-// Copyright (C) 2007-2017 CEA/DEN, EDF R&D, OPEN CASCADE
+// Copyright (C) 2007-2023 CEA, EDF, OPEN CASCADE
//
// Copyright (C) 2003-2007 OPEN CASCADE, EADS/CCR, LIP6, CEA/DEN,
// CEDRAT, EDF R&D, LEG, PRINCIPIA R&D, BUREAU VERITAS
#include "SALOME_LoadRateManager.hxx"
#include "SALOME_NamingService.hxx"
#include "SALOME_ResourcesManager_Client.hxx"
+#include "SALOME_Embedded_NamingService.hxx"
#include "SALOME_ModuleCatalog.hh"
#include "Basics_Utils.hxx"
#include "Basics_DirUtils.hxx"
+#include "PythonCppUtils.hxx"
#include <sys/types.h>
#include <sys/stat.h>
#include <signal.h>
#include "Utils_CorbaException.hxx"
#include <sstream>
#include <string>
+#include <queue>
#include <SALOMEconfig.h>
#include CORBA_CLIENT_HEADER(SALOME_Session)
#ifdef HAVE_MPI2
#include <mpi.h>
+#include <sys/wait.h>
#endif
#ifdef WIN32
#include <process.h>
#define getpid _getpid
+
+#ifndef S_ISREG
+#define S_ISREG(mode) (((mode) & S_IFMT) == S_IFREG)
+#endif
+
#endif
#ifdef WITH_PACO_PARALLEL
*/
//=============================================================================
-SALOME_ContainerManager::SALOME_ContainerManager(CORBA::ORB_ptr orb, PortableServer::POA_var poa, SALOME_NamingService *ns)
+SALOME_ContainerManager::SALOME_ContainerManager(CORBA::ORB_ptr orb, PortableServer::POA_var poa, SALOME_NamingService_Abstract *ns)
: _nbprocUsed(1)
{
MESSAGE("constructor");
threadPol->destroy();
PortableServer::ObjectId_var id = _poa->activate_object(this);
CORBA::Object_var obj = _poa->id_to_reference(id);
- Engines::ContainerManager_var refContMan =
- Engines::ContainerManager::_narrow(obj);
-
- _NS->Register(refContMan,_ContainerManagerNameInNS);
+ Engines::ContainerManager_var refContMan = Engines::ContainerManager::_narrow(obj);
+ if(_NS)
+ _NS->Register(refContMan,_ContainerManagerNameInNS);
_isAppliSalomeDefined = (GetenvThreadSafe("APPLI") != 0);
#ifdef HAVE_MPI2
urifile << GetenvThreadSafeAsString("HOME") << "/.urifile_" << getpid();
setenv("OMPI_URI_FILE",urifile.str().c_str(),1);
if( GetenvThreadSafe("OMPI_URI_FILE") != NULL ){
- // get the pid of all ompi-server
- std::set<pid_t> thepids1 = getpidofprogram("ompi-server");
- // launch a new ompi-server
- std::string command;
- command = "ompi-server -r ";
- command += GetenvThreadSafeAsString("OMPI_URI_FILE");
- int status=SystemThreadSafe(command.c_str());
- if(status!=0)
- throw SALOME_Exception("Error when launching ompi-server");
- // get the pid of all ompi-server
- std::set<pid_t> thepids2 = getpidofprogram("ompi-server");
- // my ompi-server is the new one
- std::set<pid_t>::const_iterator it;
- for(it=thepids2.begin();it!=thepids2.end();it++)
- if(thepids1.find(*it) == thepids1.end())
- _pid_mpiServer = *it;
- if(_pid_mpiServer < 0)
- throw SALOME_Exception("Error when getting ompi-server id");
+ // Linux specific code
+ pid_t pid = fork(); // spawn a child process, following code is executed in both processes
+ if ( pid == 0 ) // I'm a child, replace myself with a new ompi-server
+ {
+ std::string uriarg = GetenvThreadSafeAsString("OMPI_URI_FILE");
+ execlp( "ompi-server", "ompi-server", "-r", uriarg.c_str(), NULL );
+ throw SALOME_Exception("Error when launching ompi-server"); // execlp failed
+ }
+ else if ( pid < 0 )
+ {
+ throw SALOME_Exception("fork() failed");
+ }
+ else // I'm a parent
+ {
+ //wait(NULL); // wait(?) for a child end
+ _pid_mpiServer = pid;
+ }
}
#elif defined(MPICH)
_pid_mpiServer = -1;
- // get the pid of all hydra_nameserver
- std::set<pid_t> thepids1 = getpidofprogram("hydra_nameserver");
- // launch a new hydra_nameserver
- std::string command;
- command = "hydra_nameserver &";
- SystemThreadSafe(command.c_str());
- // get the pid of all hydra_nameserver
- std::set<pid_t> thepids2 = getpidofprogram("hydra_nameserver");
- // my hydra_nameserver is the new one
- std::set<pid_t>::const_iterator it;
- for(it=thepids2.begin();it!=thepids2.end();it++)
- if(thepids1.find(*it) == thepids1.end())
- _pid_mpiServer = *it;
+ // Linux specific code
+ pid_t pid = fork(); // spawn a child process, following code is executed in both processes
