ATLAS Offline Software
Loading...
Searching...
No Matches
SharedWriterTool Class Referencefinalabstract

#include <SharedWriterTool.h>

Inheritance diagram for SharedWriterTool:
Collaboration diagram for SharedWriterTool:

Public Member Functions

 SharedWriterTool (const std::string &type, const std::string &name, const IInterface *parent)
virtual ~SharedWriterTool () override
virtual StatusCode initialize () override
virtual StatusCode finalize () override
virtual int makePool ATLAS_NOT_THREAD_SAFE (int maxevt, int nprocs, const std::string &topdir) override
virtual StatusCode exec ATLAS_NOT_THREAD_SAFE () override
virtual void subProcessLogs (std::vector< std::string > &) override
virtual AthenaMP::AllWorkerOutputs_ptr generateOutputReport () override
virtual std::unique_ptr< AthenaInterprocess::ScheduledWorkbootstrap_func () override
virtual std::unique_ptr< AthenaInterprocess::ScheduledWorkexec_func () override
virtual std::unique_ptr< AthenaInterprocess::ScheduledWorkfin_func () override
virtual StatusCode wait_once ATLAS_NOT_THREAD_SAFE (pid_t &pid) override
virtual void reportSubprocessStatuses () override
virtual void useFdsRegistry (std::shared_ptr< AthenaInterprocess::FdsRegistry >) override
virtual void setRandString (const std::string &randStr) override
virtual void setMaxEvt (int maxEvt) override
virtual void setMPRunStop (const AthenaInterprocess::IMPRunStop *runStop) override
virtual void killChildren () override
virtual std::unique_ptr< ScheduledWork > operator () ATLAS_NOT_THREAD_SAFE(const ScheduledWork &)=0

Protected Types

enum  ESRange_Status {
  ESRANGE_SUCCESS , ESRANGE_NOTFOUND , ESRANGE_SEEKFAILED , ESRANGE_PROCFAILED ,
  ESRANGE_FILENOTMADE , ESRANGE_BADINPFILE
}
enum  Func_Flag { FUNC_BOOTSTRAP , FUNC_EXEC , FUNC_FIN }

Protected Member Functions

int mapAsyncFlag ATLAS_NOT_THREAD_SAFE (Func_Flag flag, pid_t pid=0)
int redirectLog (const std::string &rundir, bool addTimeStamp=true)
int updateIoReg (const std::string &rundir)
std::string fmterror (int errnum)
int reopenFds ()
int handleSavedPfc (const std::filesystem::path &dest_path)
void waitForSignal ()
IEvtSelector * evtSelector ()

Protected Attributes

int m_nprocs {-1}
 Number of workers spawned by the master process.
int m_maxEvt {-1}
 Maximum number of events assigned to the job.
std::string m_subprocTopDir
 Top run directory for subprocesses.
std::string m_subprocDirPrefix
 For ex. "worker__".
std::string m_evtSelName
 Name of the event selector.
AthenaInterprocess::ProcessGroupm_processGroup {nullptr}
const AthenaInterprocess::IMPRunStopm_mpRunStop {nullptr}
ServiceHandle< IEventProcessor > m_evtProcessor
ServiceHandle< IAppMgrUI > m_appMgr
ServiceHandle< IFileMgr > m_fileMgr
ServiceHandle< IIoComponentMgr > m_ioMgr
SmartIF< IEvtSelector > m_evtSelector
std::string m_fileMgrLog
std::shared_ptr< AthenaInterprocess::FdsRegistrym_fdsRegistry
std::string m_randStr
Gaudi::Property< bool > m_isPileup {this, "IsPileup", false, "Flag for configuring PileUpEventLoopMgr"}

Private Member Functions

 SharedWriterTool ()
 SharedWriterTool (const SharedWriterTool &)
SharedWriterTooloperator= (const SharedWriterTool &)
int reopenFd (int fd, const std::string &name)

Private Attributes

Gaudi::Property< bool > m_nMotherProcess
Gaudi::Property< bool > m_debug
int m_rankId
AthenaInterprocess::SharedQueuem_sharedRankQueue
SmartIF< IConversionSvc > m_cnvSvc

Detailed Description

Definition at line 12 of file SharedWriterTool.h.

