1 // Copyright (C) 2007-2011 CEA/DEN, EDF R&D, OPEN CASCADE
3 // Copyright (C) 2003-2007 OPEN CASCADE, EADS/CCR, LIP6, CEA/DEN,
4 // CEDRAT, EDF R&D, LEG, PRINCIPIA R&D, BUREAU VERITAS
6 // This library is free software; you can redistribute it and/or
7 // modify it under the terms of the GNU Lesser General Public
8 // License as published by the Free Software Foundation; either
9 // version 2.1 of the License.
11 // This library is distributed in the hope that it will be useful,
12 // but WITHOUT ANY WARRANTY; without even the implied warranty of
13 // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 // Lesser General Public License for more details.
16 // You should have received a copy of the GNU Lesser General Public
17 // License along with this library; if not, write to the Free Software
18 // Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 // See http://www.salome-platform.org/ or email : webmaster.salome@opencascade.com
24 #include <Batch/Batch_Date.hxx>
25 #include <Batch/Batch_BatchManagerCatalog.hxx>
26 #include <Batch/Batch_FactBatchManager_eClient.hxx>
27 #include <Batch/Batch_BatchManager_eClient.hxx>
30 #include "Basics_Utils.hxx"
31 #include "Basics_DirUtils.hxx"
32 #include "SALOME_Launcher_Handler.hxx"
33 #include "Launcher.hxx"
34 #include "Launcher_Job_Command.hxx"
40 //=============================================================================
44 * Define a CORBA single thread policy for the server, which avoid to deal
45 * with non thread-safe usage like Change_Directory in SALOME naming service
47 //=============================================================================
48 Launcher_cpp::Launcher_cpp()
50 LAUNCHER_MESSAGE("Launcher_cpp constructor");
52 _job_cpt_mutex = new pthread_mutex_t();
53 pthread_mutex_init(_job_cpt_mutex, NULL);
56 //=============================================================================
60 //=============================================================================
61 Launcher_cpp::~Launcher_cpp()
63 LAUNCHER_MESSAGE("Launcher_cpp destructor");
65 std::map<int, Launcher::Job *>::const_iterator it_job;
66 for(it_job = _launcher_job_map.begin(); it_job != _launcher_job_map.end(); it_job++)
67 delete it_job->second;
68 std::map <int, Batch::BatchManager_eClient * >::const_iterator it1;
69 for(it1=_batchmap.begin();it1!=_batchmap.end();it1++)
73 pthread_mutex_destroy(_job_cpt_mutex);
74 delete _job_cpt_mutex;
79 //=============================================================================
81 * Add a job into the launcher - check resource and choose one
83 //=============================================================================
85 Launcher_cpp::createJob(Launcher::Job * new_job)
87 LAUNCHER_MESSAGE("Creating a new job");
88 // Add job to the jobs map
89 pthread_mutex_lock(_job_cpt_mutex);
90 new_job->setNumber(_job_cpt);
92 pthread_mutex_unlock(_job_cpt_mutex);
93 std::map<int, Launcher::Job *>::const_iterator it_job = _launcher_job_map.find(new_job->getNumber());
94 if (it_job == _launcher_job_map.end())
95 _launcher_job_map[new_job->getNumber()] = new_job;
98 LAUNCHER_INFOS("A job as already the same id: " << new_job->getNumber());
100 throw LauncherException("A job as already the same id - job is not created !");
102 LAUNCHER_MESSAGE("New Job created");
105 //=============================================================================
109 //=============================================================================
111 Launcher_cpp::launchJob(int job_id)
113 LAUNCHER_MESSAGE("Launch a job");
115 // Check if job exist
116 std::map<int, Launcher::Job *>::const_iterator it_job = _launcher_job_map.find(job_id);
117 if (it_job == _launcher_job_map.end())
119 LAUNCHER_INFOS("Cannot find the job, is it created ? job number: " << job_id);
120 throw LauncherException("Cannot find the job, is it created ?");
123 Launcher::Job * job = it_job->second;
125 // Check job state (cannot launch a job already launched...)
