X-Git-Url: http://git.salome-platform.org/gitweb/?a=blobdiff_plain;f=src%2FContainer%2FSALOME_ContainerManager.cxx;h=f960e152c1c96d41d06afc39c2dd2fce51890ede;hb=110bda7bc3b5ebcb94cfdf9c2d93f580ae773822;hp=989ca2c4bf858d7abdaf58db4146931bb5af520a;hpb=6f4ae012f30e3e0685a39eb747b1ea5b33c085a6;p=modules%2Fkernel.git diff --git a/src/Container/SALOME_ContainerManager.cxx b/src/Container/SALOME_ContainerManager.cxx index 989ca2c4b..f960e152c 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" @@ -26,29 +27,31 @@ #include "Basics_DirUtils.hxx" #include #include +#include #ifndef WIN32 #include #endif #include #include "Utils_CorbaException.hxx" -#include "Batch_Date.hxx" #include +#ifdef WNT +#include +#define getpid _getpid +#endif + #ifdef WITH_PACO_PARALLEL #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 @@ -58,7 +61,7 @@ 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; @@ -81,6 +84,37 @@ SALOME_ContainerManager::SALOME_ContainerManager(CORBA::ORB_ptr orb, PortableSer _NS->Register(refContMan,_ContainerManagerNameInNS); _isAppliSalomeDefined = (getenv("APPLI") != 0); + +#ifdef HAVE_MPI2 +#ifdef WITHOPENMPI + _pid_ompiServer = -1; + // the urifile name depends on pid of the process + std::stringstream urifile; + urifile << getenv("HOME") << "/.urifile_" << getpid(); + setenv("OMPI_URI_FILE",urifile.str().c_str(),1); + if( getenv("OMPI_URI_FILE") != NULL ){ + // get the pid of all ompi-server + std::set thepids1 = getpidofprogram("ompi-server"); + // launch a new 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"); + // get the pid of all ompi-server + std::set thepids2 = getpidofprogram("ompi-server"); + // my ompi-server is the new one + std::set::const_iterator it; + for(it=thepids2.begin();it!=thepids2.end();it++) + if(thepids1.find(*it) == thepids1.end()) + _pid_ompiServer = *it; + if(_pid_ompiServer < 0) + throw SALOME_Exception("Error when getting ompi-server id"); + } +#endif +#endif + MESSAGE("constructor end"); } @@ -93,6 +127,19 @@ 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 ){ + // kill my ompi-server + if( kill(_pid_ompiServer,SIGTERM) != 0 ) + throw SALOME_Exception("Error when killing ompi-server"); + // delete my urifile + int status=system("rm -f ${OMPI_URI_FILE}"); + if(status!=0) + throw SALOME_Exception("Error when removing urifile"); + } +#endif +#endif } //============================================================================= @@ -122,9 +169,9 @@ 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()); @@ -140,35 +187,35 @@ 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++) + 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)); - cont->Shutdown(); - } - else - MESSAGE("ShutdownContainers: no container ref for " << (*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)) + { + MESSAGE("ShutdownContainers: " << (*iter)); + cont->Shutdown(); + } + else + MESSAGE("ShutdownContainers: no container ref for " << (*iter)); } catch(CORBA::SystemException& e) { - INFOS("CORBA::SystemException ignored : " << e); + INFOS("CORBA::SystemException ignored : " << e); } catch(CORBA::Exception&) { - INFOS("CORBA::Exception ignored."); + INFOS("CORBA::Exception ignored."); } catch(...) { - INFOS("Unknown exception ignored."); + INFOS("Unknown exception ignored."); } } } @@ -177,144 +224,234 @@ void SALOME_ContainerManager::ShutdownContainers() //============================================================================= //! Give a suitable Container given constraints /*! CORBA Method: - * \param params Machine Parameters required for the container + * \param params Container Parameters required for the container * \return the container or nil */ //============================================================================= - Engines::Container_ptr -SALOME_ContainerManager::GiveContainer(const Engines::MachineParameters& params) +SALOME_ContainerManager::GiveContainer(const Engines::ContainerParameters& params) { - char *valenv=getenv("SALOME_BATCH"); - if(valenv) - if (strcmp(valenv,"1")==0) + std::string machFile; + Engines::Container_ptr ret = Engines::Container::_nil(); + + // 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") { - if(_batchLaunchedContainers.empty()) - fillBatchLaunchedContainers(); + MESSAGE("[GiveContainer] no container found"); + return ret; + } + else + { + mode = "start"; + } + } + } - if (_batchLaunchedContainersIter == _batchLaunchedContainers.end()) - _batchLaunchedContainersIter = _batchLaunchedContainers.begin(); + // 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; - Engines::Container_ptr rtn = Engines::Container::_duplicate(*_batchLaunchedContainersIter); - _batchLaunchedContainersIter++; - return rtn; + // 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])); } - return StartContainer(params); -} + catch(CORBA::Exception&) {} + } -//============================================================================= -//! Start a suitable Container in a list of machines with constraints -/*! C++ Method: - * Constraints are given by a machine parameters struct - * \param params Machine Parameters required for the container - * \param possibleComputers list of machines usable for start - * \param container_exe specific container executable (default=SALOME_Container) - */ -//============================================================================= + // 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 + for(unsigned int i=0; i < possibleResources->length(); i++) + local_resources.push_back(std::string(possibleResources[i])); -Engines::Container_ptr -SALOME_ContainerManager::StartContainer(const Engines::MachineParameters& params, - const Engines::MachineList& possibleComputers, - const std::string& container_exe) -{ -#ifdef WITH_PACO_PARALLEL - std::string parallelLib(params.parallelLib); - if (parallelLib != "") + // Step 4: select the resource where to get/start the container + std::string