Member Enumeration Documentation

◆ ESRange_Status

enum AthenaMPToolBase::ESRange_Status
protectedinherited
Enumerator
ESRANGE_SUCCESS 
ESRANGE_NOTFOUND 
ESRANGE_SEEKFAILED 
ESRANGE_PROCFAILED 
ESRANGE_FILENOTMADE 
ESRANGE_BADINPFILE 

Definition at line 58 of file AthenaMPToolBase.h.

◆ Func_Flag

enum AthenaMPToolBase::Func_Flag
protectedinherited
Enumerator
FUNC_BOOTSTRAP 
FUNC_EXEC 
FUNC_FIN 

Definition at line 67 of file AthenaMPToolBase.h.

Constructor & Destructor Documentation

◆ SharedWriterTool() [1/3]

SharedWriterTool::SharedWriterTool ( const std::string & type,
const std::string & name,
const IInterface * parent )

Definition at line 19 of file SharedWriterTool.cxx.

22 : AthenaMPToolBase(type,name,parent)
23 , m_rankId(0)
24 , m_sharedRankQueue(nullptr)
25{
26 m_subprocDirPrefix = "shared_writer";
27}
std::string m_subprocDirPrefix
For ex. "worker__".
AthenaInterprocess::SharedQueue * m_sharedRankQueue

◆ ~SharedWriterTool()

SharedWriterTool::~SharedWriterTool ( )
overridevirtual

Definition at line 29 of file SharedWriterTool.cxx.

30{
31}

◆ SharedWriterTool() [2/3]

SharedWriterTool::SharedWriterTool ( )
private

◆ SharedWriterTool() [3/3]

SharedWriterTool::SharedWriterTool ( const SharedWriterTool & )
private

Member Function Documentation

◆ ATLAS_NOT_THREAD_SAFE() [1/4]

int mapAsyncFlag AthenaMPToolBase::ATLAS_NOT_THREAD_SAFE ( Func_Flag flag,
pid_t pid = 0 )
protectedinherited

◆ ATLAS_NOT_THREAD_SAFE() [2/4]

virtual StatusCode wait_once AthenaMPToolBase::ATLAS_NOT_THREAD_SAFE ( pid_t & pid)
overridevirtualinherited

◆ ATLAS_NOT_THREAD_SAFE() [3/4]

virtual StatusCode exec SharedWriterTool::ATLAS_NOT_THREAD_SAFE ( )
overridevirtual

◆ ATLAS_NOT_THREAD_SAFE() [4/4]

virtual int makePool SharedWriterTool::ATLAS_NOT_THREAD_SAFE ( int maxevt,
int nprocs,
const std::string & topdir )
overridevirtual

◆ bootstrap_func()

std::unique_ptr< AthenaInterprocess::ScheduledWork > SharedWriterTool::bootstrap_func ( )
overridevirtual

Implements AthenaMPToolBase.

Definition at line 135 of file SharedWriterTool.cxx.