+ if ( pid == 0 ) // I'm a child, replace myself with a new hydra_nameserver
+ {
+ execlp( "hydra_nameserver", "hydra_nameserver", NULL );
+ throw SALOME_Exception("Error when launching hydra_nameserver"); // execlp failed
+ }
+ else if ( pid < 0 )
+ {
+ throw SALOME_Exception("fork() failed");
+ }
+ else // I'm a parent
+ {
+ //wait(NULL);
+ _pid_mpiServer = pid;
+ }
#endif
#endif
{
MESSAGE("Shutdown");
ShutdownContainers();
- _NS->Destroy_Name(_ContainerManagerNameInNS);
+ if(_NS)
+ _NS->Destroy_Name(_ContainerManagerNameInNS);
PortableServer::ObjectId_var oid = _poa->servant_to_id(this);
_poa->deactivate_object(oid);
}
void SALOME_ContainerManager::ShutdownContainers()
{
MESSAGE("ShutdownContainers");
-
+ if(!_NS)
+ return ;
SALOME::Session_var session = SALOME::Session::_nil();
CORBA::Long pid = 0;
CORBA::Object_var objS = _NS->Resolve("/Kernel/Session");
if(!CORBA::is_nil(cont) && pid != cont->getPID())
lstCont.push_back((*iter));
}
- catch(const CORBA::Exception& e)
+ catch(const CORBA::Exception&)
{
// ignore this entry and continue
}
}
}
+void SALOME_ContainerManager::SetOverrideEnvForContainers(const Engines::KeyValDict& env)
+{
+ this->_override_env.clear();
+ auto sz = env.length();
+ for(auto i = 0 ; i < sz ; ++i)
+ _override_env.emplace_back(env[i].key.in(), env[i].val.in());
+}
+
+Engines::KeyValDict *SALOME_ContainerManager::GetOverrideEnvForContainers()
+{
+ std::unique_ptr<Engines::KeyValDict> ret( new Engines::KeyValDict );
+ auto sz = _override_env.size();
+ ret->length(sz);
+ for(auto i = 0 ; i < sz ; ++i)
+ {
+ (*ret)[i].key = CORBA::string_dup( _override_env[i].first.c_str() );
+ (*ret)[i].val = CORBA::string_dup( _override_env[i].second.c_str() );
+ }
+ return ret.release();
+}
+
//=============================================================================
//! Give a suitable Container given constraints
/*! CORBA Method:
break;
}
}
- catch(const SALOME_Exception &ex)
+ catch(const SALOME_Exception &ex) //!< TODO: unused variable
{
MESSAGE("[GiveContainer] Exception in ResourceManager find !: " << ex.what());
return ret;
std::string hostname(resource_definition.HostName);
std::string containerNameInNS;
if(params.isMPI){
- int nbproc;
- if ( params.nb_proc <= 0 )
- nbproc = 1;
- else
- nbproc = params.nb_proc;
+ int nbproc = params.nb_proc <= 0 ? 1 : params.nb_proc;
try
{
if( GetenvThreadSafe("LIBBATCH_NODEFILE") != NULL )
if (!CORBA::is_nil(cont))
{
INFOS("[GiveContainer] container " << containerNameInNS << " launched");
+ std::ostringstream envInfo;
+ std::for_each( _override_env.begin(), _override_env.end(), [&envInfo](const std::pair<std::string,std::string>& p) { envInfo << p.first << " = " << p.second << std::endl; } );
+ INFOS("[GiveContainer] container " << containerNameInNS << " override " << envInfo.str());
+ Engines::FieldsDict envCorba;
+ {
+ auto sz = _override_env.size();
+ envCorba.length(sz);
+ for(auto i = 0 ; i < sz ; ++i)
+ {
+ envCorba[i].key = CORBA::string_dup( _override_env[i].first.c_str() );
+ envCorba[i].value <<= CORBA::string_dup( _override_env[i].second.c_str() );
+ }
+ }
+ cont->override_environment_python( envCorba );
return cont._retn();
}
else
return ret;
}
+std::string SALOME_ContainerManager::GetCppBinaryOfKernelContainer() const
+{
+ std::string ret = this->_isSSL ? "SALOME_Container_No_NS_Serv" : "SALOME_Container";
+ return ret;
+}
+
+std::string SALOME_ContainerManager::GetRunRemoteExecutableScript() const
+{
+ std::string ret = this->_isSSL ? "runRemoteSSL.sh" : "runRemote.sh";
+ return ret;
+}
+
Engines::Container_ptr
SALOME_ContainerManager::LaunchContainer(const Engines::ContainerParameters& params,
const std::string & resource_selected,
// 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
+ std::string container_exe = this->GetCppBinaryOfKernelContainer();
Engines::ContainerParameters local_params(params);
int found=0;
try
// Only if an application directory is set
if(hostname != Kernel_Utils::GetHostname() && _isAppliSalomeDefined)
{
- // Preparing remote command
- std::string command = "";