resource_selected; + try { - Engines::MachineParameters myparams(params); - myparams.computerList=possibleComputers; - return StartParallelContainer(myparams); + resource_selected = _ResManager->GetImpl()->Find(params.resource_params.policy.in(), local_resources); } -#endif - string containerNameInNS; - Engines::Container_ptr ret = Engines::Container::_nil(); + catch(const SALOME_Exception &ex) + { + MESSAGE("[GiveContainer] Exception in ResourceManager find !: " << ex.what()); + return ret; + } + MESSAGE("[GiveContainer] Resource selected is: " << resource_selected); - MESSAGE("SALOME_ContainerManager::StartContainer " << possibleComputers.length()); + // 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, hostname.c_str()); + MESSAGE("[GiveContainer] Container name in the naming service: " << containerNameInNS); - vector lm; -// if mode is "get" keep only machines with existing containers - if(std::string(params.mode.in())=="get") + // 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)) + { + try { - for(unsigned int i=0;i_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 { - Engines::Container_ptr cont = FindContainer(params,possibleComputers[i]); - try - { - if(!cont->_non_existent()) - lm.push_back(string(possibleComputers[i])); - } - catch(CORBA::Exception&) - { - // CORBA::Exception ignored. - } + INFOS("[GiveContainer] A container is already registered with the name: " << containerNameInNS << ", shutdown the existing container"); + cont->Shutdown(); // shutdown the registered container if it exists } + } } - else + catch(CORBA::Exception&) { - for(unsigned int i=0;i 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)) { - theMachine=_ResManager->GetImpl()->Find(params.policy.in(),lm); - } - catch( const SALOME_Exception &ex ) - { - MESSAGE(ex.what()); - return Engines::Container::_nil(); + INFOS("[GiveContainer] Module Catalog is not found -> cannot launch a container"); + return ret; } - - //If the machine name is localhost use the real name - if(theMachine == "localhost") - theMachine=Kernel_Utils::GetHostname(); - - //check if an entry exists in Naming service - //if params.mode == "start" or "" shutdown the existing container before launching a new one with that name - //if params.mode == "getorstart" or "get" use the existing container - containerNameInNS = _NS->BuildContainerNameForNS(params,theMachine.c_str()); - - SCRUTE(containerNameInNS); - CORBA::Object_var obj = _NS->Resolve(containerNameInNS.c_str()); - if ( !CORBA::is_nil(obj) ) + // Loop through component list + for(unsigned int i=0; i < local_params.resource_params.componentList.length(); i++) { - 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 - { - INFOS("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&) + 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; + } - //try to launch a new container - MESSAGE("try to launch it on " << theMachine); - - string command; - if(theMachine==""){ - MESSAGE("SALOME_ContainerManager::StartContainer : no possible computer"); - return Engines::Container::_nil(); + // 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; } - else if(theMachine==Kernel_Utils::GetHostname()) - command = BuildCommandToLaunchLocalContainer(params,container_exe); + // 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(theMachine,params,container_exe); + 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()); @@ -322,124 +459,47 @@ SALOME_ContainerManager::StartContainer(const Engines::MachineParameters& params if (status == -1){ MESSAGE("SALOME_ContainerManager::StartContainer rsh failed (system command status -1)"); RmTmpFile(_TmpFileName); // command file can be removed here + _TmpFileName=""; return Engines::Container::_nil(); } else if (status == 217){ MESSAGE("SALOME_ContainerManager::StartContainer rsh failed (system command status 217)"); RmTmpFile(_TmpFileName); // command file can be removed here + _TmpFileName=""; 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_ContainerManager::StartContainer rsh failed"); - } - else - { - logFilename=":"+logFilename; - logFilename="@"+Kernel_Utils::GetHostname()+logFilename; - logFilename=getenv( "USER" )+logFilename; - ret->logfilename(logFilename.c_str()); - } - - RmTmpFile(_TmpFileName); // command file can be removed here - return ret; - } -} - -//============================================================================= -//! Start a suitable Container given constraints -/*! CORBA Method: - * \param params Machine Parameters required for the container - */ -//============================================================================= - -Engines::Container_ptr -SALOME_ContainerManager::StartContainer(const Engines::MachineParameters& params) -{ - Engines::MachineList_var possibleComputers = _ResManager->GetFittingResources(params); - - // Look into ModulCatalog if a specific container must be launched - CORBA::String_var container_exe; - int found=0; - try - { - 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&) + if (CORBA::is_nil(ret)) { - 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(); + // Setting log file name + logFilename=":"+logFilename; + logFilename="@"+Kernel_Utils::GetHostname()+logFilename; + logFilename=user+logFilename; + ret->logfilename(logFilename.c_str()); + RmTmpFile(_TmpFileName); // command file can be removed here + _TmpFileName=""; } - - if(found) - return StartContainer(params,possibleComputers,container_exe.in()); - else - return StartContainer(params,possibleComputers); -} - -//============================================================================= -//! Find or start a suitable Container given some constraints -/*! CORBA Method: - * \param params Machine Parameters required for the container - * \return the container or nil - */ -//============================================================================= - -Engines::Container_ptr -SALOME_ContainerManager::FindOrStartContainer(const