136{
137 // It's possible to debug SharedWriter just like any other AthenaMP worker.
138 // The following procedure provides a minimal explanation on how this can be achieved:
139 //
140 // Terminal #1:
141 // * Run athena w/ debugging enabled, e.g. athena.py --debugWorker --stdcmalloc --nprocs=8 [...]
142 // * In this mode, workers will be stopped after fork(), waiting for SIGUSR1 to be resumed
143 // * Find the PID of the worker to be debugged (printed by the job in stdout)
144 //
145 // Terminal #2:
146 // * Attach gdb to the relevant worker, i.e. gdb python PID
147 // * Once the symbols are loaded, one can perform any gdb action such as setting breakpoints etc.
148 // * Once ready, send SIGUSR1 to the worker to resume work, i.e. signal SIGUSR1 (in gdb)
149 //
150 // Terminal #3:
151 // * Send SIGUSR1 to the remaining workers (easiest to use htop)
152 //
153 // However, note that sometimes Shared I/O infrastructure struggles with timing problems,
154 // such as server/client(s) starting/stopping too early/later. Debugging can change this
155 // behavior so please keep this in mind.
157
158 std::unique_ptr<AthenaInterprocess::ScheduledWork> outwork(new AthenaInterprocess::ScheduledWork);
159 outwork->data = CxxUtils::xmalloc(sizeof(int));
160 *(int*)(outwork->data) = 1; // Error code: for now use 0 success, 1 failure
161 outwork->size = sizeof(int);
162
163 // ...
164 // (possible) TODO: extend outwork with some error message, which will be eventually
165 // reported in the master proces
166 // ...
167
168 // ________________________ Get RankID ________________________
169 //
170 if(!m_sharedRankQueue->receive_basic<int>(m_rankId)) {
171 ATH_MSG_ERROR("Unable to get rank ID!");
172 return outwork;
173 }
174 // Writer dir: mkdir
175 std::filesystem::path writer_rundir(m_subprocTopDir);
176 writer_rundir /= std::filesystem::path(m_subprocDirPrefix);
177
178 if(mkdir(writer_rundir.string().c_str(),S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH)==-1) {
179 ATH_MSG_ERROR("Unable to make writer run directory: " << writer_rundir.string() << ". " << fmterror(errno));
180 return outwork;
181 }
182
183 // __________ Redirect logs unless we want to attach debugger ____________
184 if(!m_debug) {
185 if(redirectLog(writer_rundir.string()))
186 return outwork;
187
188 ATH_MSG_INFO("Logs redirected in the AthenaMP Shared Writer PID=" << getpid());
189 }
190
191 // Update Io Registry
192 if(updateIoReg(writer_rundir.string()))
193 return outwork;
194
195 ATH_MSG_INFO("Io registry updated in the AthenaMP Shared Writer PID=" << getpid());
196
197 // _______________________ Handle saved PFC (if any) ______________________
198 std::filesystem::path abs_writer_rundir = std::filesystem::absolute(writer_rundir);
199 if(handleSavedPfc(abs_writer_rundir))
200 return outwork;
201
202 // Reopen file descriptors
203 if(reopenFds())
204 return outwork;
205
206 ATH_MSG_INFO("File descriptors re-opened in the AthenaMP Shared Writer PID=" << getpid());
207
208 // Try to initialize AthenaRootSharedWriterSvc early on
209 SmartIF<IAthenaSharedWriterSvc> sharedWriterSvc(serviceLocator()->service("AthenaRootSharedWriterSvc"));
210 if(!sharedWriterSvc) {
211 ATH_MSG_WARNING("Error retrieving AthenaRootSharedWriterSvc from SharedWriterTool::bootstrap_func()");
212 }
213
214 // Use IDataShare to make ConversionSvc a Share Server
215 SmartIF<IDataShare> cnvSvc(m_cnvSvc);
216 if (!cnvSvc || !cnvSvc->makeServer(-m_nprocs - 1 - 1024 * m_rankId).isSuccess()) {
217 ATH_MSG_ERROR("Failed to make the conversion service a share server");
218 return outwork;
219 }
220 else {
221 ATH_MSG_DEBUG("Successfully made the conversion service a share server");
222 }
223
224 // ________________________ I/O reinit ________________________
225 if(!m_ioMgr->io_reinitialize().isSuccess()) {
226 ATH_MSG_ERROR("Failed to reinitialize I/O");
227 return outwork;
228 } else {
229 ATH_MSG_DEBUG("Successfully reinitialized I/O");
230 }
231
232 // Writer dir: chdir
233 if(chdir(writer_rundir.string().c_str())==-1) {
234 ATH_MSG_ERROR("Failed to chdir to " << writer_rundir.string());
235 return outwork;
236 }
237
238 // Declare success and return
239 *(int*)(outwork->data) = 0;
240 return outwork;
241}
#define ATH_MSG_ERROR(x)
#define ATH_MSG_INFO(x)
#define ATH_MSG_WARNING(x)
#define ATH_MSG_DEBUG(x)
std::string m_subprocTopDir
Top run directory for subprocesses.
int handleSavedPfc(const std::filesystem::path &dest_path)
int updateIoReg(const std::string &rundir)
int redirectLog(const std::string &rundir, bool addTimeStamp=true)
int m_nprocs
Number of workers spawned by the master process.
ServiceHandle< IIoComponentMgr > m_ioMgr
std::string fmterror(int errnum)
SmartIF< IConversionSvc > m_cnvSvc
Gaudi::Property< bool > m_debug
void * xmalloc(size_t size)
Trapping version of malloc.
Definition xmalloc.cxx:31
mkdir(path, recursive=True)

◆ evtSelector()

IEvtSelector * AthenaMPToolBase::evtSelector ( )
inlineprotectedinherited

Definition at line 83 of file AthenaMPToolBase.h.

