CORBA::ORB::_duplicate(orb) ; @@ -80,8 +82,27 @@ SALOME_ContainerManager::SALOME_ContainerManager(CORBA::ORB_ptr orb, PortableSer Engines::ContainerManager::_narrow(obj); _NS->Register(refContMan,_ContainerManagerNameInNS); - _MpiStarted = false; _isAppliSalomeDefined = (getenv("APPLI") != 0); + +#ifdef HAVE_MPI2 +#ifdef WITHOPENMPI + std::string urifile = getenv("HOME"); + std::ostringstream mypid; + mypid << getpid(); + urifile += "/.urifile_" + mypid.str(); + setenv("OMPI_URI_FILE",urifile.c_str(),0); + if( getenv("OMPI_URI_FILE") != NULL ){ + system("killall ompi-server"); + std::string command; + command = "ompi-server -r "; + command += getenv("OMPI_URI_FILE"); + int status=system(command.c_str()); + if(status!=0) + throw SALOME_Exception("Error when launching ompi-server"); + } +#endif +#endif + MESSAGE("constructor end"); } @@ -94,11 +115,17 @@ SALOME_ContainerManager::SALOME_ContainerManager(CORBA::ORB_ptr orb, PortableSer SALOME_ContainerManager::~SALOME_ContainerManager() { MESSAGE("destructor"); +#ifdef HAVE_MPI2 +#ifdef WITHOPENMPI + if( getenv("OMPI_URI_FILE") != NULL ) + system("killall ompi-server"); +#endif +#endif } //============================================================================= +//! shutdown all the containers, then the ContainerManager servant /*! CORBA method: - * shutdown all the containers, then the ContainerManager servant */ //============================================================================= @@ -109,13 +136,11 @@ void SALOME_ContainerManager::Shutdown() _NS->Destroy_Name(_ContainerManagerNameInNS); PortableServer::ObjectId_var oid = _poa->servant_to_id(this); _poa->deactivate_object(oid); - //_remove_ref() has already been done at creation - //_remove_ref(); } //============================================================================= +//! Loop on all the containers listed in naming service, ask shutdown on each /*! CORBA Method: - * Loop on all the containers listed in naming service, ask shutdown on each */ //============================================================================= @@ -125,9 +150,9 @@ void SALOME_ContainerManager::ShutdownContainers() bool isOK; isOK = _NS->Change_Directory("/Containers"); if( isOK ){ - vector vec = _NS->list_directory_recurs(); - list lstCont; - for(vector::iterator iter = vec.begin();iter!=vec.end();iter++) + std::vector vec = _NS->list_directory_recurs(); + std::list lstCont; + for(std::vector::iterator iter = vec.begin();iter!=vec.end();iter++) { SCRUTE((*iter)); CORBA::Object_var obj=_NS->Resolve((*iter).c_str()); @@ -135,7 +160,7 @@ void SALOME_ContainerManager::ShutdownContainers() { Engines::Container_var cont=Engines::Container::_narrow(obj); if(!CORBA::is_nil(cont)) - lstCont.push_back((*iter)); + lstCont.push_back((*iter)); } catch(const CORBA::Exception& e) { @@ -143,632 +168,377 @@ void SALOME_ContainerManager::ShutdownContainers() } } MESSAGE("Container list: "); - for(list::iterator iter=lstCont.begin();iter!=lstCont.end();iter++){ + for(std::list::iterator iter=lstCont.begin();iter!=lstCont.end();iter++){ SCRUTE((*iter)); } - for(list::iterator iter=lstCont.begin();iter!=lstCont.end();iter++){ - SCRUTE((*iter)); - CORBA::Object_var obj=_NS->Resolve((*iter).c_str()); - Engines::Container_var cont=Engines::Container::_narrow(obj); - if(!CORBA::is_nil(cont)) + for(std::list::iterator iter=lstCont.begin();iter!=lstCont.end();iter++) + { + try + { + SCRUTE((*iter)); + CORBA::Object_var obj=_NS->Resolve((*iter).c_str()); + Engines::Container_var cont=Engines::Container::_narrow(obj); + if(!CORBA::is_nil(cont)) { - MESSAGE("ShutdownContainers: " << (*iter)); - try - { - cont->Shutdown(); - } - catch(CORBA::SystemException& e) - { - INFOS("CORBA::SystemException ignored : " << e); - } - catch(CORBA::Exception&) - { - INFOS("CORBA::Exception ignored."); - } - catch(...) - { - INFOS("Unknown exception ignored."); - } + MESSAGE("ShutdownContainers: " << (*iter)); + cont->Shutdown(); } - else - MESSAGE("ShutdownContainers: no container ref for " << (*iter)); + else + MESSAGE("ShutdownContainers: no container ref for " << (*iter)); + } + catch(CORBA::SystemException& e) + { + INFOS("CORBA::SystemException ignored : " << e); + } + catch(CORBA::Exception&) + { + INFOS("CORBA::Exception ignored."); + } + catch(...) + { + INFOS("Unknown exception ignored."); + } } } } //============================================================================= +//! Give a suitable Container given constraints /*! CORBA Method: - * Find a suitable Container in a list of machines, or start one - * \param params Machine Parameters required for the container - * \param possibleComputers list of machines usable for find or start + * \param params Container Parameters required for the container + * \return the container or nil */ //============================================================================= - Engines::Container_ptr -SALOME_ContainerManager:: -FindOrStartContainer(const Engines::MachineParameters& params, - const Engines::MachineList& possibleComputers) +SALOME_ContainerManager::GiveContainer(const Engines::ContainerParameters& params) { - Engines::Container_ptr ret = FindContainer(params,possibleComputers); - if(!CORBA::is_nil(ret)) - return ret; - MESSAGE("Container doesn't exist try to launch it ..."); - - return StartContainer(params,possibleComputers,Engines::P_FIRST); - -} - -//============================================================================= -/*! CORBA Method: - * Start a suitable Container in a list of machines - * \param params Machine Parameters required for the container - * \param possibleComputers list of machines usable for start - */ -//============================================================================= - -Engines::Container_ptr -SALOME_ContainerManager:: -StartContainer(const Engines::MachineParameters& params, - const Engines::MachineList& possibleComputers, - Engines::ResPolicy policy,const std::string& container_exe) -{ -#ifdef WITH_PACO_PARALLEL - std::string parallelLib(params.parallelLib); - if (parallelLib != "") - return FindOrStartParallelContainer(params, possibleComputers); -#endif - long id; - string containerNameInNS; - char idc[3*sizeof(long)]; + std::string machFile; Engines::Container_ptr ret = Engines::Container::_nil(); - MESSAGE("SALOME_ContainerManager::StartContainer " << - possibleComputers.length()); - - vector lm; - for(unsigned int i=0;iGetImpl()->FindFirst(lm); - break; - case Engines::P_CYCL: - theMachine=_ResManager->GetImpl()->FindNext(lm); - break; - case Engines::P_BEST: - theMachine=_ResManager->GetImpl()->FindBest(lm); - break; + // Step 0: Default mode is start + Engines::ContainerParameters local_params(params); + if (std::string(local_params.mode.in()) == "") + local_params.mode = CORBA::string_dup("start"); + std::string mode = local_params.mode.in(); + MESSAGE("[GiveContainer] starting with mode: " << mode); + + // Step 1: Find Container for find and findorstart mode + if (mode == "find" || mode == "findorstart") + { + ret = FindContainer(params, params.resource_params.resList); + if(!CORBA::is_nil(ret)) + return ret; + else + { + if (mode == "find") + { + MESSAGE("[GiveContainer] no container found"); + return ret; + } + else + { + mode = "start"; + } } } - catch( const SALOME_Exception &ex ){ - MESSAGE(ex.what()); - return Engines::Container::_nil(); - } - - //If the machine name is localhost use the real name - if(theMachine == "localhost") - theMachine=Kernel_Utils::GetHostname(); - MESSAGE("try to launch it on " << theMachine); + // Step 2: Get all possibleResources from the parameters + Engines::ResourceList_var possibleResources = _ResManager->GetFittingResources(local_params.resource_params); + MESSAGE("[GiveContainer] - length of possible resources " << possibleResources->length()); + std::vector local_resources; - // Get Id for container: a parallel container registers in Naming Service - // on the machine where is process 0. ContainerManager does'nt know the name - // of this machine before the launch of the parallel container. So to get - // the IOR of the parallel container in Naming Service, ContainerManager - // gives a unique Id. The parallel container registers his name under - // /ContainerManager/Id directory in NamingService - - id = GetIdForContainer(); + // Step 3: if mode is "get" keep only machines with existing containers + if(mode == "get") + { + for(unsigned int i=0; i < possibleResources->length(); i++) + { + Engines::Container_ptr cont = FindContainer(params, possibleResources[i].in()); + try + { + if(!cont->_non_existent()) + local_resources.push_back(std::string(possibleResources[i])); + } + catch(CORBA::Exception&) {} + } - string command; - if(theMachine==""){ - MESSAGE("SALOME_ContainerManager::StartContainer : " << - "no possible computer"); - return Engines::Container::_nil(); + // if local_resources is empty, we cannot give a container + if (local_resources.size() == 0) + { + MESSAGE("[GiveContainer] cannot find a container for mode get"); + return ret; + } } - else if(theMachine==Kernel_Utils::GetHostname()) - command = BuildCommandToLaunchLocalContainer(params,id,container_exe); else - command = BuildCommandToLaunchRemoteContainer(theMachine,params,id,container_exe); + for(unsigned int i=0; i < possibleResources->length(); i++) + local_resources.push_back(std::string(possibleResources[i])); - // RmTmpFile(); Too