Engines::MachineParameters& params) -{ - Engines::Container_ptr ret = FindContainer(params,params.computerList); - if(!CORBA::is_nil(ret)) - return ret; - MESSAGE("Container doesn't exist try to launch it ..."); - - return StartContainer(params); + } + return ret; } //============================================================================= @@ -450,18 +510,17 @@ SALOME_ContainerManager::FindOrStartContainer(const Engines::MachineParameters& //============================================================================= 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(params,theMachine)); + 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); - } - catch(const CORBA::Exception& e) - { + { + if(obj->_non_existent()) return Engines::Container::_nil(); - } + else + return Engines::Container::_narrow(obj); + } + catch(const CORBA::Exception& e) + { + return Engines::Container::_nil(); + } } -#ifdef WITH_PACO_PARALLEL + +bool isPythonContainer(const char* ContainerName); + //============================================================================= -/*! CORBA Method: - * Find or Start a suitable PaCO++ Parallel Container in a list of machines. - * \param params Machine Parameters required for the container - * \return CORBA container reference. - */ +/*! + * This is no longer valid (C++ container are also python containers) + */ //============================================================================= -Engines::Container_ptr -SALOME_ContainerManager::StartParallelContainer(const Engines::MachineParameters& params_const) +bool isPythonContainer(const char* ContainerName) { - CORBA::Object_var obj; - PaCO::InterfaceManager_var container_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, params.computerList); - if(CORBA::is_nil(ret)) { - // Step 2 : Starting a new parallel container ! - INFOS("[StartParallelContainer] Starting a PaCO++ parallel container"); - - // Step 3 : Choose a computer - std::string theMachine = _ResManager->FindFirst(params.computerList); - //If the machine name is localhost use the real name - if(theMachine == "localhost") - theMachine=Kernel_Utils::GetHostname(); - - if(theMachine == "") { - INFOS("[StartParallelContainer] !!!!!!!!!!!!!!!!!!!!!!!!!!"); - INFOS("[StartParallelContainer] No possible computer found"); - INFOS("[StartParallelContainer] !!!!!!!!!!!!!!!!!!!!!!!!!!"); - return ret; - } - INFOS("[StartParallelContainer] on machine : " << theMachine); - params.hostname = CORBA::string_dup(theMachine.c_str()); + bool ret = false; + int len = strlen(ContainerName); - // Step 4 : starting parallel container proxy - Engines::MachineParameters params_proxy(params); - std::string command_proxy; - SALOME_ContainerManager::actual_launch_machine_t proxy_machine; - try + if (len >= 2) + if (strcmp(ContainerName + len - 2, "Py") == 0) + ret = true; + + return ret; +} + +//============================================================================= +/*! + * Builds the script to be launched + * + * If SALOME Application not defined ($APPLI), + * see BuildTempFileToLaunchRemoteContainer() + * + * Else rely on distant configuration. Command is under the form (example): + * ssh user@machine distantPath/runRemote.sh hostNS portNS WORKINGDIR workingdir \ + * SALOME_Container containerName &" + + * - where user is ommited if not specified in CatalogResources, + * - where distant path is always relative to user@machine $HOME, and + * equal to $APPLI if not specified in CatalogResources, + * - where hostNS is the hostname of CORBA naming server (set by scripts to + * use to launch SALOME and servers in $APPLI: runAppli.sh, runRemote.sh) + * - where portNS is the port used by CORBA naming server (set by scripts to + * use to launch SALOME and servers in $APPLI: runAppli.sh, runRemote.sh) + * - where workingdir is the requested working directory for the container. + * If WORKINGDIR (and workingdir) is not present the working dir will be $HOME + */ +//============================================================================= + +std::string +SALOME_ContainerManager::BuildCommandToLaunchRemoteContainer +(const std::string& resource_name, + const Engines::ContainerParameters& params, const std::string& container_exe) +{ + + std::string command; + if (!_isAppliSalomeDefined) + command = BuildTempFileToLaunchRemoteContainer(resource_name, params); + else + { + 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 (params.isMPI) { - command_proxy = BuildCommandToLaunchParallelContainer("SALOME_ParallelContainerProxy", params_proxy, proxy_machine); + if ( params.nb_proc <= 0 ) + nbproc = 1; + else + nbproc = params.nb_proc; } - catch(const SALOME_Exception & ex) + + // "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"); + + if (resInfo.UserName != "") { - INFOS("[StartParallelContainer] Exception in BuildCommandToLaunchParallelContainer"); - INFOS(ex.what()); - return ret; + command += "-l "; + command += resInfo.UserName; + command += " "; } - params_proxy.nb_component_nodes = 0; // LaunchParallelContainer uses this value to know if it launches the proxy or the nodes - obj = LaunchParallelContainer(command_proxy, params_proxy, _NS->ContainerName(params_proxy), proxy_machine); - if (CORBA::is_nil(obj)) + + command += resInfo.HostName; + command += " "; + + if (resInfo.AppliPath != "") + command += resInfo.AppliPath; // path relative to user@machine $HOME + else { - INFOS("[StartParallelContainer] LaunchParallelContainer for proxy returns NIL !"); - return ret; + ASSERT(getenv("APPLI")); + command += getenv("APPLI"); // path relative to user@machine $HOME } - try + + 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 + + std::string wdir = params.workingdir.in(); + if(wdir != "") { - container_proxy = PaCO::InterfaceManager::_narrow(obj); - } - catch(CORBA::SystemException& e) - { - INFOS("[StartParallelContainer] Exception in _narrow after