1 // Copyright (C) 2007-2013 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
30 #include <libbatch/BatchManagerCatalog.hxx>
31 #include <libbatch/FactBatchManager.hxx>
32 #include <libbatch/BatchManager.hxx>
35 #include "Basics_Utils.hxx"
36 #include "Basics_DirUtils.hxx"
37 #include "SALOME_Launcher_Handler.hxx"
38 #include "Launcher.hxx"
39 #include "Launcher_Job_Command.hxx"
40 #include "Launcher_XML_Persistence.hxx"
44 //=============================================================================
48 * Define a CORBA single thread policy for the server, which avoid to deal
49 * with non thread-safe usage like Change_Directory in SALOME naming service
51 //=============================================================================
52 Launcher_cpp::Launcher_cpp()
54 LAUNCHER_MESSAGE("Launcher_cpp constructor");
56 _job_cpt_mutex = new pthread_mutex_t();
57 pthread_mutex_init(_job_cpt_mutex, NULL);
60 //=============================================================================
64 //=============================================================================
65 Launcher_cpp::~Launcher_cpp()
67 LAUNCHER_MESSAGE("Launcher_cpp destructor");
69 std::map<int, Launcher::Job *>::const_iterator it_job;
70 for(it_job = _launcher_job_map.begin(); it_job != _launcher_job_map.end(); it_job++)
71 delete it_job->second;
72 std::map <int, Batch::BatchManager * >::const_iterator it1;
73 for(it1=_batchmap.begin();it1!=_batchmap.end();it1++)
77 pthread_mutex_destroy(_job_cpt_mutex);
78 delete _job_cpt_mutex;
83 //=============================================================================
85 * Add a job into the launcher - check resource and choose one
87 //=============================================================================
89 Launcher_cpp::createJob(Launcher::Job * new_job)
91 LAUNCHER_MESSAGE("Creating a new job");
92 // Add job to the jobs map
93 pthread_mutex_lock(_job_cpt_mutex);
94 new_job->setNumber(_job_cpt);
96 pthread_mutex_unlock(_job_cpt_mutex);
97 std::map<int, Launcher::Job *>::const_iterator it_job = _launcher_job_map.find(new_job->getNumber());
98 if (it_job == _launcher_job_map.end())
99 _launcher_job_map[new_job->getNumber()] = new_job;
102 LAUNCHER_INFOS("A job as already the same id: " << new_job->getNumber());
104 throw LauncherException("A job as already the same id - job is not created !");
106 LAUNCHER_MESSAGE("New Job created");
109 //=============================================================================
113 //=============================================================================
115 Launcher_cpp::launchJob(int job_id)
117 LAUNCHER_MESSAGE("Launch a job");
119 // Check if job exist
120 std::map<int, Launcher::Job *>::const_iterator it_job = _launcher_job_map.find(job_id);
121 if (it_job == _launcher_job_map.end())
123 LAUNCHER_INFOS("Cannot find the job, is it created ? job number: " << job_id);
124 throw LauncherException("Cannot find the job, is it created ?");
127 Launcher::Job * job = it_job->second;
129 // Check job state (cannot launch a job already launched...)