126 if (job->getState() != "CREATED")
128 LAUNCHER_INFOS("Bad state of the job: " << job->getState());
129 throw LauncherException("Bad state of the job: " + job->getState());
132 // Third step search batch manager for the job into the map -> instanciate one if does not exist
134 std::map<int, Batch::BatchManager_eClient *>::const_iterator it = _batchmap.find(job_id);
135 if(it == _batchmap.end())
137 createBatchManagerForJob(job);
142 Batch::JobId batch_manager_job_id = _batchmap[job_id]->submitJob(*(job->getBatchJob()));
143 job->setBatchManagerJobId(batch_manager_job_id);
144 job->setState("QUEUED");
146 catch(const Batch::EmulationException &ex)
148 LAUNCHER_INFOS("Job is not launched, exception in submitJob: " << ex.message);
149 throw LauncherException(ex.message.c_str());
151 LAUNCHER_MESSAGE("Job launched");
154 //=============================================================================
158 //=============================================================================
160 Launcher_cpp::getJobState(int job_id)
162 LAUNCHER_MESSAGE("Get job state");
164 // Check if job exist
165 std::map<int, Launcher::Job *>::const_iterator it_job = _launcher_job_map.find(job_id);
166 if (it_job == _launcher_job_map.end())
168 LAUNCHER_INFOS("Cannot find the job, is it created ? job number: " << job_id);
169 throw LauncherException("Cannot find the job, is it created ?");
172 Launcher::Job * job = it_job->second;
173 std::string state = job->updateJobState();
175 return state.c_str();
178 //=============================================================================
180 * Get Job result - the result directory could be changed
182 //=============================================================================
184 Launcher_cpp::getJobResults(int job_id, std::string directory)
186 LAUNCHER_MESSAGE("Get Job results");
188 // Check if job exist
189 std::map<int, Launcher::Job *>::const_iterator it_job = _launcher_job_map.find(job_id);
190 if (it_job == _launcher_job_map.end())
192 LAUNCHER_INFOS("Cannot find the job, is it created ? job number: " << job_id);
193 throw LauncherException("Cannot find the job, is it created ?");
196 Launcher::Job * job = it_job->second;
197 std::string resource_name = job->getResourceDefinition().Name;
201 _batchmap[job_id]->importOutputFiles(*(job->getBatchJob()), directory);
203 _batchmap[job_id]->importOutputFiles(*(job->getBatchJob()), job->getResultDirectory());
205 catch(const Batch::EmulationException &ex)
207 LAUNCHER_INFOS("getJobResult is maybe incomplete, exception: " << ex.message);
208 throw LauncherException(ex.message.c_str());
210 LAUNCHER_MESSAGE("getJobResult ended");
213 //=============================================================================
215 * Get Job dump state - the result directory could be changed
217 //=============================================================================
219 Launcher_cpp::getJobDumpState(int job_id, std::string directory)
222 LAUNCHER_MESSAGE("Get Job dump state");
224 // Check if job exist
225 std::map<int, Launcher::Job *>::const_iterator it_job = _launcher_job_map.find(job_id);
226 if (it_job == _launcher_job_map.end())
228 LAUNCHER_INFOS("Cannot find the job, is it created ? job number: " << job_id);
229 throw LauncherException("Cannot find the job, is it created ?");
232 Launcher::Job * job = it_job->second;
233 std::string resource_name = job->getResourceDefinition().Name;
237 rtn = _batchmap[job_id]->importDumpStateFile(*(job->getBatchJob()), directory);
239 rtn = _batchmap[job_id]->importDumpStateFile(*(job->getBatchJob()), job->getResultDirectory());
241 catch(const Batch::EmulationException &ex)
243 LAUNCHER_INFOS("getJobResult is maybe incomplete, exception: " << ex.message);
244 throw LauncherException(ex.message.c_str());
246 LAUNCHER_MESSAGE("getJobResult ended");
250 //=============================================================================
252 * Remove the job - into the Launcher and its batch manager
254 //=============================================================================
256 Launcher_cpp::removeJob(int job_id)
258 LAUNCHER_MESSAGE("Remove Job");
260 // Check if job exist
261 std::map<int, Launcher::Job *>::iterator it_job = _launcher_job_map.find(job_id);
262 if (it_job == _launcher_job_map.end())
264 LAUNCHER_INFOS("Cannot find the job, is it created ? job number: " << job_id);
265 throw LauncherException("Cannot find the job, is it created ?");
268 it_job->second->removeJob();
269 delete it_job->second;
270 _launcher_job_map.erase(it_job);
273 //=============================================================================
275 * create a launcher job based on a file
276 * \param xmlExecuteFile : to define the execution on the batch cluster
278 //=============================================================================
280 Launcher_cpp::createJobWithFile(const std::string xmlExecuteFile,
281 const std::string clusterName)
283 LAUNCHER_MESSAGE("Begin of Launcher_cpp::createJobWithFile");
286 ParserLauncherType job_params = ParseXmlFile(xmlExecuteFile);
288 // Creating a new job
289 Launcher::Job_Command * new_job = new Launcher::Job_Command();
291 std::string cmdFile = Kernel_Utils::GetTmpFileName();
298 os.open(cmdFile.c_str(), std::ofstream::out );
299 os << "#! /bin/sh" << std::endl;
300 os << job_params.Command;
303 new_job->setJobFile(cmdFile);
304 new_job->setLocalDirectory(job_params.RefDirectory);
305 new_job->setWorkDirectory(job_params.MachinesList[clusterName].WorkDirectory);
306 new_job->setEnvFile(job_params.MachinesList[clusterName].EnvFile);
308 for(int i=0; i < job_params.InputFile.size(); i++)
309 new_job->add_in_file(job_params.InputFile[i]);
310 for(int i=0; i < job_params.OutputFile.size();i++)
311 new_job->add_out_file(job_params.OutputFile[i]);
314 p.hostname = clusterName;
317 p.nb_proc = job_params.NbOfProcesses;
319 p.nb_proc_per_node = 0;
322 new_job->setResourceRequiredParams(p);
325 return new_job->getNumber();
328 //=============================================================================
330 * Factory to instanciate the good batch manager for choosen cluster.