LaunchParallelContainer for proxy !"); - INFOS("CORBA::SystemException : " << e); - return ret; - } - catch(CORBA::Exception& e) - { - INFOS("[StartParallelContainer] Exception in _narrow after LaunchParallelContainer for proxy !"); - INFOS("CORBA::Exception" << e); - return ret; - } - catch(...) - { - INFOS("[StartParallelContainer] Exception in _narrow after LaunchParallelContainer for proxy !"); - INFOS("Unknown exception !"); - return ret; - } - if (CORBA::is_nil(container_proxy)) - { - INFOS("[StartParallelContainer] PaCO::InterfaceManager::_narrow returns NIL !"); - return ret; - } - - // Step 5 : starting parallel container nodes - std::string command_nodes; - Engines::MachineParameters params_nodes(params); - SALOME_ContainerManager::actual_launch_machine_t nodes_machines; - try - { - command_nodes = BuildCommandToLaunchParallelContainer("SALOME_ParallelContainerNode", params_nodes, nodes_machines, proxy_machine[0]); - } - catch(const SALOME_Exception & ex) - { - INFOS("[StartParallelContainer] Exception in BuildCommandToLaunchParallelContainer"); - INFOS(ex.what()); - return ret; - } - std::string container_generic_node_name = _NS->ContainerName(params) + "Node"; - obj = LaunchParallelContainer(command_nodes, params_nodes, container_generic_node_name, nodes_machines); - if (CORBA::is_nil(obj)) - { - INFOS("[StartParallelContainer] LaunchParallelContainer for nodes returns NIL !"); - // Il faut tuer le proxy - try - { - Engines::Container_var proxy = Engines::Container::_narrow(container_proxy); - proxy->Shutdown(); - } - catch (...) - { - INFOS("[StartParallelContainer] Exception catched from proxy Shutdown..."); - } - return ret; + command += " WORKINGDIR "; + command += " '"; + if(wdir == "$TEMPDIR") + wdir="\\$TEMPDIR"; + command += wdir; // requested working directory + command += "'"; } - // Step 6 : connecting nodes and the proxy to actually create a parallel container - for (int i = 0; i < params.nb_component_nodes; i++) + if(params.isMPI) { - 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("[StartParallelContainer] CONNECTION FAILED From Naming Service !"); - INFOS("[StartParallelContainer] Container name is " << containerNameInNS); - return ret; - } - try - { - MESSAGE("[StartParallelContainer] Deploying node : " << container_node_name); - PaCO::InterfaceParallel_var node = PaCO::InterfaceParallel::_narrow(obj); - node->deploy(); - MESSAGE("[StartParallelContainer] node " << container_node_name << " is deployed"); - } - catch(CORBA::SystemException& e) - { - INFOS("[StartParallelContainer] Exception in deploying node : " << containerNameInNS); - INFOS("CORBA::SystemException : " << e); - return ret; - } - catch(CORBA::Exception& e) - { - INFOS("[StartParallelContainer] Exception in deploying node : " << containerNameInNS); - INFOS("CORBA::Exception" << e); - return ret; - } - catch(...) - { - INFOS("[StartParallelContainer] Exception in deploying node : " << containerNameInNS); - INFOS("Unknown exception !"); - return ret; + command += " mpirun -np "; + std::ostringstream o; + o << nbproc << " "; + command += o.str(); +#ifdef WITHLAM + 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+ " "; - // Step 7 : starting parallel container - try - { - MESSAGE ("[StartParallelContainer] Starting parallel object"); - container_proxy->start(); - MESSAGE ("[StartParallelContainer] Parallel object is started"); - ret = Engines::Container::_narrow(container_proxy); - } - 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 - "<ContainerName(params); + command += " -"; + AddOmninamesParams(command); - INFOS("[LaunchParallelContainer] Begin"); - int status = system(command.c_str()); - if (status == -1) { - INFOS("[LaunchParallelContainer] failed : system command status -1"); - return obj; - } - else if (status == 217) { - INFOS("[LaunchParallelContainer] failed : system command status 217"); - return obj; + MESSAGE("command =" << command); } - if (params.nb_component_nodes == 0) - { - std::string theMachine(vect_machine[0]); - // Proxy We have launch a proxy - containerNameInNS = _NS->BuildContainerNameForNS((char*) name.c_str(), theMachine.c_str()); - INFOS("[LaunchParallelContainer] Waiting for Parallel Container proxy " << containerNameInNS << " 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"); - // We are waiting all the nodes - for (int i = 0; i < params.nb_component_nodes; i++) - { - 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; - containerNameInNS = _NS->BuildContainerNameForNS((char*) container_node_name.c_str(), theMachine.c_str()); - INFOS("[LaunchParallelContainer] Waiting for Parallel Container node " << containerNameInNS << " on " << theMachine); - while (CORBA::is_nil(obj) && count) { -#ifndef WIN32 - sleep(1) ; -#else - Sleep(1000); -#endif - count-- ; - obj = _NS->Resolve(containerNameInNS.c_str()); - } - if (CORBA::is_nil(obj)) - { - INFOS("[LaunchParallelContainer] Launch of node failed (or not found) !"); - return obj; - } - } - } - if (CORBA::is_nil(obj)) - INFOS("[LaunchParallelContainer] failed"); - - return obj; -} - -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(); -} - -//============================================================================= -/*! - * This is no longer valid (C++ container are also python containers) - */ -//============================================================================= - -bool isPythonContainer(const char* ContainerName) -{ - bool ret = false; - int len = strlen(ContainerName); - - if (len >= 2) - if (strcmp(ContainerName + len - 2, "Py") == 0) - ret = true; - - return ret; -} - -//============================================================================= -/*! - * Builds the script to be launched - * - * If SALOME Application not defined ($APPLI), - * see BuildTempFileToLaunchRemoteContainer() - * - * Else rely on distant configuration. Command is under the form (example): - * ssh