+
const ParserResourcesType resInfo(_resManager->GetResourceDefinition(resource_selected));
- command = getCommandToRunRemoteProcess(resInfo.Protocol, resInfo.HostName, resInfo.UserName);
- if (resInfo.AppliPath != "")
- command += resInfo.AppliPath;
- else
- {
- ASSERT(GetenvThreadSafe("APPLI"));
- command += GetenvThreadSafeAsString("APPLI");
- }
- command += "/runRemote.sh ";
- ASSERT(GetenvThreadSafe("NSHOST"));
- command += GetenvThreadSafeAsString("NSHOST"); // hostname of CORBA name server
- command += " ";
- ASSERT(GetenvThreadSafe("NSPORT"));
- command += GetenvThreadSafeAsString("NSPORT"); // port of CORBA name server
- command += " \"ls /tmp >/dev/null 2>&1\"";
+ std::string command = getCommandToRunRemoteProcess(resInfo.Protocol, resInfo.HostName,
+ resInfo.UserName, resInfo.AppliPath);
// Launch remote command
- int status = SystemThreadSafe(command.c_str());
+ command += " \"ls /tmp >/dev/null 2>&1\"";
+ // Anthony : command is NO MORE launched to improve dramatically time to launch containers
+ int status = 0;
if (status != 0)
{
// Error on resource - cannot launch commands
MESSAGE("[GiveContainer] Try to launch a new container on " << resource_selected);
// if a parallel container is launched in batch job, command is: "mpirun -np nbproc -machinefile nodesfile SALOME_MPIContainer"
if( GetenvThreadSafe("LIBBATCH_NODEFILE") != NULL && params.isMPI )
+ {
command = BuildCommandToLaunchLocalContainer(params, machFile, container_exe, tmpFileName);
+ MESSAGE("[LaunchContainer] LIBBATCH_NODEFILE : \"" << command << "\"");
+ }
// 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, tmpFileName);
+ MESSAGE("[LaunchContainer] hostname local : \"" << command << "\"");
+ }
// 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);
+ MESSAGE("[LaunchContainer] remote : \"" << command << "\"");
+ }
//redirect stdout and stderr in a file
#ifdef WIN32
logFilename += ".log" ;
command += " > " + logFilename + " 2>&1";
MakeTheCommandToBeLaunchedASync(command);
-
+
+ MESSAGE("[LaunchContainer] SYSTEM COMMAND that will be launched : \"" << command << "\"");
// launch container with a system call
status=SystemThreadSafe(command.c_str());
}//end of critical of section
else
return Engines::Container::_narrow(obj);
}
- catch(const CORBA::Exception& e)
+ catch(const CORBA::Exception&)
{
return Engines::Container::_nil();
}
{
return false; // VSR 02/08/2013: Python containers are no more supported
bool ret = false;
- int len = strlen(ContainerName);
+ size_t len = strlen(ContainerName);
if (len >= 2)
if (strcmp(ContainerName + len - 2, "Py") == 0)
SALOME_ContainerManager::BuildCommandToLaunchRemoteContainer(const std::string& resource_name, const Engines::ContainerParameters& params, const std::string& container_exe) const
{
std::string command,tmpFileName;
+ const ParserResourcesType resInfo(_resManager->GetResourceDefinition(resource_name));
+ std::string wdir = params.workingdir.in();
if (!_isAppliSalomeDefined)
- command = BuildTempFileToLaunchRemoteContainer(resource_name, params, tmpFileName);
+ {
+ MESSAGE("[BuildCommandToLaunchRemoteContainer] NO APPLI MODE : " << " Protocol :" << resInfo.Protocol << " hostname :" << resInfo.HostName << " username : " << resInfo.UserName << " appli : " << resInfo.AppliPath << " wdir : \"" << wdir << "\"");
+ command = getCommandToRunRemoteProcessNoAppli(resInfo.Protocol, resInfo.HostName,
+ resInfo.UserName, resInfo.AppliPath,
+ wdir);
+ }
else
{
- int nbproc;
- const ParserResourcesType resInfo(_resManager->GetResourceDefinition(resource_name));
-
- if (params.isMPI)
- {
- if ( params.nb_proc <= 0 )
- nbproc = 1;
- else
- nbproc = params.nb_proc;
+ MESSAGE("[BuildCommandToLaunchRemoteContainer] WITH APPLI MODE : " << " Protocol :" << resInfo.Protocol << " hostname :" << resInfo.HostName << " username : " << resInfo.UserName << " appli : " << resInfo.AppliPath << " wdir : \"" << wdir << "\"");
+ // "ssh -l user machine distantPath/runRemote.sh hostNS portNS WORKINGDIR workingdir
+ // SALOME_Container containerName -ORBInitRef NameService=IOR:01000..."