130 if (job->getState() != "CREATED")
132 LAUNCHER_INFOS("Bad state of the job: " << job->getState());
133 throw LauncherException("Bad state of the job: " + job->getState());
136 // Third step search batch manager for the job into the map -> instanciate one if does not exist
138 std::map<int, Batch::BatchManager *>::const_iterator it = _batchmap.find(job_id);
139 if(it == _batchmap.end())
141 createBatchManagerForJob(job);
146 Batch::JobId batch_manager_job_id = _batchmap[job_id]->submitJob(*(job->getBatchJob()));
147 job->setBatchManagerJobId(batch_manager_job_id);
148 job->setState("QUEUED");
150 catch(const Batch::GenericException &ex)
152 LAUNCHER_INFOS("Job is not launched, exception in submitJob: " << ex.message);
153 throw LauncherException(ex.message.c_str());
155 LAUNCHER_MESSAGE("Job launched");
158 //=============================================================================
162 //=============================================================================
164 Launcher_cpp::getJobState(int job_id)
166 LAUNCHER_MESSAGE("Get job state");
168 // Check if job exist
169 std::map<int, Launcher::Job *>::const_iterator it_job = _launcher_job_map.find(job_id);
170 if (it_job == _launcher_job_map.end())
172 LAUNCHER_INFOS("Cannot find the job, is it created ? job number: " << job_id);
173 throw LauncherException("Cannot find the job, is it created ?");
176 Launcher::Job * job = it_job->second;
181 state = job->updateJobState();
183 catch(const Batch::GenericException &ex)
185 LAUNCHER_INFOS("getJobState failed, exception: " << ex.message);
186 throw LauncherException(ex.message.c_str());
189 return state.c_str();
192 //=============================================================================
194 * Get job assigned hostnames
196 //=============================================================================
198 Launcher_cpp::getAssignedHostnames(int job_id)
200 LAUNCHER_MESSAGE("Get job assigned hostnames");
202 // Check if job exist
203 std::map<int, Launcher::Job *>::const_iterator it_job = _launcher_job_map.find(job_id);
204 if (it_job == _launcher_job_map.end())
206 LAUNCHER_INFOS("Cannot find the job, is it created ? job number: " << job_id);
207 throw LauncherException("Cannot find the job, is it created ?");
210 Launcher::Job * job = it_job->second;
211 std::string assigned_hostnames = job->getAssignedHostnames();
213 return assigned_hostnames.c_str();
216 //=============================================================================
218 * Get Job result - the result directory could be changed
220 //=============================================================================
222 Launcher_cpp::getJobResults(int job_id, std::string directory)
224 LAUNCHER_MESSAGE("Get Job results");
226 // Check if job exist
227 std::map<int, Launcher::Job *>::const_iterator it_job = _launcher_job_map.find(job_id);
228 if (it_job == _launcher_job_map.end())
230 LAUNCHER_INFOS("Cannot find the job, is it created ? job number: " << job_id);
231 throw LauncherException("Cannot find the job, is it created ?");
234 Launcher::Job * job = it_job->second;
235 std::string resource_name = job->getResourceDefinition().Name;
239 _batchmap[job_id]->importOutputFiles(*(job->getBatchJob()), directory);
241 _batchmap[job_id]->importOutputFiles(*(job->getBatchJob()), job->getResultDirectory());
243 catch(const Batch::GenericException &ex)
245 LAUNCHER_INFOS("getJobResult is maybe incomplete, exception: " << ex.message);
246 throw LauncherException(ex.message.c_str());
248 LAUNCHER_MESSAGE("getJobResult ended");
251 //=============================================================================
253 * Get Job dump state - the result directory could be changed
255 //=============================================================================
257 Launcher_cpp::getJobDumpState(int job_id, std::string directory)
260 LAUNCHER_MESSAGE("Get Job dump state");
262 // Check if job exist
263 std::map<int, Launcher::Job *>::const_iterator it_job = _launcher_job_map.find(job_id);
264 if (it_job == _launcher_job_map.end())
266 LAUNCHER_INFOS("Cannot find the job, is it created ? job number: " << job_id);
267 throw LauncherException("Cannot find the job, is it created ?");
270 Launcher::Job * job = it_job->second;
271 std::string resource_name = job->getResourceDefinition().Name;
275 rtn = _batchmap[job_id]->importDumpStateFile(*(job->getBatchJob()), directory);
277 rtn = _batchmap[job_id]->importDumpStateFile(*(job->getBatchJob()), job->getResultDirectory());