user@machine distantPath/runRemote.sh hostNS portNS WORKINGDIR workingdir \ - * SALOME_Container containerName &" - - * - where user is ommited if not specified in CatalogResources, - * - where distant path is always relative to user@machine $HOME, and - * equal to $APPLI if not specified in CatalogResources, - * - where hostNS is the hostname of CORBA naming server (set by scripts to - * use to launch SALOME and servers in $APPLI: runAppli.sh, runRemote.sh) - * - where portNS is the port used by CORBA naming server (set by scripts to - * use to launch SALOME and servers in $APPLI: runAppli.sh, runRemote.sh) - * - where workingdir is the requested working directory for the container. - * If WORKINGDIR (and workingdir) is not present the working dir will be $HOME - */ -//============================================================================= - -string -SALOME_ContainerManager::BuildCommandToLaunchRemoteContainer -(const string& machine, - const Engines::MachineParameters& params, const std::string& container_exe) -{ - string command; - int nbproc; - - if ( ! _isAppliSalomeDefined ) - command = BuildTempFileToLaunchRemoteContainer(machine, 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 &" - - if (resInfo.Protocol == rsh) - command = "rsh "; - else if (resInfo.Protocol == ssh) - command = "ssh "; - else - throw SALOME_Exception("Unknown protocol"); - - if (resInfo.UserName != "") - { - command += resInfo.UserName; - command += "@"; - } - - command += machine; - 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 $HOME - } - - command += "/runRemote.sh "; - - ASSERT(getenv("NSHOST")); - command += getenv("NSHOST"); // hostname of CORBA name server - - command += " "; - ASSERT(getenv("NSPORT")); - command += getenv("NSPORT"); // port of CORBA name server - - std::string wdir=params.workingdir.in(); - if(wdir != "") - { - command += " WORKINGDIR "; - command += " '"; - if(wdir == "$TEMPDIR") - wdir="\\$TEMPDIR"; - command += wdir; // requested working directory - command += "'"; - } - - if(params.isMPI) - { - command += " mpirun -np "; - std::ostringstream o; - o << nbproc << " "; - command += o.str(); -#ifdef WITHLAM - 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 += " -"; - AddOmninamesParams(command); - - MESSAGE("command =" << command); - } - return command; } @@ -957,62 +703,50 @@ SALOME_ContainerManager::BuildCommandToLaunchRemoteContainer * builds the command to be launched. */ //============================================================================= - -string +std::string SALOME_ContainerManager::BuildCommandToLaunchLocalContainer -(const Engines::MachineParameters& params, 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; - 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) - //command += "-x PATH -x LD_LIBRARY_PATH -x OMNIORB_CONFIG -x SALOME_trace "; if( getenv("OMPI_URI_FILE") == NULL ) - command_file << "-x PATH -x LD_LIBRARY_PATH -x OMNIORB_CONFIG -x SALOME_trace"; + o << "-x PATH -x LD_LIBRARY_PATH -x OMNIORB_CONFIG -x SALOME_trace"; else { - command_file << "-x PATH -x LD_LIBRARY_PATH -x OMNIORB_CONFIG -x SALOME_trace -ompi-server file:"; - command_file << getenv("OMPI_URI_FILE"); + 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 != "") { @@ -1020,13 +754,11 @@ 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 } @@ -1034,27 +766,26 @@ 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 + " "; } - command_file << _NS->ContainerName(params); - command_file << " -"; - AddOmninamesParams(command_file); + o << _NS->ContainerName(params); + o << " -"; + AddOmninamesParams(o); + + std::ofstream command_file( _TmpFileName.c_str() ); + command_file << o.str(); command_file.close(); #ifndef WIN32 @@ -1063,6 +794,7 @@ SALOME_ContainerManager::BuildCommandToLaunchLocalContainer command = _TmpFileName; MESSAGE("Command is file ... " << command); + MESSAGE("Command is ... " << o.str()); return command; } @@ -1079,9 +811,9 @@ void SALOME_ContainerManager::RmTmpFile(std::string& tmpFileName) 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 ); @@ -1090,7 +822,7 @@ void SALOME_ContainerManager::RmTmpFile(std::string& 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 @@ -1109,37 +841,49 @@ void SALOME_ContainerManager::RmTmpFile(std::string& tmpFileName) */ //============================================================================= -void SALOME_ContainerManager::AddOmninamesParams(string& command) const +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(ofstream& fileStream) const +void SALOME_ContainerManager::AddOmninamesParams(std::ofstream& fileStream) const { CORBA::String_var iorstr = _NS->getIORaddr(); fileStream << "ORBInitRef NameService="; fileStream << iorstr; } +//============================================================================= +/*! + * add to command all options relative to naming service. + */ +//============================================================================= + +void SALOME_ContainerManager::AddOmninamesParams(std::ostringstream& oss) const +{ + CORBA::String_var iorstr = _NS->getIORaddr(); + oss << "ORBInitRef NameService="; + oss << iorstr; +} + //============================================================================= /*! * generate a file name in /tmp directory */ //============================================================================= -string SALOME_ContainerManager::BuildTemporaryFileName() const +std::string SALOME_ContainerManager::BuildTemporaryFileName() const { //build more complex file name to support multiple salome session - string aFileName = Kernel_Utils::GetTmpFileName(); + std::string aFileName = Kernel_Utils::GetTmpFileName(); #ifndef WIN32 aFileName += ".sh"; #else @@ -1148,7 +892,6 @@ string SALOME_ContainerManager::BuildTemporaryFileName() const return aFileName; } - //============================================================================= /*! * Builds in a