+ // or
+ // "ssh -l user machine distantLauncher remote -p hostNS -m portNS -d dir
+ // -- SALOME_Container contName -ORBInitRef NameService=IOR:01000..."
+ command = getCommandToRunRemoteProcess(resInfo.Protocol, resInfo.HostName,
+ resInfo.UserName, resInfo.AppliPath,
+ wdir);
+ }
+ if(params.isMPI)
+ {
+ int nbproc = params.nb_proc <= 0 ? 1 : params.nb_proc;
+ command += " mpirun -np ";
+ std::ostringstream o;
+ o << nbproc << " ";
+ command += o.str();
+#ifdef LAM_MPI
+ command += "-x PATH,LD_LIBRARY_PATH,OMNIORB_CONFIG,SALOME_trace ";
+#elif defined(OPEN_MPI)
+ if( GetenvThreadSafe("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 += GetenvThreadSafeAsString("OMPI_URI_FILE");
}
+#elif defined(MPICH)
+ command += "-nameserver " + Kernel_Utils::GetHostname();
+#endif
+ command += " SALOME_MPIContainer ";
+ }
+ else
+ command += " " +container_exe+ " ";
- // "ssh -l user machine distantPath/runRemote.sh hostNS portNS WORKINGDIR workingdir
- // SALOME_Container containerName &"
- command = getCommandToRunRemoteProcess(resInfo.Protocol, resInfo.HostName, resInfo.UserName);
+ command += _NS->ContainerName(params) + " ";
+ if(this->_isSSL)
+ {
+ Engines::EmbeddedNamingService_var ns = GetEmbeddedNamingService();
+ CORBA::String_var iorNS = _orb->object_to_string(ns);
+ command += std::string(iorNS);
+ }
+ else //if(!this->_isSSL)
+ {
+ command += " -";
+ AddOmninamesParams(command);
+ }
+ MESSAGE("command =" << command);
- if (resInfo.AppliPath != "")
- command += resInfo.AppliPath; // path relative to user@machine $HOME
- else
+ return command;
+}
+
+//=============================================================================
+//! Return a path to the directory with scripts templates
+/*!
+ * \return the path pointed by SALOME_KERNEL_SCRIPTS_DIR environment variable, if it is defined,
+ * ${KERNEL_ROOT_DIR}/share/salome/resources/separator/kernel/ScriptsTemplate - otherwise
+ */
+//=============================================================================
+std::string getScriptTemplateFilePath()
+{
+ auto parseScriptTemplateFilePath = []() -> std::string
+ {
+ std::string scriptTemplateFilePath = SALOME_ContainerManager::GetenvThreadSafeAsString("SALOME_KERNEL_SCRIPTS_DIR");
+ if (!scriptTemplateFilePath.empty())
{
- ASSERT(GetenvThreadSafe("APPLI"));
- command += GetenvThreadSafeAsString("APPLI"); // path relative to user@machine $HOME
+ return scriptTemplateFilePath;
}
+ else {
+ return SALOME_ContainerManager::GetenvThreadSafeAsString("KERNEL_ROOT_DIR") +
+ "/share/salome/resources/kernel/ScriptsTemplate";
+ }
+ };
- command += "/runRemote.sh ";
+ static const std::string scriptTemplateFilePath = parseScriptTemplateFilePath();
+ return scriptTemplateFilePath;
+}
- ASSERT(GetenvThreadSafe("NSHOST"));
- command += GetenvThreadSafeAsString("NSHOST"); // hostname of CORBA name server
+//=============================================================================
+//! Return a command line constructed based on Python scripts templates
+/*!