279 catch(const Batch::GenericException &ex)
281 LAUNCHER_INFOS("getJobResult is maybe incomplete, exception: " << ex.message);
282 throw LauncherException(ex.message.c_str());
284 LAUNCHER_MESSAGE("getJobResult ended");
288 //=============================================================================
290 * Remove the job - into the Launcher and its batch manager
292 //=============================================================================
294 Launcher_cpp::removeJob(int job_id)
296 LAUNCHER_MESSAGE("Remove Job");
298 // Check if job exist
299 std::map<int, Launcher::Job *>::iterator it_job = _launcher_job_map.find(job_id);
300 if (it_job == _launcher_job_map.end())
302 LAUNCHER_INFOS("Cannot find the job, is it created ? job number: " << job_id);
303 throw LauncherException("Cannot find the job, is it created ?");
306 it_job->second->removeJob();
307 delete it_job->second;
308 _launcher_job_map.erase(it_job);
311 //=============================================================================
315 //=============================================================================
317 Launcher_cpp::stopJob(int job_id)
319 LAUNCHER_MESSAGE("Stop Job");
321 // Check if job exist
322 std::map<int, Launcher::Job *>::iterator it_job = _launcher_job_map.find(job_id);
323 if (it_job == _launcher_job_map.end())
325 LAUNCHER_INFOS("Cannot find the job, is it created ? job number: " << job_id);
326 throw LauncherException("Cannot find the job, is it created ?");
329 it_job->second->stopJob();
332 //=============================================================================
334 * create a launcher job based on a file
335 * \param xmlExecuteFile : to define the execution on the batch cluster
337 //=============================================================================
339 Launcher_cpp::createJobWithFile(const std::string xmlExecuteFile,
340 const std::string clusterName)
342 LAUNCHER_MESSAGE("Begin of Launcher_cpp::createJobWithFile");
345 ParserLauncherType job_params = ParseXmlFile(xmlExecuteFile);
347 // Creating a new job
348 Launcher::Job_Command * new_job = new Launcher::Job_Command();
350 std::string cmdFile = Kernel_Utils::GetTmpFileName();
357 os.open(cmdFile.c_str(), std::ofstream::out );
358 os << "#! /bin/sh" << std::endl;
359 os << job_params.Command;
362 new_job->setJobFile(cmdFile);
363 new_job->setLocalDirectory(job_params.RefDirectory);
364 new_job->setWorkDirectory(job_params.MachinesList[clusterName].WorkDirectory);
365 new_job->setEnvFile(job_params.MachinesList[clusterName].EnvFile);
367 for(int i=0; i < job_params.InputFile.size(); i++)
368 new_job->add_in_file(job_params.InputFile[i]);
369 for(int i=0; i < job_params.OutputFile.size();i++)
370 new_job->add_out_file(job_params.OutputFile[i]);
373 p.hostname = clusterName;
376 p.nb_proc = job_params.NbOfProcesses;
378 p.nb_proc_per_node = 0;
381 new_job->setResourceRequiredParams(p);
384 return new_job->getNumber();
387 //=============================================================================
389 * Factory to instanciate the good batch manager for choosen cluster.
391 //=============================================================================
392 Batch::BatchManager *
393 Launcher_cpp::FactoryBatchManager(ParserResourcesType& params)
396 Batch::CommunicationProtocolType protocol;
397 Batch::FactBatchManager * fact;
399 std::string hostname = params.HostName;
401 switch(params.Protocol)
404 protocol = Batch::SH;
407 protocol = Batch::RSH;
410 protocol = Batch::SSH;
413 throw LauncherException("Unknown protocol for this resource");
445 switch( params.Batch )
478 LAUNCHER_MESSAGE("Bad batch description of the resource: Batch = " << params.Batch);
479 throw LauncherException("No batchmanager for that cluster - Bad batch description of the resource");
481 Batch::BatchManagerCatalog & cata = Batch::BatchManagerCatalog::getInstance();
482 fact = dynamic_cast<Batch::FactBatchManager*>(cata(bmType));
484 LAUNCHER_MESSAGE("Cannot find batch manager factory for " << bmType << ". Check your version of libBatch.");
485 throw LauncherException("Cannot find batch manager factory");
487 LAUNCHER_MESSAGE("Instanciation of batch manager of type: " << bmType);
488 Batch::BatchManager * batch_client = (*fact)(hostname.c_str(), params.UserName.c_str(),
489 protocol, mpi.c_str());
493 //----------------------------------------------------------
494 // Without LIBBATCH - Launcher_cpp do nothing...