temporary file the script to be launched. @@ -1159,22 +902,22 @@ 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 @@ -1184,14 +927,10 @@ 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; @@ -1200,10 +939,10 @@ SALOME_ContainerManager::BuildTempFileToLaunchRemoteContainer 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"; + 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"); + tempOutputFile << "-x PATH -x LD_LIBRARY_PATH -x OMNIORB_CONFIG -x SALOME_trace -ompi-server file:"; + tempOutputFile << getenv("OMPI_URI_FILE"); } #endif } @@ -1228,7 +967,7 @@ SALOME_ContainerManager::BuildTempFileToLaunchRemoteContainer tempOutputFile << _NS->ContainerName(params) << " -"; AddOmninamesParams(tempOutputFile); - tempOutputFile << " &" << endl; + tempOutputFile << " &" << std::endl; tempOutputFile.flush(); tempOutputFile.close(); #ifndef WIN32 @@ -1237,15 +976,15 @@ 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()); @@ -1254,10 +993,10 @@ 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()); @@ -1268,7 +1007,7 @@ SALOME_ContainerManager::BuildTempFileToLaunchRemoteContainer if(status) throw SALOME_Exception("Error of connection on remote host"); - command += machine; + command += resInfo.HostName; _CommandForRemAccess = command; command += " "; command += _TmpFileName; @@ -1279,395 +1018,790 @@ 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; + +} + +std::set SALOME_ContainerManager::getpidofprogram(const std::string program) +{ + std::set thepids; + std::string tmpFile = Kernel_Utils::GetTmpFileName(); + std::string cmd; + std::string thepid; + cmd = "pidof " + program + " > " + tmpFile; + system(cmd.c_str()); + std::ifstream fpi(tmpFile.c_str(),std::ios::in); + while(fpi >> thepid){ + thepids.insert(atoi(thepid.c_str())); + } + return thepids; +} + +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::BuildCommandToLaunchParallelContainer(const std::string& exe_name, - const Engines::MachineParameters& params, - SALOME_ContainerManager::actual_launch_machine_t & vect_machine, - const std::string proxy_hostname) +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. - - //command = "gdb --args "; - //command = "valgrind --tool=memcheck --log-file=val_log "; - //command += real_exe_name; - - // Step 0 : init some variables... - std::string parallelLib(CORBA::string_dup(params.parallelLib)); - std::string real_exe_name = exe_name + parallelLib; - std::string machine_file_name(""); - bool remote = false; - bool is_a_proxy = false; - std::string hostname(CORBA::string_dup(params.hostname)); + 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()); - std::ostringstream tmp_string; - CORBA::Long nb_nodes = params.nb_component_nodes; - tmp_string << nb_nodes; - std::string nbproc = tmp_string.str(); - - 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; - - // Step 1 : local or remote launch ? - if (hostname != std::string(Kernel_Utils::GetHostname()) ) + // Step 0 : Check parameters + if (!checkPaCOParameters(params, resource_selected)) + { + INFOS("[StartPaCOPPContainer] check parameters failed ! see logs..."); + return ret; + } + + // Step 1 : Starting a new parallel container ! + INFOS("[StartPaCOPPContainer] Starting a PaCO++ parallel container"); + INFOS("[StartPaCOPPContainer] on resource : " << resource_selected); + + // 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) { - MESSAGE("[BuildCommandToLaunchParallelContainer] remote machine case detected !"); - remote = true; + INFOS("[StarPaCOPPContainer] Exception in BuildCommandToLaunchPaCONodeContainer"); + INFOS(ex.what()); + return ret; } - // Step 2 : proxy or nodes launch ? - std::string::size_type loc_proxy = exe_name.find("Proxy"); - if( loc_proxy != string::npos ) { - is_a_proxy = true; - } + 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) + { + INFOS("[StarPaCOPPContainer] LaunchPaCONodeContainer failed !"); + // Il faut tuer le proxy + try + { + Engines::Container_var proxy = Engines::Container::_narrow(container_proxy); + proxy->Shutdown(); + } + catch (...) + { + INFOS("[StarPaCOPPContainer] Exception catched from proxy Shutdown..."); + } + return ret; + } - // Step 3 : Depending of the parallelLib, getting the machine file - // ParallelLib Dummy has is own machine for this method - if (remote) + // Step 4 : connecting nodes and the proxy to actually create a parallel container + for (int i = 0; i < params.nb_proc; i++) { - if (is_a_proxy) + 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)) { - machine_file_name = _ResManager->getMachineFile(hostname, - 1, - parallelLib); + INFOS("[StarPaCOPPContainer] CONNECTION FAILED From Naming Service !"); + INFOS("[StarPaCOPPContainer] Container name is " << containerNameInNS); + return ret; } - else + 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) + { + INFOS("[StarPaCOPPContainer] Exception in deploying node : " << containerNameInNS); + INFOS("CORBA::SystemException : " << e); + return ret; + } + catch(CORBA::Exception& e) { - machine_file_name = _ResManager->getMachineFile(hostname, - params.nb_component_nodes, - parallelLib); + INFOS("[StarPaCOPPContainer] Exception in deploying node : " << containerNameInNS); + INFOS("CORBA::Exception" << e); + return ret; } - if (machine_file_name == "") + catch(...) { - INFOS("[BuildCommandToLaunchParallelContainer] Error machine_file was not generated for machine " << hostname); - throw SALOME_Exception("Error machine_file was not generated"); + INFOS("[StarPaCOPPContainer] Exception in deploying node : " << containerNameInNS); + INFOS("Unknown exception !"); + return ret; } - MESSAGE("[BuildCommandToLaunchParallelContainer] machine_file_name is : " << machine_file_name); } - // Step 4 : Log type choosen by the user - std::string log_env(""); - char * get_val = getenv("PARALLEL_LOG"); - if (get_val) - log_env = get_val; - std::string command_begin(""); - std::string command_end(""); - if(log_env == "xterm") + // Step 5 : starting parallel container + try { - command_begin = "/usr/X11R6/bin/xterm -e \"export LD_LIBRARY_PATH=$LD_LIBRARY_PATH; export PATH=$PATH;"; - command_end = "\"&"; + MESSAGE ("[StarPaCOPPContainer] Starting parallel object"); + container_proxy->start(); + MESSAGE ("[StarPaCOPPContainer] Parallel object is started"); + ret = Engines::Container::_narrow(container_proxy); } - else if(log_env == "xterm_debug") + catch(CORBA::SystemException& e) { - command_begin = "/usr/X11R6/bin/xterm -e \"export LD_LIBRARY_PATH=$LD_LIBRARY_PATH; export PATH=$PATH;"; - command_end = "; cat \" &"; + INFOS("Caught CORBA::SystemException. : " << e); } - else + catch(PortableServer::POA::ServantAlreadyActive&) { - // default into a file... - std::string logFilename = "/tmp/" + _NS->ContainerName(params) + "_" + hostname; - if (is_a_proxy) - logFilename += "_Proxy_"; - else - logFilename += "_Node_"; - logFilename += std::string(getenv("USER")) + ".log"; - command_end = " > " + logFilename + " 2>&1 & "; + 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())) { - if (is_a_proxy) + MESSAGE("[BuildCommandToLaunchPaCOProxyContainer] remote machine case detected !"); + remote_execution = true; + } + + // 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) + { + 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 + } + + 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++) { - std::string command_remote(""); - if (remote) + // 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())) { - std::string machine_name; - std::ifstream machine_file(machine_file_name.c_str()); - std::getline(machine_file, machine_name); - MESSAGE("[BuildCommandToLaunchParallelContainer] machine file name extracted is " << machine_name) - - // We want to launch a command like : - // ssh user@machine distantPath/runRemote.sh hostNS portNS - const ParserResourcesType& resInfo = _ResManager->GetImpl()->GetResourcesList(machine_name); - if (resInfo.Protocol == rsh) - command_remote = "rsh "; - else - command_remote = "ssh "; - command_remote += resInfo.UserName; - command_remote += "@"; - command_remote += machine_name; - command_remote += " "; - command_remote += resInfo.AppliPath; // path relative to user@machine $HOME - command_remote += "/runRemote.sh "; - ASSERT(getenv("NSHOST")); - command_remote += getenv("NSHOST"); // hostname of CORBA name server - command_remote += " "; - ASSERT(getenv("NSPORT")); - command_remote += getenv("NSPORT"); // port of CORBA name server - command_remote += " "; - - hostname = machine_name; + MESSAGE("[BuildCommandToLaunchPaCONodeContainer] remote machine case detected !"); + remote_execution = true; } - command = real_exe_name; - command += " " + _NS->ContainerName(rtn); - command += " " + parallelLib; - command += " " + hostname; - command += " " + nbproc; - command += " -"; - AddOmninamesParams(command); - - command = command_begin + command_remote + command + command_end; - vect_machine.push_back(hostname); - } - else - { - std::ifstream * machine_file = NULL; - if (remote) - machine_file = new std::ifstream(machine_file_name.c_str()); - for (int i= 0; i < nb_nodes; i++) + // 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) { - std::string command_remote(""); - if (remote) - { - std::string machine_name; - std::getline(*machine_file, machine_name); - MESSAGE("[BuildCommandToLaunchParallelContainer] machine file name extracted is " << machine_name) - - // We want to launch a command like : - // ssh user@machine distantPath/runRemote.sh hostNS portNS - const ParserResourcesType& resInfo = _ResManager->GetImpl()->GetResourcesList(machine_name); - if (resInfo.Protocol == rsh) - command_remote = "rsh "; - else - command_remote = "ssh "; - command_remote += resInfo.UserName; - command_remote += "@"; - command_remote += machine_name; - command_remote += " "; - command_remote += resInfo.AppliPath; // path relative to user@machine $HOME - command_remote += "/runRemote.sh "; - ASSERT(getenv("NSHOST")); - command_remote += getenv("NSHOST"); // hostname of CORBA name server - command_remote += " "; - ASSERT(getenv("NSPORT")); - command_remote += getenv("NSPORT"); // port of CORBA name server - command_remote += " "; - - hostname = machine_name; - } - - std::ostringstream tmp; - tmp << i; - std::string proc_number = tmp.str(); - - std::string command_tmp(""); - command_tmp += real_exe_name; - command_tmp += " " + _NS->ContainerName(rtn); - command_tmp += " " + parallelLib; - command_tmp += " " + proxy_hostname; - command_tmp += " " + proc_number; - command_tmp += " -"; - AddOmninamesParams(command_tmp); - - // On change _Node_ par _Nodex_ pour avoir chaque noeud - // sur un fichier - std::string command_end_tmp = command_end; - std::string::size_type loc_node = command_end_tmp.find("_Node_"); - if (loc_node != std::string::npos) - command_end_tmp.insert(loc_node+5, proc_number); - command += command_begin + command_remote + command_tmp + command_end_tmp; - vect_machine.push_back(hostname); + 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 } - if (machine_file) - delete machine_file; + + 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 (parallelLib == "Mpi") + + else if (std::string(params.parallelLib.in()) == "Mpi") { - // Step 0: if remote we have to copy the file - // to the first machine of the file - std::string remote_machine(""); - if (remote) + // 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())) { - std::ifstream * machine_file = NULL; - machine_file = new std::ifstream(machine_file_name.c_str()); - // Get first word of the line - // For MPI implementation the first word is the - // machine name - std::getline(*machine_file, remote_machine, ' '); - machine_file->close(); - MESSAGE("[BuildCommandToLaunchParallelContainer] machine file name extracted is " << remote_machine) - - // We want to launch a command like : - // scp mpi_machine_file