+ * \param theScriptName the name of Python script template
+ * \param theScriptParameters the queue of parameter values
+ * \return the command line constructed according to the given parameters
+ */
+//=============================================================================
+std::string GetCommandFromTemplate(const std::string& theScriptName,
+ std::queue<std::string>& theScriptParameters)
+{
+ std::string command;
+ AutoGIL agil;
+ // manage GIL
- command += " ";
- ASSERT(GetenvThreadSafe("NSPORT"));
- command += GetenvThreadSafeAsString("NSPORT"); // port of CORBA name server
+ PyObject* mod(PyImport_ImportModule(theScriptName.c_str()));
+ if (!mod)
+ {
+ PyObject* sys = PyImport_ImportModule("sys");
+ PyObject* sys_path = PyObject_GetAttrString(sys, "path");
+ PyObject* folder_path = PyUnicode_FromString(getScriptTemplateFilePath().c_str());
+ PyList_Append(sys_path, folder_path);
- std::string wdir = params.workingdir.in();
- if(wdir != "")
- {
- command += " WORKINGDIR ";
- command += " '";
- if(wdir == "$TEMPDIR")
- wdir="\\$TEMPDIR";
- command += wdir; // requested working directory
- command += "'";
- }
+ mod = PyImport_ImportModule(theScriptName.c_str());
- if(params.isMPI)
+ Py_XDECREF(folder_path);
+ Py_XDECREF(sys_path);
+ Py_XDECREF(sys);
+ }
+
+ if (mod)
+ {
+ PyObject* meth(PyObject_GetAttrString(mod, "command"));
+ if (!meth)
{
- command += " mpirun -np ";
- std::ostringstream o;
- o << nbproc << " ";
- command += o.str();
-#ifdef LAM_MPI
- command += "-x PATH,LD_LIBRARY_PATH,OMNIORB_CONFIG,SALOME_trace ";
-#elif defined(OPEN_MPI)
- if( GetenvThreadSafe("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 += GetenvThreadSafeAsString("OMPI_URI_FILE");
- }
-#elif defined(MPICH)
- command += "-nameserver " + Kernel_Utils::GetHostname();
-#endif
- command += " SALOME_MPIContainer ";
+ Py_XDECREF(mod);
}
else
- command += " " +container_exe+ " ";
+ {
+ int id = -1;
+ PyObject* tuple(PyTuple_New(theScriptParameters.size()));
- command += _NS->ContainerName(params);
- command += " -";
- AddOmninamesParams(command);
+ auto insert_parameter = [&tuple, &theScriptParameters, &id]()
+ {
+ if (!theScriptParameters.empty())
+ {
+ PyTuple_SetItem(tuple, ++id, PyUnicode_FromString(theScriptParameters.front().c_str()));
+ theScriptParameters.pop();
+ }
+ };
+
+ while (!theScriptParameters.empty())
+ {
+ insert_parameter();
+ }
+
+ PyObject *args(PyTuple_New(1));
+ PyTuple_SetItem(args, 0, tuple);
+
+ PyObject *res(PyObject_CallObject(meth, args));
+ if (res)
+ {
+ command = PyUnicode_AsUTF8(res);
+ Py_XDECREF(res);
+ }
- MESSAGE("command =" << command);
+ Py_XDECREF(args);
+ Py_XDECREF(tuple);
+ Py_XDECREF(meth);
+ Py_XDECREF(mod);
+ }
}
+ MESSAGE("Command from template is ... " << command << std::endl);
return command;
}
+//=============================================================================
//=============================================================================
/*!
//=============================================================================
std::string SALOME_ContainerManager::BuildCommandToLaunchLocalContainer(const Engines::ContainerParameters& params, const std::string& machinesFile, const std::string& container_exe, std::string& tmpFileName) const
{
- tmpFileName = BuildTemporaryFileName();
- std::string command;
- int nbproc = 0;
-
- std::ostringstream o;
-
+ // Prepare name of the script to be used
+ std::string script_name = "SALOME_CM_LOCAL_NO_MPI";
if (params.isMPI)
- {
- o << "mpirun -np ";
-
- if ( params.nb_proc <= 0 )
- nbproc = 1;
- else
- nbproc = params.nb_proc;
-
- o << nbproc << " ";
-
- if( GetenvThreadSafe("LIBBATCH_NODEFILE") != NULL )
- o << "-machinefile " << machinesFile << " ";
-
+ {
#ifdef LAM_MPI
- o << "-x PATH,LD_LIBRARY_PATH,OMNIORB_CONFIG,SALOME_trace ";
+ script_name = "SALOME_CM_LOCAL_MPI_LAN";
#elif defined(OPEN_MPI)
- if( GetenvThreadSafe("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 << GetenvThreadSafeAsString("OMPI_URI_FILE");
- }
+ script_name = "SALOME_CM_LOCAL_MPI_OPENMPI";
#elif defined(MPICH)
- o << "-nameserver " + Kernel_Utils::GetHostname();
+ script_name = "SALOME_CM_LOCAL_MPI_MPICH";
#endif
+ }
+
+ // Prepare parameters to use in the Python script:
+ // 1. All parameters are strings.
+ // 2. For some booleans use "1" = True, "0" = False.
+ // 3. If a parameter is NULL, then its value is "NULL".