495 //----------------------------------------------------------
499 Launcher_cpp::createJob(Launcher::Job * new_job)
501 LAUNCHER_INFOS("Launcher compiled without LIBBATCH - cannot create a job !!!");
503 throw LauncherException("Method Launcher_cpp::createJob is not available "
504 "(libBatch was not present at compilation time)");
508 Launcher_cpp::launchJob(int job_id)
510 LAUNCHER_INFOS("Launcher compiled without LIBBATCH - cannot launch a job !!!");
511 throw LauncherException("Method Launcher_cpp::launchJob is not available "
512 "(libBatch was not present at compilation time)");
516 Launcher_cpp::getJobState(int job_id)
518 LAUNCHER_INFOS("Launcher compiled without LIBBATCH - cannot get job state!!!");
519 throw LauncherException("Method Launcher_cpp::getJobState is not available "
520 "(libBatch was not present at compilation time)");
524 Launcher_cpp::getAssignedHostnames(int job_id)
526 LAUNCHER_INFOS("Launcher compiled without LIBBATCH - cannot get job assigned hostnames!!!");
527 throw LauncherException("Method Launcher_cpp::getAssignedHostnames is not available "
528 "(libBatch was not present at compilation time)");
532 Launcher_cpp::getJobResults(int job_id, std::string directory)
534 LAUNCHER_INFOS("Launcher compiled without LIBBATCH - cannot get job results!!!");
535 throw LauncherException("Method Launcher_cpp::getJobResults is not available "
536 "(libBatch was not present at compilation time)");
540 Launcher_cpp::getJobDumpState(int job_id, std::string directory)
542 LAUNCHER_INFOS("Launcher compiled without LIBBATCH - cannot get job dump state!!!");
543 throw LauncherException("Method Launcher_cpp::getJobDumpState is not available "
544 "(libBatch was not present at compilation time)");
548 Launcher_cpp::removeJob(int job_id)
550 LAUNCHER_INFOS("Launcher compiled without LIBBATCH - cannot remove job!!!");
551 throw LauncherException("Method Launcher_cpp::removeJob is not available "
552 "(libBatch was not present at compilation time)");
556 Launcher_cpp::stopJob(int job_id)
558 throw LauncherException("Method Launcher_cpp::stopJob is not available "
559 "(libBatch was not present at compilation time)");
563 Launcher_cpp::createJobWithFile( const std::string xmlExecuteFile, std::string clusterName)
565 throw LauncherException("Method Launcher_cpp::createJobWithFile is not available "
566 "(libBatch was not present at compilation time)");
573 Launcher_cpp::ParseXmlFile(std::string xmlExecuteFile)
575 ParserLauncherType job_params;
576 SALOME_Launcher_Handler * handler = new SALOME_Launcher_Handler(job_params);
578 const char* aFilePath = xmlExecuteFile.c_str();
579 FILE* aFile = fopen(aFilePath, "r");
582 xmlDocPtr aDoc = xmlReadFile(aFilePath, NULL, 0);
584 handler->ProcessXmlDocument(aDoc);
587 std::string message = "ResourcesManager_cpp: could not parse file: " + xmlExecuteFile;
588 LAUNCHER_MESSAGE(message);
590 throw LauncherException(message);
598 std::string message = "ResourcesManager_cpp: file is not readable: " + xmlExecuteFile;
599 LAUNCHER_MESSAGE(message);
601 throw LauncherException(message);
609 std::map<int, Launcher::Job *>
610 Launcher_cpp::getJobs()
612 return _launcher_job_map;
616 Launcher_cpp::createBatchManagerForJob(Launcher::Job * job)
618 int job_id = job->getNumber();
620 // Select a ressource for the job
621 std::vector<std::string> ResourceList;
622 resourceParams params = job->getResourceRequiredParams();
623 // Consider only resources that can launch batch jobs
624 params.can_launch_batch_jobs = true;
627 ResourceList = _ResManager->GetFittingResources(params);
629 catch(const ResourcesException &ex)
631 throw LauncherException(ex.msg.c_str());
633 if (ResourceList.size() == 0)
635 LAUNCHER_INFOS("No adequate resource found for the job, number " << job->getNumber());
636 job->setState("ERROR");
637 throw LauncherException("No resource found the job");
640 // Configure the job with the resource selected - the first of the list
641 ParserResourcesType resource_definition = _ResManager->GetResourcesDescr(ResourceList[0]);
643 // Set resource definition to the job
644 // The job will check if the definitions needed
647 job->setResourceDefinition(resource_definition);
649 catch(const LauncherException &ex)
651 LAUNCHER_INFOS("Error in the definition of the resource, mess: " << ex.msg);
652 job->setState("ERROR");
656 // Step 2: We can now add a Factory if the resource is correctly define
658 std::map<int, Batch::BatchManager *>::const_iterator it = _batchmap.find(job_id);
659 if(it == _batchmap.end())
663 // Warning cannot write on one line like this, because map object is constructed before
664 // the method is called...