332 //=============================================================================
333 Batch::BatchManager_eClient *
334 Launcher_cpp::FactoryBatchManager(ParserResourcesType& params)
337 Batch::CommunicationProtocolType protocol;
338 Batch::FactBatchManager_eClient* fact;
340 int nb_proc_per_node = params.DataForSort._nbOfProcPerNode;
341 std::string hostname = params.HostName;
343 switch(params.Protocol)
346 protocol = Batch::RSH;
349 protocol = Batch::SSH;
352 throw LauncherException("Unknown protocol for this resource");
381 switch( params.Batch )
405 LAUNCHER_MESSAGE("Bad batch description of the resource: Batch = " << params.Batch);
406 throw LauncherException("No batchmanager for that cluster - Bad batch description of the resource");
408 Batch::BatchManagerCatalog & cata = Batch::BatchManagerCatalog::getInstance();
409 fact = dynamic_cast<Batch::FactBatchManager_eClient*>(cata(bmType));
411 LAUNCHER_MESSAGE("Cannot find batch manager factory for " << bmType << ". Check your version of libBatch.");
412 throw LauncherException("Cannot find batch manager factory");
414 LAUNCHER_MESSAGE("Instanciation of batch manager of type: " << bmType);
415 Batch::BatchManager_eClient * batch_client = (*fact)(hostname.c_str(), params.UserName.c_str(),
416 protocol, mpi.c_str(), nb_proc_per_node);
420 //----------------------------------------------------------
421 // Without LIBBATCH - Launcher_cpp do nothing...
422 //----------------------------------------------------------
426 Launcher_cpp::createJob(Launcher::Job * new_job)
428 LAUNCHER_INFOS("Launcher compiled without LIBBATCH - cannot create a job !!!");
430 throw LauncherException("Method Launcher_cpp::createJob is not available "
431 "(libBatch was not present at compilation time)");
435 Launcher_cpp::launchJob(int job_id)
437 LAUNCHER_INFOS("Launcher compiled without LIBBATCH - cannot launch a job !!!");
438 throw LauncherException("Method Launcher_cpp::launchJob is not available "
439 "(libBatch was not present at compilation time)");
443 Launcher_cpp::getJobState(int job_id)
445 LAUNCHER_INFOS("Launcher compiled without LIBBATCH - cannot get job state!!!");
446 throw LauncherException("Method Launcher_cpp::getJobState is not available "
447 "(libBatch was not present at compilation time)");
451 Launcher_cpp::getJobResults(int job_id, std::string directory)
453 LAUNCHER_INFOS("Launcher compiled without LIBBATCH - cannot get job results!!!");
454 throw LauncherException("Method Launcher_cpp::getJobResults is not available "
455 "(libBatch was not present at compilation time)");
459 Launcher_cpp::getJobDumpState(int job_id, std::string directory)
461 LAUNCHER_INFOS("Launcher compiled without LIBBATCH - cannot get job dump state!!!");
462 throw LauncherException("Method Launcher_cpp::getJobDumpState is not available "
463 "(libBatch was not present at compilation time)");
467 Launcher_cpp::removeJob(int job_id)
469 LAUNCHER_INFOS("Launcher compiled without LIBBATCH - cannot remove job!!!");
470 throw LauncherException("Method Launcher_cpp::removeJob is not available "
471 "(libBatch was not present at compilation time)");
475 Launcher_cpp::createJobWithFile( const std::string xmlExecuteFile, std::string clusterName)
477 throw LauncherException("Method Launcher_cpp::createJobWithFile is not available "
478 "(libBatch was not present at compilation time)");
485 Launcher_cpp::ParseXmlFile(std::string xmlExecuteFile)
487 ParserLauncherType job_params;
488 SALOME_Launcher_Handler * handler = new SALOME_Launcher_Handler(job_params);
490 const char* aFilePath = xmlExecuteFile.c_str();
491 FILE* aFile = fopen(aFilePath, "r");
494 xmlDocPtr aDoc = xmlReadFile(aFilePath, NULL, 0);
496 handler->ProcessXmlDocument(aDoc);