+
+ std::queue<std::string> script_parameters;
+
+ // ===== Number of processes (key = "nb_proc")
+ script_parameters.push(params.isMPI ? std::to_string(params.nb_proc <= 0 ? 1 : params.nb_proc) : "NULL");
+
+ // ===== Working directory (key = "workdir") and temporary directory flag (key = "isTmpDir")
+ // A working directory is requested
+ std::string workdir = params.workingdir.in();
+ std::string isTmpDir = std::to_string(0);
+ if (workdir == "$TEMPDIR")
+ {
+ // A new temporary directory is requested
+ isTmpDir = std::to_string(1);
+ workdir = Kernel_Utils::GetTmpDir();
+ }
+ script_parameters.push(workdir);
+ script_parameters.push(isTmpDir);
+
+ // ===== Server name (key = "name_server")
+ script_parameters.push(Kernel_Utils::GetHostname());
+
+ // ===== Container (key = "container")
+ std::string container;
+ if (params.isMPI)
+ {
+ container = isPythonContainer(params.container_name) ? "pyMPI SALOME_ContainerPy.py" : "SALOME_MPIContainer";
+ }
+ else
+ {
+ container = isPythonContainer(params.container_name) ? "SALOME_ContainerPy.py" : container_exe;
+ }
+ script_parameters.push(container);
- if (isPythonContainer(params.container_name))
- o << " pyMPI SALOME_ContainerPy.py ";
- else
- o << " SALOME_MPIContainer ";
- }
+ // ===== Container name (key = "container_name")
+ script_parameters.push(_NS->ContainerName(params));
- else
- {
- std::string wdir=params.workingdir.in();
- if(wdir != "")
- {
- // a working directory is requested
- if(wdir == "$TEMPDIR")
- {
- // a new temporary directory is requested
- std::string dir = Kernel_Utils::GetTmpDir();
-#ifdef WIN32
- o << "cd /d " << dir << std::endl;
-#else
- o << "cd " << dir << ";";
-#endif
+ // ===== LIBBATCH node file (key = "libbatch_nodefile")
+ script_parameters.push(std::to_string(GetenvThreadSafe("LIBBATCH_NODEFILE") != NULL ? 1 : 0));
- }
- else
- {
- // a permanent directory is requested use it or create it
-#ifdef WIN32
- o << "mkdir " + wdir << std::endl;
- o << "cd /D " + wdir << std::endl;
-#else
- o << "mkdir -p " << wdir << " && cd " << wdir + ";";
-#endif
- }
- }
+ // ===== Machine file (key = "machine_file")
+ script_parameters.push(machinesFile.empty() ? "NULL" : machinesFile);
- if (isPythonContainer(params.container_name))
- o << "SALOME_ContainerPy.py ";
- else
- o << container_exe + " ";
+ // ===== OMPI uri file (key = "ompi_uri_file")
+ std::string ompi_uri_file = GetenvThreadSafeAsString("OMPI_URI_FILE");
+ script_parameters.push(ompi_uri_file.empty() ? "NULL" : ompi_uri_file);
- }
+ std::string command_from_template = GetCommandFromTemplate(script_name, script_parameters);
- o << _NS->ContainerName(params);
- o << " -";
- AddOmninamesParams(o);
+ std::ostringstream o;
+ o << command_from_template << " ";
+
+ //==================================================================================== */
+ if( this->_isSSL )
+ {
+ Engines::EmbeddedNamingService_var ns = GetEmbeddedNamingService();
+ CORBA::String_var iorNS = _orb->object_to_string(ns);
+ o << iorNS;
+ }
+ else
+ {
+ o << "-";
+ AddOmninamesParams(o);
+ }
+
+ tmpFileName = BuildTemporaryFileName();
std::ofstream command_file( tmpFileName.c_str() );
command_file << o.str();
command_file.close();
#ifndef WIN32
chmod(tmpFileName.c_str(), 0x1ED);
#endif
- command = tmpFileName;
-
+
+ std::string command = tmpFileName;
MESSAGE("Command is file ... " << command);
MESSAGE("Command is ... " << o.str());
return command;
void SALOME_ContainerManager::RmTmpFile(std::string& tmpFileName)
{
- int length = tmpFileName.size();
+ size_t length = tmpFileName.size();
if ( length > 0)
{
#ifdef WIN32
*/
//=============================================================================
-void SALOME_ContainerManager::AddOmninamesParams(std::ostream& fileStream, SALOME_NamingService *ns)