early! May be this function has not been used for a long time... - - //check if an entry exists in Naming service - if(params.isMPI) - { - containerNameInNS = "/ContainerManager/id"; - sprintf(idc,"%ld",id); - containerNameInNS += idc; - } + // Step 4: select the resource where to get/start the container + std::string resource_selected; + try + { + resource_selected = _ResManager->GetImpl()->Find(params.resource_params.policy.in(), local_resources); + } + catch(const SALOME_Exception &ex) + { + MESSAGE("[GiveContainer] Exception in ResourceManager find !: " << ex.what()); + return ret; + } + MESSAGE("[GiveContainer] Resource selected is: " << resource_selected); + + // Step 5: get container in the naming service + Engines::ResourceDefinition_var resource_definition = _ResManager->GetResourceDefinition(resource_selected.c_str()); + std::string hostname(resource_definition->hostname.in()); + std::string containerNameInNS; + if(params.isMPI){ + int nbproc; + if ( params.nb_proc <= 0 ) + nbproc = 1; + else + nbproc = params.nb_proc; + if( getenv("LIBBATCH_NODEFILE") != NULL ) + machFile = machinesFile(nbproc); + // A mpi parallel container register on zero node in NS + containerNameInNS = _NS->BuildContainerNameForNS(params, GetMPIZeroNode(hostname,machFile).c_str()); + } else - containerNameInNS = _NS->BuildContainerNameForNS(params,theMachine.c_str()); + containerNameInNS = _NS->BuildContainerNameForNS(params, hostname.c_str()); + MESSAGE("[GiveContainer] Container name in the naming service: " << containerNameInNS); - SCRUTE(containerNameInNS); + // Step 6: check if the name exists in naming service + //if params.mode == "getorstart" or "get" use the existing container + //if params.mode == "start" shutdown the existing container before launching a new one with that name CORBA::Object_var obj = _NS->Resolve(containerNameInNS.c_str()); - if ( !CORBA::is_nil(obj) ) + if (!CORBA::is_nil(obj)) + { + try { - try + Engines::Container_var cont=Engines::Container::_narrow(obj); + if(!cont->_non_existent()) + { + if(std::string(params.mode.in())=="getorstart" || std::string(params.mode.in())=="get"){ + return cont._retn(); /* the container exists and params.mode is getorstart or get use it*/ + } + else { - // shutdown the registered container if it exists - Engines::Container_var cont=Engines::Container::_narrow(obj); - if(!CORBA::is_nil(cont)) - cont->Shutdown(); + INFOS("[GiveContainer] A container is already registered with the name: " << containerNameInNS << ", shutdown the existing container"); + cont->Shutdown(); // shutdown the registered container if it exists } - catch(CORBA::Exception&) + } + } + catch(CORBA::Exception&) + { + INFOS("[GiveContainer] CORBA::Exception ignored when trying to get the container - we start a new one"); + } + } + + // Step 7: type of container: PaCO, Exe, Mpi or Classic + // Mpi already tested in step 5, specific code on BuildCommandToLaunch Local/Remote Container methods + // TODO -> separates Mpi from Classic/Exe + // Classic or Exe ? + std::string container_exe = "SALOME_Container"; // Classic container + int found=0; + try + { + CORBA::String_var container_exe_tmp; + CORBA::Object_var obj = _NS->Resolve("/Kernel/ModulCatalog"); + SALOME_ModuleCatalog::ModuleCatalog_var Catalog = SALOME_ModuleCatalog::ModuleCatalog::_narrow(obj) ; + if (CORBA::is_nil (Catalog)) + { + INFOS("[GiveContainer] Module Catalog is not found -> cannot launch a container"); + return ret; + } + // Loop through component list + for(unsigned int i=0; i < local_params.resource_params.componentList.length(); i++) + { + const char* compoi = local_params.resource_params.componentList[i]; + SALOME_ModuleCatalog::Acomponent_var compoInfo = Catalog->GetComponent(compoi); + if (CORBA::is_nil (compoInfo)) + { + continue; + } + SALOME_ModuleCatalog::ImplType impl=compoInfo->implementation_type(); + container_exe_tmp=compoInfo->implementation_name(); + if(impl==SALOME_ModuleCatalog::CEXE) + { + if(found) { - INFOS("CORBA::Exception ignored."); + INFOS("ContainerManager Error: you can't have 2 CEXE component in the same container" ); + return Engines::Container::_nil(); } + MESSAGE("[GiveContainer] Exe container found !: " << container_exe_tmp); + container_exe = container_exe_tmp.in(); + found=1; + } } + } + catch (ServiceUnreachable&) + { + INFOS("Caught exception: Naming Service Unreachable"); + return ret; + } + catch (...) + { + INFOS("Caught unknown exception."); + return ret; + } + + // Step 8: start a new container + // Check if a PaCO container + // PaCO++ + if (std::string(local_params.parallelLib.in()) != "") + { + ret = StartPaCOPPContainer(params, resource_selected); + return ret; + } + // Other type of containers... + MESSAGE("[GiveContainer] Try to launch a new container on " << resource_selected); + std::string command; + // if a parallel container is launched in batch job, command is: "mpirun -np nbproc -machinefile nodesfile SALOME_MPIContainer" + if( getenv("LIBBATCH_NODEFILE") != NULL && params.isMPI ) + command = BuildCommandToLaunchLocalContainer(params, machFile, container_exe); + // if a container is launched on localhost, command is "SALOME_Container" or "mpirun -np nbproc SALOME_MPIContainer" + else if(hostname == Kernel_Utils::GetHostname()) + command = BuildCommandToLaunchLocalContainer(params, machFile, container_exe); + // if a container is launched in remote mode, command is "ssh resource_selected SALOME_Container" or "ssh resource_selected mpirun -np nbproc SALOME_MPIContainer" + else + command = BuildCommandToLaunchRemoteContainer(resource_selected, params, container_exe); //redirect stdout and stderr in a file - string logFilename="/tmp/"+_NS->ContainerName(params)+"_"+ theMachine +"_"+getenv( "USER" )+".log" ; - command += " > " + logFilename + " 2>&1 &"; +#ifdef WNT + std::string logFilename=getenv("TEMP"); + logFilename += "\\"; + std::string user = getenv( "USERNAME" ); +#else + std::string user = getenv( "USER" ); + std::string logFilename="/tmp"; + char* val = getenv("SALOME_TMP_DIR"); + if(val) + { + struct stat file_info; + stat(val, &file_info); + bool is_dir = S_ISDIR(file_info.st_mode); + if (is_dir)logFilename=val; + else std::cerr << "SALOME_TMP_DIR environment variable is not a directory use /tmp instead" << std::endl; + } + logFilename += "/"; +#endif + logFilename += _NS->ContainerName(params)+"_"+ resource_selected +"_"+user; + std::ostringstream tmp; + tmp << "_" << getpid(); + logFilename += tmp.str(); + logFilename += ".log" ; + command += " > " + logFilename + " 2>&1"; +#ifdef WNT + command = "%PYTHONBIN% -c \"import win32pm ; win32pm.spawnpid(r'" + command + "', '')\""; +#else + command += " &"; +#endif // launch container with a system call int status=system(command.c_str()); - if (status == -1){ - MESSAGE("SALOME_LifeCycleCORBA::StartOrFindContainer rsh failed " << - "(system command status -1)"); - RmTmpFile(); // command file can be removed here + MESSAGE("SALOME_ContainerManager::StartContainer rsh failed (system command status -1)"); + RmTmpFile(_TmpFileName); // command file can be removed here return Engines::Container::_nil(); } else if (status == 217){ - MESSAGE("SALOME_LifeCycleCORBA::StartOrFindContainer rsh failed " << - "(system command status 217)"); - RmTmpFile(); // command file can be removed here + MESSAGE("SALOME_ContainerManager::StartContainer rsh failed (system command status 217)"); + RmTmpFile(_TmpFileName); // command file can be removed here return Engines::Container::_nil(); } - else{ - int count=TIME_OUT_TO_LAUNCH_CONT; - MESSAGE("count = "<Resolve(containerNameInNS.c_str()); ret=Engines::Container::_narrow(obj); } - - if ( CORBA::is_nil(ret) ) - { - MESSAGE("SALOME_LifeCycleCORBA::StartOrFindContainer rsh failed"); - } - else - { - logFilename=":"+logFilename; - logFilename="@"+Kernel_Utils::GetHostname()+logFilename; - logFilename=getenv( "USER" )+logFilename; - ret->logfilename(logFilename.c_str()); - } - - RmTmpFile(); // command file can be removed here - return ret; - } -} - -//============================================================================= -/*! CORBA Method: - * Start a suitable Container in a list of machines - * \param params Machine Parameters required for the container - * \param possibleComputers list of machines usable for start - */ -//============================================================================= - -Engines::Container_ptr -SALOME_ContainerManager:: -StartContainer(const Engines::MachineParameters& params, - Engines::ResPolicy policy, - const Engines::CompoList& componentList) -{ - Engines::MachineList_var possibleComputers = _ResManager->GetFittingResources(params,componentList); - - // Look into ModulCatalog if a specific container must be launched - CORBA::String_var container_exe; - int found=0; - try + if (CORBA::is_nil(ret)) { - CORBA::Object_var obj = _NS->Resolve("/Kernel/ModulCatalog"); - SALOME_ModuleCatalog::ModuleCatalog_var Catalog = SALOME_ModuleCatalog::ModuleCatalog::_narrow(obj) ; - if (CORBA::is_nil (Catalog)) - return Engines::Container::_nil(); - // Loop through component list - for(unsigned int i=0;iGetComponent(compoi); - if (CORBA::is_nil (compoInfo)) - { - continue; - } - SALOME_ModuleCatalog::ImplType