499 std::string message = "ResourcesManager_cpp: could not parse file: " + xmlExecuteFile;
500 LAUNCHER_MESSAGE(message);
502 throw LauncherException(message);
510 std::string message = "ResourcesManager_cpp: file is not readable: " + xmlExecuteFile;
511 LAUNCHER_MESSAGE(message);
513 throw LauncherException(message);
521 std::map<int, Launcher::Job *>
522 Launcher_cpp::getJobs()
524 return _launcher_job_map;
528 Launcher_cpp::createBatchManagerForJob(Launcher::Job * job)
530 int job_id = job->getNumber();
532 // Select a ressource for the job
533 std::vector<std::string> ResourceList;
534 resourceParams params = job->getResourceRequiredParams();
537 ResourceList = _ResManager->GetFittingResources(params);
539 catch(const ResourcesException &ex)
541 throw LauncherException(ex.msg.c_str());
543 if (ResourceList.size() == 0)
545 LAUNCHER_INFOS("No adequate resource found for the job, number " << job->getNumber());
546 job->setState("ERROR");
547 throw LauncherException("No resource found the job");
550 // Configure the job with the resource selected - the first of the list
551 ParserResourcesType resource_definition = _ResManager->GetResourcesDescr(ResourceList[0]);
553 // Set resource definition to the job
554 // The job will check if the definitions needed
557 job->setResourceDefinition(resource_definition);
559 catch(const LauncherException &ex)
561 LAUNCHER_INFOS("Error in the definition of the resource, mess: " << ex.msg);
562 job->setState("ERROR");
566 // Step 2: We can now add a Factory if the resource is correctly define
568 std::map<int, Batch::BatchManager_eClient *>::const_iterator it = _batchmap.find(job_id);
569 if(it == _batchmap.end())
573 // Warning cannot write on one line like this, because map object is constructed before
574 // the method is called...
575 //_batchmap[job_id] = FactoryBatchManager(resource_definition);
576 Batch::BatchManager_eClient * batch_client = FactoryBatchManager(resource_definition);
577 _batchmap[job_id] = batch_client;
579 catch(const LauncherException &ex)
581 LAUNCHER_INFOS("Error during creation of the batch manager of the job, mess: " << ex.msg);
584 catch(const Batch::EmulationException &ex)
586 LAUNCHER_INFOS("Error during creation of the batch manager of the job, mess: " << ex.message);
587 throw LauncherException(ex.message);
589 catch(const Batch::InvalidArgumentException &ex)
591 LAUNCHER_INFOS("Error during creation of the batch manager of the job, mess: " << ex.message);
592 throw LauncherException(ex.message);
599 Launcher_cpp::addJobDirectlyToMap(Launcher::Job * new_job, const std::string job_reference)
601 // Step 0: Calculated job_id
602 pthread_mutex_lock(_job_cpt_mutex);
604 int job_id = _job_cpt;
605 new_job->setNumber(job_id);
606 pthread_mutex_unlock(_job_cpt_mutex);
608 // Step 1: check if resource is already in the map
609 createBatchManagerForJob(new_job);
611 // Step 2: add the job to the batch manager
615 Batch::JobId batch_manager_job_id = _batchmap[job_id]->addJob(*(new_job->getBatchJob()),
617 new_job->setBatchManagerJobId(batch_manager_job_id);
619 catch(const Batch::EmulationException &ex)
621 LAUNCHER_INFOS("Job cannot be added, exception in addJob: " << ex.message);
622 throw LauncherException(ex.message.c_str());
625 // Step 3: add job to launcher map
626 std::map<int, Launcher::Job *>::const_iterator it_job = _launcher_job_map.find(new_job->getNumber());
627 if (it_job == _launcher_job_map.end())
628 _launcher_job_map[new_job->getNumber()] = new_job;
631 LAUNCHER_INFOS("A job as already the same id: " << new_job->getNumber());
633 throw LauncherException("A job as already the same id - job is not created !");
635 LAUNCHER_MESSAGE("New job added");