+void SALOME_ContainerManager::AddOmninamesParams(std::ostream& fileStream, SALOME_NamingService_Abstract *ns)
{
- CORBA::String_var iorstr(ns->getIORaddr());
- fileStream << "ORBInitRef NameService=";
- fileStream << iorstr;
+ SALOME_NamingService *nsTrad(dynamic_cast<SALOME_NamingService *>(ns));
+ if(nsTrad)
+ {
+ CORBA::String_var iorstr(nsTrad->getIORaddr());
+ fileStream << "ORBInitRef NameService=";
+ fileStream << iorstr;
+ }
}
void SALOME_ContainerManager::MakeTheCommandToBeLaunchedASync(std::string& command)
if (params.isMPI)
{
- tempOutputFile << "mpirun -np ";
- int nbproc;
+ int nbproc = params.nb_proc <= 0 ? 1 : params.nb_proc;
- if ( params.nb_proc <= 0 )
- nbproc = 1;
- else
- nbproc = params.nb_proc;
-
- std::ostringstream o;
+ tempOutputFile << "mpirun -np ";
tempOutputFile << nbproc << " ";
#ifdef LAM_MPI
else if (resInfo.Protocol == srun)
{
- command = "srun -n 1 -N 1 --share --nodelist=";
+ command = "srun -n 1 -N 1 -s --mem-per-cpu=0 --cpu-bind=none --nodelist=";
std::string commandRcp = "rcp ";
commandRcp += tmpFileName;
commandRcp += " ";
{
if (_isAppliSalomeDefined)
{
-
- if (resInfo.Protocol == rsh)
- command = "rsh ";
- else if (resInfo.Protocol == ssh)
- command = "ssh ";
- else if (resInfo.Protocol == srun)
- command = "srun -n 1 -N 1 --share --nodelist=";
- 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(GetenvThreadSafe("APPLI"));
- command += GetenvThreadSafeAsString("APPLI"); // path relative to user@machine $HOME
- }
-
- command += "/runRemote.sh ";
-
- ASSERT(GetenvThreadSafe("NSHOST"));
- command += GetenvThreadSafeAsString("NSHOST"); // hostname of CORBA name server
-
- command += " ";
- ASSERT(GetenvThreadSafe("NSPORT"));
- command += GetenvThreadSafeAsString("NSPORT"); // port of CORBA name server
-
+ command = getCommandToRunRemoteProcess(resInfo.Protocol, resInfo.HostName,
+ resInfo.UserName, resInfo.AppliPath);
command += " mpirun -np 1 hostname -s > " + tmpFile;
}
else
}
-std::set<pid_t> SALOME_ContainerManager::getpidofprogram(const std::string program)
+std::string SALOME_ContainerManager::getCommandToRunRemoteProcessNoAppli(AccessProtocolType protocol, const std::string & hostname, const std::string & username, const std::string & applipath, const std::string & workdir) const
{
- std::set<pid_t> thepids;
- std::string tmpFile = Kernel_Utils::GetTmpFileName();
- std::string cmd;
- std::string thepid;
- cmd = "pidof " + program + " > " + tmpFile;
- SystemThreadSafe(cmd.c_str());
- std::ifstream fpi(tmpFile.c_str(),std::ios::in);
- while(fpi >> thepid){
- thepids.insert(atoi(thepid.c_str()));
- }
- return thepids;
+ return getCommandToRunRemoteProcessCommon("SALOME_CM_REMOTE","salome shell --",protocol,hostname,username,applipath,workdir);
}
-std::string SALOME_ContainerManager::getCommandToRunRemoteProcess(AccessProtocolType protocol,
+std::string SALOME_ContainerManager::getCommandToRunRemoteProcess(AccessProtocolType protocol, const std::string & hostname, const std::string & username, const std::string & applipath, const std::string & workdir) const
+{
+ return getCommandToRunRemoteProcessCommon("SALOME_CM_REMOTE_OLD",this->GetRunRemoteExecutableScript(),protocol,hostname,username,applipath,workdir);
+}
+
+std::string SALOME_ContainerManager::getCommandToRunRemoteProcessCommon(const std::string& templateName,
+ const std::string& remoteScript,
+ AccessProtocolType protocol,
const std::string & hostname,
- const std::string & username)
+ const std::string & username,
+ const std::string & applipath,
+ const std::string & workdir) const
{
std::ostringstream command;
+
+ // Prepare parameters to use in the Python script:
+ // 1. All parameters are strings.
+ // 2. For some booleans use "1" = True, "0" = False.
+ // 3. If a parameter is NULL, then its value is "NULL".