impl=compoInfo->implementation_type(); - container_exe=compoInfo->implementation_name(); - if(impl==SALOME_ModuleCatalog::CEXE) - { - if(found) - { - INFOS("ContainerManager Error: you can't have 2 CEXE component in the same container" ); - return Engines::Container::_nil(); - } - found=1; - } - } - } - catch (ServiceUnreachable&) - { - INFOS("Caught exception: Naming Service Unreachable"); - return Engines::Container::_nil(); + INFOS("[GiveContainer] was not able to launch container " << containerNameInNS); } - catch (...) + else { - INFOS("Caught unknown exception."); - return Engines::Container::_nil(); - } - - if(found) - return StartContainer(params,possibleComputers,policy,container_exe.in()); - else - return StartContainer(params,possibleComputers,policy); -} - -#ifdef WITH_PACO_PARALLEL -//============================================================================= -/*! CORBA Method: - * Find or Start a suitable PaCO++ Parallel Container in a list of machines. - * \param params Machine Parameters required for the container - * \param possibleComputers list of machines usable for find or start - * - * \return CORBA container reference. - */ -//============================================================================= -Engines::Container_ptr -SALOME_ContainerManager:: -FindOrStartParallelContainer(const Engines::MachineParameters& params_const, - const Engines::MachineList& possibleComputers) -{ - CORBA::Object_var obj; - PaCO::InterfaceManager_var proxy; - Engines::Container_ptr ret = Engines::Container::_nil(); - Engines::MachineParameters params(params_const); - - // Step 1 : Try to find a suitable container - // Currently not as good as could be since - // we have to verified the number of nodes of the container - // if a user tell that. - ret = FindContainer(params, possibleComputers); - - if(CORBA::is_nil(ret)) { - // Step 2 : Starting a new parallel container - INFOS("[FindOrStartParallelContainer] Starting a parallel container"); - - // Step 2.1 : Choose a computer - string theMachine = _ResManager->FindFirst(possibleComputers); - if(theMachine == "") { - INFOS("[FindOrStartParallelContainer] !!!!!!!!!!!!!!!!!!!!!!!!!!"); - INFOS("[FindOrStartParallelContainer] No possible computer found"); - INFOS("[FindOrStartParallelContainer] !!!!!!!!!!!!!!!!!!!!!!!!!!"); - } - else { - INFOS("[FindOrStartParallelContainer] on machine : " << theMachine); - string command; - if(theMachine == Kernel_Utils::GetHostname()) { - // Step 3 : starting parallel container proxy - params.hostname = CORBA::string_dup(theMachine.c_str()); - Engines::MachineParameters params_proxy(params); - try { - command = BuildCommandToLaunchLocalParallelContainer("SALOME_ParallelContainerProxy", params_proxy, "xterm"); - } - catch(const SALOME_Exception & ex){ - MESSAGE(ex.what()); - return Engines::Container::_nil(); - } - // LaunchParallelContainer uses this value to know if it launches the proxy or the nodes - params_proxy.nb_component_nodes = 0; - obj = LaunchParallelContainer(command, params_proxy, _NS->ContainerName(params)); - ret = Engines::Container::_narrow(obj); - proxy = PaCO::InterfaceManager::_narrow(obj); - - // Step 4 : starting parallel container nodes - command = BuildCommandToLaunchLocalParallelContainer("SALOME_ParallelContainerNode", params, "xterm"); - string name = _NS->ContainerName(params) + "Node"; - LaunchParallelContainer(command, params, name); - // Step 5 : connecting nodes and the proxy to actually create a parallel container - try { - for (int i = 0; i < params.nb_component_nodes; i++) { - - char buffer [5]; -#ifndef WIN32 - snprintf(buffer,5,"%d",i); -#else - _snprintf(buffer,5,"%d",i); -#endif - string name_cont = name + string(buffer); - - string theNodeMachine(CORBA::string_dup(params.hostname)); - string containerNameInNS = _NS->BuildContainerNameForNS(name_cont.c_str(),theNodeMachine.c_str()); - int count = TIME_OUT_TO_LAUNCH_CONT; - obj = _NS->Resolve(containerNameInNS.c_str()); - while (CORBA::is_nil(obj) && count) { - INFOS("[FindOrStartParallelContainer] CONNECTION FAILED !!!!!!!!!!!!!!!!!!!!!!!!"); -#ifndef WIN32 - sleep(1) ; -#else - Sleep(1000); -#endif - count-- ; - obj = _NS->Resolve(containerNameInNS.c_str()); - } - - PaCO::InterfaceParallel_var node = PaCO::InterfaceParallel::_narrow(obj); - MESSAGE("[FindOrStartParallelContainer] Deploying node : " << name); - node->deploy(); - } - proxy->start(); - } - catch(CORBA::SystemException& e) - { - INFOS("Caught CORBA::SystemException. : " << e); - } - catch(PortableServer::POA::ServantAlreadyActive&) - { - INFOS("Caught CORBA::ServantAlreadyActiveException"); - } - catch(CORBA::Exception&) - { - INFOS("Caught CORBA::Exception."); - } - catch(std::exception& exc) - { - INFOS("Caught std::exception - "<logfilename(logFilename.c_str()); + RmTmpFile(_TmpFileName); // command file can be removed here } -} -return ret; -} -#else -//============================================================================= -/*! CORBA Method: - * Find or Start a suitable PaCO++ Parallel Container in a list of machines. - * \param params Machine Parameters required for the container - * \param possibleComputers list of machines usable for find or start - * - * \return CORBA container reference. - */ -//============================================================================= -Engines::Container_ptr -SALOME_ContainerManager:: -FindOrStartParallelContainer(const Engines::MachineParameters& params, - const Engines::MachineList& possibleComputers) -{ - Engines::Container_ptr ret = Engines::Container::_nil(); - INFOS("[FindOrStartParallelContainer] is disabled !"); - INFOS("[FindOrStartParallelContainer] recompile SALOME Kernel to enable parallel extension"); + } return ret; } -#endif //============================================================================= -/*! CORBA Method: - * Give a suitable Container in a list of machines - * \param params Machine Parameters required for the container - * \param possibleComputers list of machines usable for start - */ -//============================================================================= - -Engines::Container_ptr -SALOME_ContainerManager:: -GiveContainer(const Engines::MachineParameters& params, - Engines::ResPolicy policy, - const Engines::CompoList& componentList) -{ - char *valenv=getenv("SALOME_BATCH"); - if(valenv) - if (strcmp(valenv,"1")==0) - { - if(_batchLaunchedContainers.empty()) - fillBatchLaunchedContainers(); - - if (_batchLaunchedContainersIter == _batchLaunchedContainers.end()) - _batchLaunchedContainersIter = _batchLaunchedContainers.begin(); - - Engines::Container_ptr rtn = Engines::Container::_duplicate(*_batchLaunchedContainersIter); - _batchLaunchedContainersIter++; - return rtn; - } - return StartContainer(params,policy,componentList); -} - -//============================================================================= -/*! - * - */ -//============================================================================= - -Engines::Container_ptr -SALOME_ContainerManager:: -FindContainer(const Engines::MachineParameters& params, - const char *theMachine) -{ - string containerNameInNS(_NS->BuildContainerNameForNS(params,theMachine)); - CORBA::Object_var obj = _NS->Resolve(containerNameInNS.c_str()); - if( !CORBA::is_nil(obj) ) - return Engines::Container::_narrow(obj); - else - return Engines::Container::_nil(); -} - -//============================================================================= -/*! - * +//! Find a container given constraints (params) on a list of machines (possibleComputers) +/*! + * */ //============================================================================= Engines::Container_ptr -SALOME_ContainerManager:: -FindContainer(const Engines::MachineParameters& params, - const Engines::MachineList& possibleComputers) +SALOME_ContainerManager::FindContainer(const Engines::ContainerParameters& params, + const Engines::ResourceList& possibleResources) { - MESSAGE("FindContainer "<BuildContainerNameForNS((char*) name.c_str(),theMachine.c_str()); - - INFOS("[LaunchParallelContainer] Waiting for Parallel Container proxy on " << theMachine); - while (CORBA::is_nil(obj) && count) { -#ifndef WIN32 - sleep(1) ; -#else - Sleep(1000); -#endif - count-- ; - obj = _NS->Resolve(containerNameInNS.c_str()); - } - } - else { - INFOS("[LaunchParallelContainer] launching the nodes of the parallel container"); - int status = system(command.c_str()); - if (status == -1) { - INFOS("[LaunchParallelContainer] failed : system command status -1"); - } - else if (status == 217) { - INFOS("[LaunchParallelContainer] failed : system command status 217"); - } - // We are waiting all the nodes - for (int i = 0; i < params.nb_component_nodes; i++) { - obj = CORBA::Object::_nil(); - int count = TIME_OUT_TO_LAUNCH_CONT; - - // Name of the node - char buffer [5]; -#ifndef WIN32 - snprintf(buffer,5,"%d",i); -#else - _snprintf(buffer,5,"%d",i); -#endif - - string name_cont = name + string(buffer); - - // I don't like this... - string theMachine(CORBA::string_dup(params.hostname)); - containerNameInNS = _NS->BuildContainerNameForNS((char*) name_cont.c_str(),theMachine.c_str()); - cerr << "[LaunchContainer] Waiting for Parllel Container node " << containerNameInNS << " on " << theMachine << endl; - while (CORBA::is_nil(obj) && count) { -#ifndef WIN32 - sleep(1) ; -#else - Sleep(1000); -#endif - count-- ; - obj = _NS->Resolve(containerNameInNS.c_str()); - } - } +Engines::Container_ptr +SALOME_ContainerManager::FindContainer(const Engines::ContainerParameters& params, + const std::string& resource) +{ + Engines::ResourceDefinition_var resource_definition = _ResManager->GetResourceDefinition(resource.c_str()); + std::string hostname(resource_definition->hostname.in()); + std::string containerNameInNS(_NS->BuildContainerNameForNS(params, hostname.c_str())); + MESSAGE("[FindContainer] Try to find a container " << containerNameInNS << " on resource " << resource); + CORBA::Object_var obj = _NS->Resolve(containerNameInNS.c_str()); + try + { + if(obj->_non_existent()) + return Engines::Container::_nil(); + else + return Engines::Container::_narrow(obj); } - - if ( CORBA::is_nil(obj) ) { - INFOS("[LaunchParallelContainer] failed"); + catch(const CORBA::Exception& e) + { + return Engines::Container::_nil(); } - return obj; } -//============================================================================= -/*! - * Get Id for container: a parallel container registers in Naming Service - * on the machine where is process 0. ContainerManager does'nt know the name - * of this machine before the launch of the parallel container. So to get - * the IOR of the parallel container in Naming Service, ContainerManager - * gives a unique Id. The parallel container registers his name under - * /ContainerManager/Id directory in NamingService - */ -//============================================================================= - - -long SALOME_ContainerManager::GetIdForContainer(void) -{ - _id++; - return _id; -} -void SALOME_ContainerManager::fillBatchLaunchedContainers() -{ - _batchLaunchedContainers.clear(); - _NS->Change_Directory("/Containers"); - vector vec = _NS->list_directory_recurs(); - for(vector::iterator iter = vec.begin();iter!=vec.end();iter++){ - CORBA::Object_var obj=_NS->Resolve((*iter).c_str()); - Engines::Container_ptr cont=Engines::Container::_narrow(obj); - if(!CORBA::is_nil(cont)){ - _batchLaunchedContainers.push_back(cont); - } - } - _batchLaunchedContainersIter=_batchLaunchedContainers.begin(); -} +bool isPythonContainer(const char* ContainerName); //============================================================================= /*! * This is no longer valid (C++ container are also python containers) */ //============================================================================= - bool isPythonContainer(const char* ContainerName) { bool ret = false; @@ -804,104 +574,104 @@ bool isPythonContainer(const char* ContainerName) */ //============================================================================= -string +std::string SALOME_ContainerManager::BuildCommandToLaunchRemoteContainer -(const string& machine, - const Engines::MachineParameters& params, const long id,const std::string& container_exe) +(const std::string& resource_name, + const Engines::ContainerParameters& params, const std::string& container_exe) { - string command; - int nbproc; - char idc[3*sizeof(long)]; - - if ( ! _isAppliSalomeDefined ) - command = BuildTempFileToLaunchRemoteContainer(machine, params); - + + std::string command; + if (!_isAppliSalomeDefined) + command = BuildTempFileToLaunchRemoteContainer(resource_name, params); else - { - const ParserResourcesType& resInfo = _ResManager->GetImpl()->GetResourcesList(machine); - - if (params.isMPI) - { - if ( (params.nb_node <= 0) && (params.nb_proc_per_node <= 0) ) - nbproc = 1; - else if ( params.nb_node == 0 ) - nbproc = params.nb_proc_per_node; - else if ( params.nb_proc_per_node == 0 ) - nbproc = params.nb_node; - else - nbproc = params.nb_node * params.nb_proc_per_node; - } - - // "ssh user@machine distantPath/runRemote.sh hostNS portNS WORKINGDIR workingdir \ - // SALOME_Container containerName &" + { + int nbproc; + Engines::ResourceDefinition_var resource_definition = _ResManager->GetResourceDefinition(resource_name.c_str()); + std::string hostname(resource_definition->hostname.in()); + const ParserResourcesType& resInfo = _ResManager->GetImpl()->GetResourcesDescr(resource_name); - if (resInfo.Protocol == rsh) - command = "rsh "; - else if (resInfo.Protocol == ssh) - command = "ssh "; + if (params.isMPI) + { + if ( params.nb_proc <= 0 ) + nbproc = 1; else - throw SALOME_Exception("Unknown protocol"); + nbproc = params.nb_proc; + } - if (resInfo.UserName != "") - { - command += resInfo.UserName; - command += "@"; - } + // "ssh -l user machine distantPath/runRemote.sh hostNS portNS WORKINGDIR workingdir \ + // SALOME_Container containerName &" + if (resInfo.Protocol == rsh) + command = "rsh "; + else if (resInfo.Protocol == ssh) + command = "ssh "; + else + throw SALOME_Exception("Unknown protocol"); - command += machine; + if (resInfo.UserName != "") + { + command += "-l "; + command += resInfo.UserName; command += " "; + } - if (resInfo.AppliPath != "") - command += resInfo.AppliPath; // path relative to user@machine $HOME - else - { - ASSERT(getenv("APPLI")); - command += getenv("APPLI"); // path relative to user@machine $HOME - } + command += resInfo.HostName; + command += " "; - command += "/runRemote.sh "; + if (resInfo.AppliPath != "") + command += resInfo.AppliPath; // path relative to user@machine $HOME + else + { + ASSERT(getenv("APPLI")); + command += getenv("APPLI"); // path relative to user@machine $HOME + } - ASSERT(getenv("NSHOST")); - command += getenv("NSHOST"); // hostname of CORBA name server + command += "/runRemote.sh "; - command += " "; - ASSERT(getenv("NSPORT")); - command += getenv("NSPORT"); // port of CORBA name server + ASSERT(getenv("NSHOST")); + command += getenv("NSHOST"); // hostname of CORBA name server - std::string wdir=params.workingdir.in(); - if(wdir != "") - { - command += " WORKINGDIR "; - command += " '"; - if(wdir == "$TEMPDIR") - wdir="\\$TEMPDIR"; - command += wdir; // requested working directory - command += "'"; - } + command += " "; + ASSERT(getenv("NSPORT")); + command += getenv("NSPORT"); // port of CORBA name server + + std::string wdir = params.workingdir.in(); + if(wdir != "") + { + command += " WORKINGDIR "; + command += " '"; + if(wdir == "$TEMPDIR") + wdir="\\$TEMPDIR"; + command += wdir; // requested working directory + command += "'"; + } - if(params.isMPI) - { - command += " mpirun -np "; - std::ostringstream o; - o << nbproc << " "; - command += o.str(); + if(params.isMPI) + { + command += " mpirun -np "; + std::ostringstream o; + o << nbproc << " "; + command += o.str(); #ifdef WITHLAM - command += "-x PATH,LD_LIBRARY_PATH,OMNIORB_CONFIG,SALOME_trace "; -#endif - command += " SALOME_MPIContainer "; - } - else - command += " " +container_exe+ " "; + command += "-x PATH,LD_LIBRARY_PATH,OMNIORB_CONFIG,SALOME_trace "; +#elif defined(WITHOPENMPI) + if( getenv("OMPI_URI_FILE") == NULL ) + command += "-x PATH -x LD_LIBRARY_PATH -x OMNIORB_CONFIG -x SALOME_trace"; + else{ + command += "-x PATH -x LD_LIBRARY_PATH -x OMNIORB_CONFIG -x SALOME_trace -ompi-server file:"; + command += getenv("OMPI_URI_FILE"); + } +#endif + command += " SALOME_MPIContainer "; + } + else + command += " " +container_exe+ " "; - command += _NS->ContainerName(params); - command += " -id "; - sprintf(idc,"%ld",id); - command += idc; - command += " -"; - AddOmninamesParams(command); + command += _NS->ContainerName(params); + command += " -"; + AddOmninamesParams(command); - MESSAGE("command =" << command); - } + MESSAGE("command =" << command); + } return command; } @@ -911,54 +681,50 @@ SALOME_ContainerManager::BuildCommandToLaunchRemoteContainer * builds the command to be launched. */ //============================================================================= - -string +std::string SALOME_ContainerManager::BuildCommandToLaunchLocalContainer -(const Engines::MachineParameters& params, const long id,const std::string& container_exe) +(const Engines::ContainerParameters& params, const std::string& machinesFile, const std::string& container_exe) { _TmpFileName = BuildTemporaryFileName(); - string command; + std::string command; int nbproc = 0; - //char idc[3*sizeof(long)]; - ofstream command_file( _TmpFileName.c_str() ); + std::ostringstream o; if (params.isMPI) { - //command = "mpirun -np "; - command_file << "mpirun -np "; + o << "mpirun -np "; - if ( (params.nb_node <= 0) && (params.nb_proc_per_node <= 0) ) + if ( params.nb_proc <= 0 ) nbproc = 1; - else if ( params.nb_node == 0 ) - nbproc = params.nb_proc_per_node; - else if ( params.nb_proc_per_node == 0 ) - nbproc = params.nb_node; else - nbproc = params.nb_node * params.nb_proc_per_node; + nbproc = params.nb_proc; - //std::ostringstream o; + o << nbproc << " "; - //o << nbproc << " "; - command_file << nbproc << " "; + if( getenv("LIBBATCH_NODEFILE") != NULL ) + o << "-machinefile " << machinesFile << " "; - //command += o.str(); #ifdef WITHLAM - //command += "-x PATH,LD_LIBRARY_PATH,OMNIORB_CONFIG,SALOME_trace "; - command_file << "-x PATH,LD_LIBRARY_PATH,OMNIORB_CONFIG,SALOME_trace "; + o << "-x PATH,LD_LIBRARY_PATH,OMNIORB_CONFIG,SALOME_trace "; +#elif defined(WITHOPENMPI) + if( getenv("OMPI_URI_FILE") == NULL ) + o << "-x PATH -x LD_LIBRARY_PATH -x OMNIORB_CONFIG -x SALOME_trace"; + else + { + o << "-x PATH -x LD_LIBRARY_PATH -x OMNIORB_CONFIG -x SALOME_trace -ompi-server file:"; + o << getenv("OMPI_URI_FILE"); + } #endif if (isPythonContainer(params.container_name)) - //command += "pyMPI SALOME_ContainerPy.py "; - command_file << "pyMPI SALOME_ContainerPy.py "; + o << " pyMPI SALOME_ContainerPy.py "; else - //command += "SALOME_MPIContainer "; - command_file << "SALOME_MPIContainer "; + o << " SALOME_MPIContainer "; } else { - //command=""; std::string wdir=params.workingdir.in(); if(wdir != "") { @@ -966,13 +732,11 @@ SALOME_ContainerManager::BuildCommandToLaunchLocalContainer if(wdir == "$TEMPDIR") { // a new temporary directory is requested - string dir = Kernel_Utils::GetTmpDir(); + std::string dir = Kernel_Utils::GetTmpDir(); #ifdef WIN32 - //command += "cd /d "+ dir +";"; - command_file << "cd /d " << dir << endl; + o << "cd /d " << dir << std::endl; #else - //command = "cd "+ dir +";"; - command_file << "cd " << dir << ";"; + o << "cd " << dir << ";"; #endif } @@ -980,35 +744,26 @@ SALOME_ContainerManager::BuildCommandToLaunchLocalContainer { // a permanent directory is requested use it or create it #ifdef WIN32 - //command="mkdir " + wdir; - command_file << "mkdir " + wdir << endl; - command_file << "cd /D " + wdir << endl; + o << "mkdir " + wdir << std::endl; + o << "cd /D " + wdir << std::endl; #else - //command="mkdir -p " + wdir + " && cd " + wdir + ";"; - command_file << "mkdir -p " << wdir << " && cd " << wdir + ";"; + o << "mkdir -p " << wdir << " && cd " << wdir + ";"; #endif } } if (isPythonContainer(params.container_name)) - //command += "SALOME_ContainerPy.py "; - command_file << "SALOME_ContainerPy.py "; + o << "SALOME_ContainerPy.py "; else - //command += container_exe + " "; - command_file << container_exe + " "; + o << container_exe + " "; } + o << _NS->ContainerName(params); + o << " -"; + AddOmninamesParams(o); - /*command += _NS->ContainerName(params); - command += " -id "; - sprintf(idc,"%ld",id); - command += idc; - command += " -"; - AddOmninamesParams(command);*/ - - command_file << _NS->ContainerName(params); - command_file << " -id " << id << " -"; - AddOmninamesParams(command_file); + std::ofstream command_file( _TmpFileName.c_str() ); + command_file << o.str(); command_file.close(); #ifndef WIN32 @@ -1017,6 +772,7 @@ SALOME_ContainerManager::BuildCommandToLaunchLocalContainer command = _TmpFileName; MESSAGE("Command is file ... " << command); + MESSAGE("Command is ... " << o.str()); return command; } @@ -1027,24 +783,24 @@ SALOME_ContainerManager::BuildCommandToLaunchLocalContainer */ //============================================================================= -void SALOME_ContainerManager::RmTmpFile() +void SALOME_ContainerManager::RmTmpFile(std::string& tmpFileName) { - int lenght = _TmpFileName.size(); + int lenght = tmpFileName.size(); if ( lenght > 0) { #ifdef WIN32 - string command = "del /F "; + std::string command = "del /F "; #else - string command = "rm "; + std::string command = "rm "; #endif if ( lenght > 4 ) - command += _TmpFileName.substr(0, lenght - 3 ); + command += tmpFileName.substr(0, lenght - 3 ); else - command += _TmpFileName; + command += tmpFileName; command += '*'; system(command.c_str()); //if dir is empty - remove it - string tmp_dir = Kernel_Utils::GetDirByPath( _TmpFileName ); + std::string tmp_dir = Kernel_Utils::GetDirByPath( tmpFileName ); if ( Kernel_Utils::IsEmptyDir( tmp_dir ) ) { #ifdef WIN32 @@ -1063,13 +819,25 @@ void SALOME_ContainerManager::RmTmpFile() */ //============================================================================= -void SALOME_ContainerManager::AddOmninamesParams(string& command) const - { - CORBA::String_var iorstr = _NS->getIORaddr(); - command += "ORBInitRef NameService="; - command += iorstr; - } +void SALOME_ContainerManager::AddOmninamesParams(std::string& command) const +{ + CORBA::String_var iorstr = _NS->getIORaddr(); + command += "ORBInitRef NameService="; + command += iorstr; +} +//============================================================================= +/*! + * add to command all options relative to naming service. + */ +//============================================================================= + +void SALOME_ContainerManager::AddOmninamesParams(std::ofstream& fileStream) const +{ + CORBA::String_var iorstr = _NS->getIORaddr(); + fileStream << "ORBInitRef NameService="; + fileStream << iorstr; +} //============================================================================= /*! @@ -1077,12 +845,12 @@ void SALOME_ContainerManager::AddOmninamesParams(string& command) const */ //============================================================================= -void SALOME_ContainerManager::AddOmninamesParams(ofstream& fileStream) const - { - CORBA::String_var iorstr = _NS->getIORaddr(); - fileStream << "ORBInitRef NameService="; - fileStream << iorstr; - } +void SALOME_ContainerManager::AddOmninamesParams(std::ostringstream& oss) const +{ + CORBA::String_var iorstr = _NS->getIORaddr(); + oss << "ORBInitRef NameService="; + oss << iorstr; +} //============================================================================= /*! @@ -1090,18 +858,17 @@ void SALOME_ContainerManager::AddOmninamesParams(ofstream& fileStream) const */ //============================================================================= -string SALOME_ContainerManager::BuildTemporaryFileName() const - { - //build more complex file name to support multiple salome session - string aFileName = Kernel_Utils::GetTmpFileName(); +std::string SALOME_ContainerManager::BuildTemporaryFileName() const +{ + //build more complex file name to support multiple salome session + std::string aFileName = Kernel_Utils::GetTmpFileName(); #ifndef WIN32 - aFileName += ".sh"; + aFileName += ".sh"; #else - aFileName += ".bat"; + aFileName += ".bat"; #endif - return aFileName; - } - + return aFileName; +} //============================================================================= /*! @@ -1113,22 +880,22 @@ string SALOME_ContainerManager::BuildTemporaryFileName() const */ //============================================================================= -string +std::string SALOME_ContainerManager::BuildTempFileToLaunchRemoteContainer -(const string& machine, - const Engines::MachineParameters& params) throw(SALOME_Exception) +(const std::string& resource_name, + const Engines::ContainerParameters& params) throw(SALOME_Exception) { int status; _TmpFileName = BuildTemporaryFileName(); - ofstream tempOutputFile; - tempOutputFile.open(_TmpFileName.c_str(), ofstream::out ); - const ParserResourcesType& resInfo = _ResManager->GetImpl()->GetResourcesList(machine); - tempOutputFile << "#! /bin/sh" << endl; + std::ofstream tempOutputFile; + tempOutputFile.open(_TmpFileName.c_str(), std::ofstream::out ); + const ParserResourcesType& resInfo = _ResManager->GetImpl()->GetResourcesDescr(resource_name); + tempOutputFile << "#! /bin/sh" << std::endl; // --- set env vars - tempOutputFile << "export SALOME_trace=local" << endl; // mkr : 27.11.2006 : PAL13967 - Distributed supervision graphs - Problem with "SALOME_trace" + tempOutputFile << "export SALOME_trace=local" << std::endl; // mkr : 27.11.2006 : PAL13967 - Distributed supervision graphs - Problem with "SALOME_trace" //tempOutputFile << "source " << resInfo.PreReqFilePath << endl; // ! env vars @@ -1138,20 +905,23 @@ SALOME_ContainerManager::BuildTempFileToLaunchRemoteContainer tempOutputFile << "mpirun -np "; int nbproc; - if ( (params.nb_node <= 0) && (params.nb_proc_per_node <= 0) ) + if ( params.nb_proc <= 0 ) nbproc = 1; - else if ( params.nb_node == 0 ) - nbproc = params.nb_proc_per_node; - else if ( params.nb_proc_per_node == 0 ) - nbproc = params.nb_node; else - nbproc = params.nb_node * params.nb_proc_per_node; + nbproc = params.nb_proc; std::ostringstream o; tempOutputFile << nbproc << " "; #ifdef WITHLAM tempOutputFile << "-x PATH,LD_LIBRARY_PATH,OMNIORB_CONFIG,SALOME_trace "; +#elif defined(WITHOPENMPI) + if( getenv("OMPI_URI_FILE") == NULL ) + tempOutputFile << "-x PATH -x LD_LIBRARY_PATH -x OMNIORB_CONFIG -x SALOME_trace"; + else{ + tempOutputFile << "-x PATH -x LD_LIBRARY_PATH -x OMNIORB_CONFIG -x SALOME_trace -ompi-server file:"; + tempOutputFile << getenv("OMPI_URI_FILE"); + } #endif } @@ -1160,9 +930,9 @@ SALOME_ContainerManager::BuildTempFileToLaunchRemoteContainer if (params.isMPI) { if (isPythonContainer(params.container_name)) - tempOutputFile << "pyMPI SALOME_ContainerPy.py "; + tempOutputFile << " pyMPI SALOME_ContainerPy.py "; else - tempOutputFile << "SALOME_MPIContainer "; + tempOutputFile << " SALOME_MPIContainer "; } else @@ -1175,7 +945,7 @@ SALOME_ContainerManager::BuildTempFileToLaunchRemoteContainer tempOutputFile << _NS->ContainerName(params) << " -"; AddOmninamesParams(tempOutputFile); - tempOutputFile << " &" << endl; + tempOutputFile << " &" << std::endl; tempOutputFile.flush(); tempOutputFile.close(); #ifndef WIN32 @@ -1184,15 +954,15 @@ SALOME_ContainerManager::BuildTempFileToLaunchRemoteContainer // --- Build command - string command; + std::string command; if (resInfo.Protocol == rsh) { command = "rsh "; - string commandRcp = "rcp "; + std::string commandRcp = "rcp "; commandRcp += _TmpFileName; commandRcp += " "; - commandRcp += machine; + commandRcp += resInfo.HostName; commandRcp += ":"; commandRcp += _TmpFileName; status = system(commandRcp.c_str()); @@ -1201,10 +971,10 @@ SALOME_ContainerManager::BuildTempFileToLaunchRemoteContainer else if (resInfo.Protocol == ssh) { command = "ssh "; - string commandRcp = "scp "; + std::string commandRcp = "scp "; commandRcp += _TmpFileName; commandRcp += " "; - commandRcp += machine; + commandRcp += resInfo.HostName; commandRcp += ":"; commandRcp += _TmpFileName; status = system(commandRcp.c_str()); @@ -1215,7 +985,7 @@ SALOME_ContainerManager::BuildTempFileToLaunchRemoteContainer if(status) throw SALOME_Exception("Error of connection on remote host"); - command += machine; + command += resInfo.HostName; _CommandForRemAccess = command; command += " "; command += _TmpFileName; @@ -1226,149 +996,775 @@ SALOME_ContainerManager::BuildTempFileToLaunchRemoteContainer } +std::string SALOME_ContainerManager::GetMPIZeroNode(const std::string machine, const std::string machinesFile) +{ + int status; + std::string zeronode; + std::string command; + std::string tmpFile = BuildTemporaryFileName(); + + if( getenv("LIBBATCH_NODEFILE") == NULL ) + { + if (_isAppliSalomeDefined) + { + const ParserResourcesType& resInfo = _ResManager->GetImpl()->GetResourcesDescr(machine); + + if (resInfo.Protocol == rsh) + command = "rsh "; + else if (resInfo.Protocol == ssh) + command = "ssh "; + else + throw SALOME_Exception("Unknown protocol"); + + if (resInfo.UserName != "") + { + command += "-l "; + command += resInfo.UserName; + command += " "; + } + + command += resInfo.HostName; + command += " "; + + if (resInfo.AppliPath != "") + command += resInfo.AppliPath; // path relative to user@machine $HOME + else + { + ASSERT(getenv("APPLI")); + command += getenv("APPLI"); // path relative to user@machine $HOME + } + + command += "/runRemote.sh "; + + ASSERT(getenv("NSHOST")); + command += getenv("NSHOST"); // hostname of CORBA name server + + command += " "; + ASSERT(getenv("NSPORT")); + command += getenv("NSPORT"); // port of CORBA name server + + command += " mpirun -np 1 hostname > " + tmpFile; + } + else + command = "mpirun -np 1 hostname > " + tmpFile; + } + else + command = "mpirun -np 1 -machinefile " + machinesFile + " hostname > " + tmpFile; + + status = system(command.c_str()); + if( status == 0 ){ + std::ifstream fp(tmpFile.c_str(),std::ios::in); + while(fp >> zeronode); + } + + RmTmpFile(tmpFile); + + return zeronode; +} + +std::string SALOME_ContainerManager::machinesFile(const int nbproc) +{ + std::string tmp; + std::string nodesFile = getenv("LIBBATCH_NODEFILE"); + std::string machinesFile = Kernel_Utils::GetTmpFileName(); + std::ifstream fpi(nodesFile.c_str(),std::ios::in); + std::ofstream fpo(machinesFile.c_str(),std::ios::out); + + _numInstanceMutex.lock(); + + for(int i=0;i<_nbprocUsed;i++) + fpi >> tmp; + + for(int i=0;i> tmp ) + fpo << tmp << std::endl; + else + throw SALOME_Exception("You ask more processes than batch session have allocated!"); + + _nbprocUsed += nbproc; + fpi.close(); + fpo.close(); + + _numInstanceMutex.unlock(); + + return machinesFile; + +} + +bool +SALOME_ContainerManager::checkPaCOParameters(Engines::ContainerParameters & params, std::string resource_selected) +{ + bool result = true; + + // Step 1 : check ContainerParameters + // Check container_name, has to be defined + if (std::string(params.container_name.in()) == "") + { + INFOS("[checkPaCOParameters] You must define a container_name to launch a PaCO++ container"); + result = false; + } + // Check parallelLib + std::string parallelLib = params.parallelLib.in(); + if (parallelLib != "Mpi" && parallelLib != "Dummy") + { + INFOS("[checkPaCOParameters] parallelLib is not correctly defined"); + INFOS("[checkPaCOParameters] you can chosse between: Mpi and Dummy"); + INFOS("[checkPaCOParameters] you entered: " << parallelLib); + result = false; + } + // Check nb_proc + if (params.nb_proc <= 0) + { + INFOS("[checkPaCOParameters] You must define a nb_proc > 0"); + result = false; + } + + // Step 2 : check resource_selected + Engines::ResourceDefinition_var resource_definition = _ResManager->GetResourceDefinition(resource_selected.c_str()); + std::string protocol = resource_definition->protocol.in(); + std::string username = resource_definition->username.in(); + std::string applipath = resource_definition->applipath.in(); + + if (protocol == "" || username == "" || applipath == "") + { + INFOS("[checkPaCOParameters] resource selected is not well defined"); + INFOS("[checkPaCOParameters] resource name: " << resource_definition->name.in()); + INFOS("[checkPaCOParameters] resource hostname: " << resource_definition->hostname.in()); + INFOS("[checkPaCOParameters] resource protocol: " << protocol); + INFOS("[checkPaCOParameters] resource username: " << username); + INFOS("[checkPaCOParameters] resource applipath: " << applipath); + result = false; + } + + return result; +} +#ifdef WITH_PACO_PARALLEL + //============================================================================= -/*! Creates a command line that the container manager uses to launch - * a parallel container. - */ +/*! CORBA Method: + * Start a suitable PaCO++ Parallel Container in a list of machines. + * \param params Container Parameters required for the container + * \return CORBA container reference. + */ //============================================================================= -string -SALOME_ContainerManager::BuildCommandToLaunchLocalParallelContainer(const std::string& exe_name, - const Engines::MachineParameters& params, - const std::string& log) +Engines::Container_ptr +SALOME_ContainerManager::StartPaCOPPContainer(const Engines::ContainerParameters& params_const, + std::string resource_selected) { - // This method knows the differences between the proxy and the nodes. - // nb_component_nodes is not used in the same way if it is a proxy or - // a node. - - string command; - string parallelLib(CORBA::string_dup(params.parallelLib)); - string hostname(CORBA::string_dup(params.hostname)); - int par = exe_name.find("Proxy"); - int nbproc = params.nb_component_nodes; - char buffer [33]; - sprintf(buffer,"%d",nbproc); - - Engines::MachineParameters_var rtn = new Engines::MachineParameters(); - rtn->container_name = params.container_name; - rtn->hostname = params.hostname; - rtn->OS = params.OS; - rtn->mem_mb = params.mem_mb; - rtn->cpu_clock = params.cpu_clock; - rtn->nb_proc_per_node = params.nb_proc_per_node; - rtn->nb_node = params.nb_node; - rtn->isMPI = params.isMPI; - - string real_exe_name = exe_name + parallelLib; - - if (parallelLib == "Dummy") + CORBA::Object_var obj; + PaCO::InterfaceManager_var container_proxy; + Engines::Container_ptr ret = Engines::Container::_nil(); + Engines::ContainerParameters params(params_const); + params.resource_params.name = CORBA::string_dup(resource_selected.c_str()); + + // Step 0 : Check parameters + if (!checkPaCOParameters(params, resource_selected)) { - //command = "gdb --args "; - //command = "valgrind --tool=memcheck --log-file=val_log "; - //command += real_exe_name; + INFOS("[StartPaCOPPContainer] check parameters failed ! see logs..."); + return ret; + } - command = real_exe_name; + // Step 1 : Starting a new parallel container ! + INFOS("[StartPaCOPPContainer] Starting a PaCO++ parallel container"); + INFOS("[StartPaCOPPContainer] on resource : " << resource_selected); - command += " " + _NS->ContainerName(rtn); - command += " " + parallelLib; - command += " " + hostname; - command += " -"; - AddOmninamesParams(command); + // Step 2 : Get a MachineFile for the parallel container + std::string machine_file_name = _ResManager->getMachineFile(resource_selected, + params.nb_proc, + params.parallelLib.in()); + + if (machine_file_name == "") + { + INFOS("[StartPaCOPPContainer] Machine file generation failed"); + return ret; + } + + // Step 3 : starting parallel container proxy + std::string command_proxy(""); + std::string proxy_machine; + try + { + command_proxy = BuildCommandToLaunchPaCOProxyContainer(params, machine_file_name, proxy_machine); + } + catch(const SALOME_Exception & ex) + { + INFOS("[StartPaCOPPContainer] Exception in BuildCommandToLaunchPaCOContainer"); + INFOS(ex.what()); + return ret; + } + obj = LaunchPaCOProxyContainer(command_proxy, params, proxy_machine); + if (CORBA::is_nil(obj)) + { + INFOS("[StartPaCOPPContainer] LaunchPaCOContainer for proxy returns NIL !"); + return ret; + } + container_proxy = PaCO::InterfaceManager::_narrow(obj); + MESSAGE("[StartPaCOPPContainer] PaCO container proxy is launched"); + + // Step 4 : starting parallel container nodes + std::string command_nodes(""); + SALOME_ContainerManager::actual_launch_machine_t nodes_machines; + try + { + command_nodes = BuildCommandToLaunchPaCONodeContainer(params, machine_file_name, nodes_machines, proxy_machine); + } + catch(const SALOME_Exception & ex) + { + INFOS("[StarPaCOPPContainer] Exception in BuildCommandToLaunchPaCONodeContainer"); + INFOS(ex.what()); + return ret; } - else if (parallelLib == "Mpi") + std::string container_generic_node_name = std::string(params.container_name.in()) + std::string("Node"); + bool result = LaunchPaCONodeContainer(command_nodes, params, container_generic_node_name, nodes_machines); + if (!result) { - // Step 1 : check if MPI is started - if (_MpiStarted == false) + INFOS("[StarPaCOPPContainer] LaunchPaCONodeContainer failed !"); + // Il faut tuer le proxy + try + { + Engines::Container_var proxy = Engines::Container::_narrow(container_proxy); + proxy->Shutdown(); + } + catch (...) { - startMPI(); + INFOS("[StarPaCOPPContainer] Exception catched from proxy Shutdown..."); } + return ret; + } - if (par < 0) + // Step 4 : connecting nodes and the proxy to actually create a parallel container + for (int i = 0; i < params.nb_proc; i++) + { + std::ostringstream tmp; + tmp << i; + std::string proc_number = tmp.str(); + std::string container_node_name = container_generic_node_name + proc_number; + + std::string theNodeMachine(nodes_machines[i]); + std::string containerNameInNS = _NS->BuildContainerNameForNS(container_node_name.c_str(), theNodeMachine.c_str()); + obj = _NS->Resolve(containerNameInNS.c_str()); + if (CORBA::is_nil(obj)) + { + INFOS("[StarPaCOPPContainer] CONNECTION FAILED From Naming Service !"); + INFOS("[StarPaCOPPContainer] Container name is " << containerNameInNS); + return ret; + } + try + { + MESSAGE("[StarPaCOPPContainer] Deploying node : " << container_node_name); + PaCO::InterfaceParallel_var node = PaCO::InterfaceParallel::_narrow(obj); + node->deploy(); + MESSAGE("[StarPaCOPPContainer] node " << container_node_name << " is deployed"); + } + catch(CORBA::SystemException& e) { - // Nodes case - - command = "mpiexec -np " + string(buffer) + " "; -// command += "gdb --args "; - command += real_exe_name; - command += " " + _NS->ContainerName(rtn); - command += " " + parallelLib; - command += " " + hostname; - command += " -"; - AddOmninamesParams(command); + INFOS("[StarPaCOPPContainer] Exception in deploying node : " << containerNameInNS); + INFOS("CORBA::SystemException : " << e); + return ret; } - else + catch(CORBA::Exception& e) { - // Proxy case - command = "mpiexec -np 1 "; - command += real_exe_name; - command += " " + _NS->ContainerName(rtn); - command += " " + string(buffer); - command += " " + parallelLib; - command += " " + hostname; - command += " -"; - AddOmninamesParams(command); + INFOS("[StarPaCOPPContainer] Exception in deploying node : " << containerNameInNS); + INFOS("CORBA::Exception" << e); + return ret; + } + catch(...) + { + INFOS("[StarPaCOPPContainer] Exception in deploying node : " << containerNameInNS); + INFOS("Unknown exception !"); + return ret; } } - else + + // Step 5 : starting parallel container + try + { + MESSAGE ("[StarPaCOPPContainer] Starting parallel object"); + container_proxy->start(); + MESSAGE ("[StarPaCOPPContainer] Parallel object is started"); + ret = Engines::Container::_narrow(container_proxy); + } + catch(CORBA::SystemException& e) + { + INFOS("Caught CORBA::SystemException. : " << e); + } + catch(PortableServer::POA::ServantAlreadyActive&) { - std::string message("Unknown parallelLib" + parallelLib); - throw SALOME_Exception(message.c_str()); + INFOS("Caught CORBA::ServantAlreadyActiveException"); } + catch(CORBA::Exception&) + { + INFOS("Caught CORBA::Exception."); + } + catch(std::exception& exc) + { + INFOS("Caught std::exception - "<GetResourceDefinition(params.resource_params.name); + + // Choose hostname + std::string hostname; + std::ifstream machine_file(machine_file_name.c_str()); + std::getline(machine_file, hostname, ' '); + size_t found = hostname.find('\n'); + if (found!=std::string::npos) + hostname.erase(found, 1); // Remove \n + proxy_hostname = hostname; + MESSAGE("[BuildCommandToLaunchPaCOProxyContainer] machine file name extracted is " << hostname); + + // Remote execution + bool remote_execution = false; + if (hostname != std::string(Kernel_Utils::GetHostname())) { - command += " > /tmp/"; - command += _NS->ContainerName(rtn); - command += "_"; - command += Kernel_Utils::GetHostname(); - command += "_"; - command += getenv( "USER" ) ; - command += ".log 2>&1 &" ; + MESSAGE("[BuildCommandToLaunchPaCOProxyContainer] remote machine case detected !"); + remote_execution = true; } - if (log == "xterm") + + // Log environnement + std::string log_type(""); + char * get_val = getenv("PARALLEL_LOG"); + if (get_val) + log_type = get_val; + + // Generating the command + std::string command_begin(""); + std::string command_end(""); + std::ostringstream command; + + LogConfiguration(log_type, "proxy", container_name, hostname, command_begin, command_end); + command << command_begin; + + // Adding connection command + // We can only have a remote execution with + // a SALOME application + if (remote_execution) { - command = "/usr/X11R6/bin/xterm -e \"export LD_LIBRARY_PATH=$LD_LIBRARY_PATH; export PATH=$PATH; " - + command + " \" &"; -// + command + "; echo $LD_LIBRARY_PATH; cat \" &"; + ASSERT(getenv("NSHOST")); + ASSERT(getenv("NSPORT")); + + command << resource_definition->protocol.in(); + command << " -l "; + command << resource_definition->username.in(); + command << " " << hostname; + command << " " << resource_definition->applipath.in(); + command << "/runRemote.sh "; + command << getenv("NSHOST") << " "; // hostname of CORBA name server + command << getenv("NSPORT") << " "; // port of CORBA name server } - return command; -/* if (log == "xterm") + command << exe_name; + command << " " << container_name; + command << " Dummy"; + command << " " << hostname; + command << " " << nb_proc_str; + command << " -"; + AddOmninamesParams(command); + + // Final command + command << command_end; + MESSAGE("[BuildCommandToLaunchPaCOProxyContainer] Command is: " << command.str()); + + return command.str(); +} + +std::string +SALOME_ContainerManager::BuildCommandToLaunchPaCONodeContainer(const Engines::ContainerParameters& params, + const std::string & machine_file_name, + SALOME_ContainerManager::actual_launch_machine_t & vect_machine, + const std::string & proxy_hostname) +{ + // Name of exe + std::string exe_name = "SALOME_ParallelContainerNode"; + exe_name += params.parallelLib.in(); + std::string container_name = params.container_name.in(); + + // Convert nb_proc in string + std::ostringstream nb_proc_stream; + nb_proc_stream << params.nb_proc; + + // Get resource definition + Engines::ResourceDefinition_var resource_definition = + _ResManager->GetResourceDefinition(params.resource_params.name); + + // Log environnement + std::string log_type(""); + char * get_val = getenv("PARALLEL_LOG"); + if (get_val) + log_type = get_val; + + // Now the command is different according to paralleLib + std::ostringstream command_nodes; + std::ifstream machine_file(machine_file_name.c_str()); + if (std::string(params.parallelLib.in()) == "Dummy") + { + for (int i= 0; i < params.nb_proc; i++) + { + // Choose hostname + std::string hostname; + std::getline(machine_file, hostname); + MESSAGE("[BuildCommandToLaunchPaCONodeContainer] machine file name extracted is " << hostname); + + // Remote execution + bool remote_execution = false; + if (hostname != std::string(Kernel_Utils::GetHostname())) + { + MESSAGE("[BuildCommandToLaunchPaCONodeContainer] remote machine case detected !"); + remote_execution = true; + } + + // For each node we have a new command + // Generating the command + std::ostringstream command_node_stream; + std::string command_node_begin(""); + std::string command_node_end(""); + std::ostringstream node_number; + node_number << i; + std::string container_node_name = container_name + node_number.str(); + LogConfiguration(log_type, "node", container_node_name, hostname, command_node_begin, command_node_end); + + // Adding connection command + // We can only have a remote execution with + // a SALOME application + if (remote_execution) + { + ASSERT(getenv("NSHOST")); + ASSERT(getenv("NSPORT")); + + command_node_stream << resource_definition->protocol.in(); + command_node_stream << " -l "; + command_node_stream << resource_definition->username.in(); + command_node_stream << " " << hostname; + command_node_stream << " " << resource_definition->applipath.in(); + command_node_stream << "/runRemote.sh "; + command_node_stream << getenv("NSHOST") << " "; // hostname of CORBA name server + command_node_stream << getenv("NSPORT") << " "; // port of CORBA name server + } + + command_node_stream << exe_name; + command_node_stream << " " << container_name; + command_node_stream << " " << params.parallelLib.in(); + command_node_stream << " " << proxy_hostname; + command_node_stream << " " << node_number.str(); + command_node_stream << " -"; + AddOmninamesParams(command_node_stream); + + command_nodes << command_node_begin << command_node_stream.str() << command_node_end; + vect_machine.push_back(hostname); + } + } + + else if (std::string(params.parallelLib.in()) == "Mpi") { - command = "/usr/X11R6/bin/xterm -e \"export LD_LIBRARY_PATH=$LD_LIBRARY_PATH; export PATH=$PATH; echo $LD_LIBRARY_PATH; echo $PATH; " + command + "; cat \" &"; + // Choose hostname + std::string hostname; + std::getline(machine_file, hostname, ' '); + MESSAGE("[BuildCommandToLaunchPaCONodeContainer] machine file name extracted is " << hostname); + + // Remote execution + bool remote_execution = false; + if (hostname != std::string(Kernel_Utils::GetHostname())) + { + MESSAGE("[BuildCommandToLaunchPaCONodeContainer] remote machine case detected !"); + remote_execution = true; + } + + // In case of Mpi and Remote, we copy machine_file in the applipath + // scp mpi_machine_file user@machine:Path + std::ostringstream command_remote_stream; + std::string::size_type last = machine_file_name.find_last_of("/"); + if (last == std::string::npos) + last = -1; + + std::string protocol = resource_definition->protocol.in(); + if (protocol == "rsh") + command_remote_stream << "rcp "; + else + command_remote_stream << "scp "; + command_remote_stream << machine_file_name << " "; + command_remote_stream << resource_definition->username.in() << "@"; + command_remote_stream << hostname << ":" << resource_definition->applipath.in(); + command_remote_stream << "/" << machine_file_name.substr(last+1); + + int status = system(command_remote_stream.str().c_str()); + if (status == -1) + { + INFOS("copy of the MPI machine file failed ! - sorry !"); + return ""; + } + + // Generating the command + std::string command_begin(""); + std::string command_end(""); + + LogConfiguration(log_type, "nodes", container_name, hostname, command_begin, command_end); + command_nodes << command_begin; + + // Adding connection command + // We can only have a remote execution with + // a SALOME application + if (remote_execution) + { + ASSERT(getenv("NSHOST")); + ASSERT(getenv("NSPORT")); + + command_nodes << resource_definition->protocol.in(); + command_nodes << " -l "; + command_nodes << resource_definition->username.in(); + command_nodes << " " << hostname; + command_nodes << " " << resource_definition->applipath.in(); + command_nodes << "/runRemote.sh "; + command_nodes << getenv("NSHOST") << " "; // hostname of CORBA name server + command_nodes << getenv("NSPORT") << " "; // port of CORBA name server + } + + if (std::string(resource_definition->mpiImpl.in()) == "lam") + { + command_nodes << "mpiexec -ssi boot "; + command_nodes << "-machinefile " << machine_file_name << " "; + command_nodes << "-n " << params.nb_proc; + } + else + { + command_nodes << "mpirun -np " << params.nb_proc; + } + command_nodes << " " << exe_name; + command_nodes << " " << container_name; + command_nodes << " " << params.parallelLib.in(); + command_nodes << " " << proxy_hostname; + command_nodes << " -"; + AddOmninamesParams(command_nodes); + + // We don't put hostname, because nodes are registered in the resource of the proxy + for (int i= 0; i < params.nb_proc; i++) + vect_machine.push_back(proxy_hostname); + + command_nodes << command_end; } -*/ -/* command = "cd ; rm " + fichier_commande + "; touch " + \ - fichier_commande + "; echo \" export LD_LIBRARY_PATH=$LD_LIBRARY_PATH; " + \ - command + " >& /tmp/ribes_" + fichier_commande + " & \" > " + fichier_commande + ";"; - command += "ssh cn01 sh " + fichier_commande + " &"; - cerr << "La commande : " << command << endl; -*/ + return command_nodes.str(); } -void SALOME_ContainerManager::startMPI() +void +SALOME_ContainerManager::LogConfiguration(const std::string & log_type, + const std::string & exe_type, + const std::string & container_name, + const std::string & hostname, + std::string & begin, + std::string & end) { - cerr << "----------------------------------------------" << endl; - cerr << "----------------------------------------------" << endl; - cerr << "----------------------------------------------" << endl; - cerr << "-Only Lam on Localhost is currently supported-" << endl; - cerr << "----------------------------------------------" << endl; - cerr << "----------------------------------------------" << endl; - cerr << "----------------------------------------------" << endl; - - int status = system("lamboot"); - if (status == -1) + if(log_type == "xterm") { - INFOS("lamboot failed : system command status -1"); + begin = "xterm -e \"export LD_LIBRARY_PATH=$LD_LIBRARY_PATH; export PATH=$PATH;"; + end = "\"&"; } - else if (status == 217) + else if(log_type == "xterm_debug") { - INFOS("lamboot failed : system command status 217"); + begin = "xterm -e \"export LD_LIBRARY_PATH=$LD_LIBRARY_PATH; export PATH=$PATH;"; + end = "; cat \" &"; } else { - _MpiStarted = true; + // default into a file... + std::string logFilename = "/tmp/" + container_name + "_" + hostname + "_" + exe_type + "_"; + logFilename += std::string(getenv("USER")) + ".log"; + end = " > " + logFilename + " 2>&1 & "; + } +} + +CORBA::Object_ptr +SALOME_ContainerManager::LaunchPaCOProxyContainer(const std::string& command, + const Engines::ContainerParameters& params, + const std::string & hostname) +{ + PaCO::InterfaceManager_ptr container_proxy = PaCO::InterfaceManager::_nil(); + + MESSAGE("[LaunchPaCOProxyContainer] Launch command"); + int status = system(command.c_str()); + if (status == -1) { + INFOS("[LaunchPaCOProxyContainer] failed : system command status -1"); + return container_proxy; + } + else if (status == 217) { + INFOS("[LaunchPaCOProxyContainer] failed : system command status 217"); + return container_proxy; + } + + int count = TIME_OUT_TO_LAUNCH_CONT; + CORBA::Object_var obj = CORBA::Object::_nil(); + std::string containerNameInNS = _NS->BuildContainerNameForNS(params.container_name.in(), + hostname.c_str()); + MESSAGE("[LaunchParallelContainer] Waiting for Parallel Container proxy : " << containerNameInNS); + + while (CORBA::is_nil(obj) && count) + { + sleep(1); + count--; + obj = _NS->Resolve(containerNameInNS.c_str()); + } + + try + { + container_proxy = PaCO::InterfaceManager::_narrow(obj); + } + catch(CORBA::SystemException& e) + { + INFOS("[StarPaCOPPContainer] Exception in _narrow after LaunchParallelContainer for proxy !"); + INFOS("CORBA::SystemException : " << e); + return container_proxy; + } + catch(CORBA::Exception& e) + { + INFOS("[StarPaCOPPContainer] Exception in _narrow after LaunchParallelContainer for proxy !"); + INFOS("CORBA::Exception" << e); + return container_proxy; + } + catch(...) + { + INFOS("[StarPaCOPPContainer] Exception in _narrow after LaunchParallelContainer for proxy !"); + INFOS("Unknown exception !"); + return container_proxy; + } + if (CORBA::is_nil(container_proxy)) + { + INFOS("[StarPaCOPPContainer] PaCO::InterfaceManager::_narrow returns NIL !"); + return container_proxy; + } + return obj._retn(); +} + +//============================================================================= +/*! This method launches the parallel container. + * It will may be placed on the ressources manager. + * + * \param command to launch + * \param container's parameters + * \param name of the container + * + * \return CORBA container reference + */ +//============================================================================= +bool +SALOME_ContainerManager::LaunchPaCONodeContainer(const std::string& command, + const Engines::ContainerParameters& params, + const std::string& name, + SALOME_ContainerManager::actual_launch_machine_t & vect_machine) +{ + INFOS("[LaunchPaCONodeContainer] Launch command"); + int status = system(command.c_str()); + if (status == -1) { + INFOS("[LaunchPaCONodeContainer] failed : system command status -1"); + return false; + } + else if (status == 217) { + INFOS("[LaunchPaCONodeContainer] failed : system command status 217"); + return false; + } + + INFOS("[LaunchPaCONodeContainer] Waiting for the nodes of the parallel container"); + // We are waiting all the nodes + for (int i = 0; i < params.nb_proc; i++) + { + CORBA::Object_var obj = CORBA::Object::_nil(); + std::string theMachine(vect_machine[i]); + // Name of the node + std::ostringstream tmp; + tmp << i; + std::string proc_number = tmp.str(); + std::string container_node_name = name + proc_number; + std::string containerNameInNS = _NS->BuildContainerNameForNS((char*) container_node_name.c_str(), theMachine.c_str()); + INFOS("[LaunchPaCONodeContainer] Waiting for Parallel Container node " << containerNameInNS << " on " << theMachine); + int count = TIME_OUT_TO_LAUNCH_CONT; + while (CORBA::is_nil(obj) && count) { + sleep(1) ; + count-- ; + obj = _NS->Resolve(containerNameInNS.c_str()); + } + if (CORBA::is_nil(obj)) + { + INFOS("[LaunchPaCONodeContainer] Launch of node failed (or not found) !"); + return false; + } } + return true; +} + +#else + +Engines::Container_ptr +SALOME_ContainerManager::StartPaCOPPContainer(const Engines::ContainerParameters& params, + std::string resource_selected) +{ + Engines::Container_ptr ret = Engines::Container::_nil(); + INFOS("[StarPaCOPPContainer] is disabled !"); + INFOS("[StarPaCOPPContainer] recompile SALOME Kernel to enable PaCO++ parallel extension"); + return ret; +} + +std::string +SALOME_ContainerManager::BuildCommandToLaunchPaCOProxyContainer(const Engines::ContainerParameters& params, + std::string machine_file_name, + std::string & proxy_hostname) +{ + return ""; +} + +std::string +SALOME_ContainerManager::BuildCommandToLaunchPaCONodeContainer(const Engines::ContainerParameters& params, + const std::string & machine_file_name, + SALOME_ContainerManager::actual_launch_machine_t & vect_machine, + const std::string & proxy_hostname) +{ + return ""; +} +void +SALOME_ContainerManager::LogConfiguration(const std::string & log_type, + const std::string & exe_type, + const std::string & container_name, + const std::string & hostname, + std::string & begin, + std::string & end) +{ +} + +CORBA::Object_ptr +SALOME_ContainerManager::LaunchPaCOProxyContainer(const std::string& command, + const Engines::ContainerParameters& params, + const std::string& hostname) +{ + CORBA::Object_ptr ret = CORBA::Object::_nil(); + return ret; } +bool +SALOME_ContainerManager::LaunchPaCONodeContainer(const std::string& command, + const Engines::ContainerParameters& params, + const std::string& name, + SALOME_ContainerManager::actual_launch_machine_t & vect_machine) +{ + return false; +} +#endif +