665 //_batchmap[job_id] = FactoryBatchManager(resource_definition);
666 Batch::BatchManager * batch_client = FactoryBatchManager(resource_definition);
667 _batchmap[job_id] = batch_client;
669 catch(const LauncherException &ex)
671 LAUNCHER_INFOS("Error during creation of the batch manager of the job, mess: " << ex.msg);
674 catch(const Batch::GenericException &ex)
676 LAUNCHER_INFOS("Error during creation of the batch manager of the job, mess: " << ex.message);
677 throw LauncherException(ex.message);
684 Launcher_cpp::addJobDirectlyToMap(Launcher::Job * new_job)
686 // Step 0: Calculated job_id
687 pthread_mutex_lock(_job_cpt_mutex);
688 int job_id = _job_cpt;
690 new_job->setNumber(job_id);
691 pthread_mutex_unlock(_job_cpt_mutex);
693 // Step 1: check if resource is already in the map
694 createBatchManagerForJob(new_job);
696 // Step 2: add the job to the batch manager
700 Batch::JobId batch_manager_job_id = _batchmap[job_id]->addJob(*(new_job->getBatchJob()),
701 new_job->getReference());
702 new_job->setBatchManagerJobId(batch_manager_job_id);
704 catch(const Batch::GenericException &ex)
706 LAUNCHER_INFOS("Job cannot be added, exception in addJob: " << ex.message);
707 throw LauncherException(ex.message.c_str());
710 // Step 3: add job to launcher map
711 std::map<int, Launcher::Job *>::const_iterator it_job = _launcher_job_map.find(new_job->getNumber());
712 if (it_job == _launcher_job_map.end())
713 _launcher_job_map[new_job->getNumber()] = new_job;
716 LAUNCHER_INFOS("A job as already the same id: " << new_job->getNumber());
718 throw LauncherException("A job as already the same id - job is not created !");
720 LAUNCHER_MESSAGE("New job added");
725 Launcher_cpp::loadJobs(const char* jobs_file)
727 list<int> new_jobs_id_list;
729 // Load the jobs from XML file
730 list<Launcher::Job *> jobs_list = Launcher::XML_Persistence::loadJobs(jobs_file);
732 // Create each job in the launcher
733 list<Launcher::Job *>::const_iterator it_job;
734 for (it_job = jobs_list.begin(); it_job != jobs_list.end(); it_job++)
736 Launcher::Job * new_job = *it_job;
737 string job_state = new_job->getState();
741 if (job_state == "CREATED")
743 // In this case, we ignore run_part informations
745 new_jobs_id_list.push_back(new_job->getNumber());
747 else if (job_state == "QUEUED" ||
748 job_state == "RUNNING" ||
749 job_state == "IN_PROCESS" ||
750 job_state == "PAUSED")
752 addJobDirectlyToMap(new_job);
753 new_jobs_id_list.push_back(new_job->getNumber());
755 // Step 4: We check that the BatchManager could resume
758 if (new_job->getBatchManagerJobId().getReference() != new_job->getReference())
760 LAUNCHER_INFOS("BatchManager type cannot resume a job - job state is set to ERROR");
761 new_job->setState("ERROR");
765 else if (job_state == "FINISHED" ||
766 job_state == "FAILED" ||
767 job_state == "ERROR")
769 // Step 2: We add run_part informations
770 addJobDirectlyToMap(new_job);
771 new_jobs_id_list.push_back(new_job->getNumber());
775 LAUNCHER_INFOS("A bad job is found, state unknown " << job_state);
779 catch(const LauncherException &ex)
781 LAUNCHER_INFOS("Cannot load the job. Exception: " << ex.msg.c_str());
786 return new_jobs_id_list;
790 Launcher_cpp::saveJobs(const char* jobs_file)
792 // Create a sorted list from the internal job map
793 list<const Launcher::Job *> jobs_list;
794 for (int i=0; i<_job_cpt; i++)
796 map<int, Launcher::Job *>::const_iterator it_job = _launcher_job_map.find(i);
797 if (it_job != _launcher_job_map.end())
798 jobs_list.push_back(it_job->second);
801 // Save the jobs in XML file
802 Launcher::XML_Persistence::saveJobs(jobs_file, jobs_list);