+
+ std::queue<std::string> script_parameters;
+
+ // ===== Protocol (key = "protocol")
+ std::string strProtocol;
switch (protocol)
{
- case rsh:
- command << "rsh ";
- if (username != "")
- {
- command << "-l " << username << " ";
- }
- command << hostname << " ";
- break;
- case ssh:
- command << "ssh ";
- if (username != "")
- {
- command << "-l " << username << " ";
- }
- command << hostname << " ";
- break;
- case srun:
- // no need to redefine the user with srun, the job user is taken by default
- // (note: for srun, user id can be specified with " --uid=<user>")
- command << "srun -n 1 -N 1 --share --nodelist=" << hostname << " ";
- break;
- case pbsdsh:
- command << "pbsdsh -o -h " << hostname << " ";
- break;
- case blaunch:
- command << "blaunch -no-shell " << hostname << " ";
- break;
+ case rsh: strProtocol = "rsh"; break;
+ case ssh: strProtocol = "ssh"; break;
+ case srun: strProtocol = "srun"; break;
+ case pbsdsh: strProtocol = "pbsdsh"; break;
+ case blaunch: strProtocol = "blaunch"; break;
default:
throw SALOME_Exception("Unknown protocol");
}
+ script_parameters.push(strProtocol);
+
+ // ===== User name (key = "user")
+ script_parameters.push(username.empty() ? "NULL" : username);
+
+ // ===== Host name (key = "host")
+ script_parameters.push(hostname.empty() ? "NULL" : hostname);
+
+
+ // ===== Remote APPLI path (key = "appli")
+ script_parameters.push(applipath.empty() ? GetenvThreadSafeAsString("APPLI") : applipath);
+
+ if(!this->_isSSL)
+ {
+ ASSERT(GetenvThreadSafe("NSHOST"));
+ ASSERT(GetenvThreadSafe("NSPORT"));
+ }
+
+ struct stat statbuf;
+ std::string appli_mode = (stat(GetenvThreadSafe("APPLI"), &statbuf) == 0 && S_ISREG(statbuf.st_mode)) ? "launcher" : "dir";
+
+ // ===== Working directory (key = "workdir")
+ script_parameters.push(workdir == "$TEMPDIR" ? "\\$TEMPDIR" : workdir);
+
+ // ===== SSL (key = "ssl")
+ script_parameters.push(this->_isSSL ? "1" : "0");
+
+ // ===== Hostname of CORBA name server (key = "nshost")
+ std::string nshost = GetenvThreadSafeAsString("NSHOST");
+ script_parameters.push(nshost.empty() ? "NULL" : nshost);
+
+ // ===== Port of CORBA name server (key = "nsport")
+ std::string nsport = GetenvThreadSafeAsString("NSPORT");
+ script_parameters.push(nsport.empty() ? "NULL" : nsport);
+
+ // ===== Remote script (key = "remote_script")
+ script_parameters.push(remoteScript.empty() ? "NONE" : remoteScript);
+
+ // ===== Naming service (key = "naming_service")
+ std::string namingService = "NONE";
+ if(this->_isSSL)
+ {
+ Engines::EmbeddedNamingService_var ns = GetEmbeddedNamingService();
+ CORBA::String_var iorNS = _orb->object_to_string(ns);
+ namingService = iorNS;
+ }
+ script_parameters.push(namingService);
+
+ // ===== APPLI mode (key = "appli_mode")
+ // $APPLI points either to an application directory, or to a salome launcher file
+ // we prepare the remote command according to the case
+ script_parameters.push(appli_mode);
+
+ command << GetCommandFromTemplate(templateName, script_parameters);
return command.str();
}
return system(command);
}
+long SALOME_ContainerManager::SystemWithPIDThreadSafe(const std::vector<std::string>& command)
+{
+ Utils_Locker lock(&_systemMutex);
+ if(command.size()<1)
+ throw SALOME_Exception("SystemWithPIDThreadSafe : command is expected to have a length of size 1 at least !");
+#ifndef WIN32
+ pid_t pid ( fork() ) ; // spawn a child process, following code is executed in both processes
+#else
+ pid_t pid = -1; //Throw SALOME_Exception on Windows
+#endif
+ if ( pid == 0 ) // I'm a child, replace myself with a new ompi-server
+ {
+ std::size_t sz(command.size());
+ char **args = new char *[sz+1];
+ for(std::size_t i=0;i<sz;i++)
+ args[i] = strdup(command[i].c_str());
+ args[sz] = nullptr;
+ execvp( command[0].c_str() , args );
+ std::ostringstream oss;
+ oss << "Error when launching " << command[0];
+ throw SALOME_Exception(oss.str().c_str()); // execvp failed
+ }
+ else if ( pid < 0 )
+ {
+ throw SALOME_Exception("fork() failed");
+ }
+ else // I'm a parent
+ {
+ return pid;
+ }
+}
+
#ifdef WITH_PACO_PARALLEL
//=============================================================================
remote_execution = true;
}
- // Log environnement
+ // Log environment
std::string log_type("");
char * get_val = GetenvThreadSafe("PARALLEL_LOG");
if (get_val)
ParserResourcesType resource_definition =
_resManager->GetResourceDefinition(params.resource_params.name.in());
- // Log environnement
+ // Log environment
std::string log_type("");
char * get_val = GetenvThreadSafe("PARALLEL_LOG");
if (get_val)
#else
Engines::Container_ptr
-SALOME_ContainerManager::StartPaCOPPContainer(const Engines::ContainerParameters& params,
- std::string resource_selected)
+SALOME_ContainerManager::StartPaCOPPContainer(const Engines::ContainerParameters& /*params*/,
+ std::string /*resource_selected*/)
{
Engines::Container_ptr ret = Engines::Container::_nil();
INFOS("[StarPaCOPPContainer] is disabled !");
}
std::string
-SALOME_ContainerManager::BuildCommandToLaunchPaCOProxyContainer(const Engines::ContainerParameters& params,
- std::string machine_file_name,
- std::string & proxy_hostname)
+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)
+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)
+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)
+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)
+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;
}