user@machine:Path - std::string command_remote(""); - const ParserResourcesType& resInfo = _ResManager->GetImpl()->GetResourcesList(remote_machine); - if (resInfo.Protocol == rsh) - command_remote = "rcp "; - else - command_remote = "scp "; - - command_remote += machine_file_name; - command_remote += " "; - command_remote += resInfo.UserName; - command_remote += "@"; - command_remote += remote_machine; - command_remote += ":"; - command_remote += machine_file_name; - - int status = system(command_remote.c_str()); - if (status == -1) - { - INFOS("copy of the mpi machine file failed !"); - return ""; - } + MESSAGE("[BuildCommandToLaunchPaCONodeContainer] remote machine case detected !"); + remote_execution = true; } - if (is_a_proxy) + // 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) { - std::string command_remote(""); - if (remote) - { - // We want to launch a command like : - // ssh user@machine distantPath/runRemote.sh hostNS portNS - const ParserResourcesType& resInfo = _ResManager->GetImpl()->GetResourcesList(remote_machine); - if (resInfo.Protocol == rsh) - command_remote = "rsh "; - else - command_remote = "ssh "; - command_remote += resInfo.UserName; - command_remote += "@"; - command_remote += remote_machine; - command_remote += " "; - command_remote += resInfo.AppliPath; // path relative to user@machine $HOME - command_remote += "/runRemote.sh "; - ASSERT(getenv("NSHOST")); - command_remote += getenv("NSHOST"); // hostname of CORBA name server - command_remote += " "; - ASSERT(getenv("NSPORT")); - command_remote += getenv("NSPORT"); // port of CORBA name server - command_remote += " "; - - hostname = remote_machine; - } - - // We use Dummy proxy for MPI parallel containers - real_exe_name = exe_name + "Dummy"; - command = real_exe_name; - command += " " + _NS->ContainerName(rtn); - command += " Dummy"; - command += " " + hostname; - command += " " + nbproc; - command += " -"; - AddOmninamesParams(command); - - command = command_begin + command_remote + command + command_end; - vect_machine.push_back(hostname); + INFOS("copy of the MPI machine file failed ! - sorry !"); + return ""; } - else + + // 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) { - std::string command_remote(""); - if (remote) - { - const ParserResourcesType& resInfo = _ResManager->GetImpl()->GetResourcesList(remote_machine); - if (resInfo.Protocol == rsh) - command_remote = "rsh "; - else - command_remote = "ssh "; - command_remote += resInfo.UserName; - command_remote += "@"; - command_remote += remote_machine; - command_remote += " "; - - std::string new_real_exe_name(""); - new_real_exe_name += resInfo.AppliPath; // path relative to user@machine $HOME - new_real_exe_name += "/runRemote.sh "; - ASSERT(getenv("NSHOST")); - new_real_exe_name += getenv("NSHOST"); // hostname of CORBA name server - new_real_exe_name += " "; - ASSERT(getenv("NSPORT")); - new_real_exe_name += getenv("NSPORT"); // port of CORBA name server - new_real_exe_name += " "; - - real_exe_name = new_real_exe_name + real_exe_name; - hostname = remote_machine; - } + ASSERT(getenv("NSHOST")); + ASSERT(getenv("NSPORT")); - const ParserResourcesType& resInfo = _ResManager->GetImpl()->GetResourcesList(hostname); - if (resInfo.mpi == lam) - { - command = "mpiexec -ssi boot "; - if (resInfo.Protocol == rsh) - command += "rsh "; - else - command += "ssh "; - command += "-machinefile " + machine_file_name + " "; - command += "-n " + nbproc + " "; - command += real_exe_name; - command += " " + _NS->ContainerName(rtn); - command += " " + parallelLib; - command += " " + proxy_hostname; - command += " -"; - AddOmninamesParams(command); - } - else - { - command = "mpirun -np " + nbproc + " "; - command += real_exe_name; - command += " " + _NS->ContainerName(rtn); - command += " " + parallelLib; - command += " " + proxy_hostname; - command += " -"; - AddOmninamesParams(command); - } + 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 + } - command = command_begin + command_remote + command + command_end; - for (int i= 0; i < nb_nodes; i++) - vect_machine.push_back(proxy_hostname); + 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; + } + return command_nodes.str(); +} + +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) +{ + if(log_type == "xterm") + { + begin = "xterm -e \"export LD_LIBRARY_PATH=$LD_LIBRARY_PATH; export PATH=$PATH;"; + end = "\"&"; + } + else if(log_type == "xterm_debug") + { + begin = "xterm -e \"export LD_LIBRARY_PATH=$LD_LIBRARY_PATH; export PATH=$PATH;"; + end = "; cat \" &"; } else { - std::string message("Unknown parallelLib : " + parallelLib); - throw SALOME_Exception(message.c_str()); + // default into a file... + std::string logFilename = "/tmp/" + container_name + "_" + hostname + "_" + exe_type + "_"; + logFilename += std::string(getenv("USER")) + ".log"; + end = " > " + logFilename + " 2>&1 & "; } - - MESSAGE("Parallel launch is: " << command); - return command; } -string SALOME_ContainerManager::GetMPIZeroNode(string machine) +CORBA::Object_ptr +SALOME_ContainerManager::LaunchPaCOProxyContainer(const std::string& command, + const Engines::ContainerParameters& params, + const std::string & hostname) { - int status; - string zeronode; - string cmd; - string tmpFile = BuildTemporaryFileName(); + PaCO::InterfaceManager_ptr container_proxy = PaCO::InterfaceManager::_nil(); - cmd = "ssh " + machine + " mpirun -np 1 hostname > " + tmpFile; + 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; + } - status = system(cmd.c_str()); - if( status == 0 ){ - ifstream fp(tmpFile.c_str(),ios::in); - fp >> zeronode; + 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()); } - RmTmpFile(tmpFile); + 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(); +} - return zeronode; +//============================================================================= +/*! 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 +