83{ return m_evtSelector; }
SmartIF< IEvtSelector > m_evtSelector

◆ exec_func()

std::unique_ptr< AthenaInterprocess::ScheduledWork > SharedWriterTool::exec_func ( )
overridevirtual

Implements AthenaMPToolBase.

Definition at line 243 of file SharedWriterTool.cxx.

244{
245 ATH_MSG_INFO("Exec function in the AthenaMP Shared Writer PID=" << getpid());
246 bool all_ok=true;
247
248 SmartIF<IAthenaSharedWriterSvc> sharedWriterSvc(serviceLocator()->service("AthenaRootSharedWriterSvc"));
249 if(!sharedWriterSvc) {
250 ATH_MSG_ERROR("Error retrieving AthenaRootSharedWriterSvc");
251 all_ok=false;
252 } else if(!sharedWriterSvc->share(m_nprocs, m_nMotherProcess.value()).isSuccess()) {
253 ATH_MSG_ERROR("Exec function could not share data");
254 all_ok=false;
255 }
256 AthCnvSvc* cnvSvc = dynamic_cast<AthCnvSvc*>(m_cnvSvc.get());
257 if (cnvSvc == 0 || !cnvSvc->disconnectOutput("").isSuccess()) {
258 ATH_MSG_ERROR("Exec function could not disconnectOutput");
259 all_ok=false;
260 }
261
262 if(m_appMgr->stop().isFailure()) {
263 ATH_MSG_ERROR("Unable to stop AppMgr");
264 all_ok=false;
265 }
266 else {
267 if(m_appMgr->finalize().isFailure()) {
268 std::cerr << "Unable to finalize AppMgr" << std::endl;
269 all_ok=false;
270 }
271 }
272
273 std::unique_ptr<AthenaInterprocess::ScheduledWork> outwork(new AthenaInterprocess::ScheduledWork);
274 outwork->data = CxxUtils::xmalloc(sizeof(int));
275 *(int*)(outwork->data) = (all_ok?0:1); // Error code: for now use 0 success, 1 failure
276 outwork->size = sizeof(int);
277
278 // ...
279 // (possible) TODO: extend outwork with some error message, which will be eventually
280 // reported in the master proces
281 // ...
282 return outwork;
283}
virtual StatusCode disconnectOutput(const std::string &output)
Disconnect output files from the service.
ServiceHandle< IAppMgrUI > m_appMgr
Gaudi::Property< bool > m_nMotherProcess

◆ fin_func()

std::unique_ptr< AthenaInterprocess::ScheduledWork > SharedWriterTool::fin_func ( )
overridevirtual

Implements AthenaMPToolBase.

Definition at line 285 of file SharedWriterTool.cxx.

286{
287 // Dummy
288 std::unique_ptr<AthenaInterprocess::ScheduledWork> outwork(new AthenaInterprocess::ScheduledWork);
289 outwork->data = CxxUtils::xmalloc(sizeof(int));
290 *(int*)(outwork->data) = 0; // Error code: for now use 0 success, 1 failure
291 outwork->size = sizeof(int);
292 return outwork;
293}

◆ finalize()

StatusCode SharedWriterTool::finalize ( )
overridevirtual

Reimplemented from AthenaMPToolBase.

Definition at line 44 of file SharedWriterTool.cxx.

45{
46 ATH_MSG_DEBUG("In finalize");
47
48 delete m_sharedRankQueue;
49 return StatusCode::SUCCESS;
50}

◆ fmterror()

std::string AthenaMPToolBase::fmterror ( int errnum)
protectedinherited

Definition at line 333 of file AthenaMPToolBase.cxx.

334{
335 char buf[256];
336 strerror_r(errnum, buf, sizeof(buf));
337 return std::string(buf);
338}

◆ generateOutputReport()

AthenaMP::AllWorkerOutputs_ptr SharedWriterTool::generateOutputReport ( )
overridevirtual

Reimplemented from AthenaMPToolBase.

Definition at line 129 of file SharedWriterTool.cxx.

130{
132 return jobOutputs;
133}
std::map< std::string, SingleWorkerOutputs > AllWorkerOutputs
std::unique_ptr< AllWorkerOutputs > AllWorkerOutputs_ptr

◆ handleSavedPfc()

int AthenaMPToolBase::handleSavedPfc ( const std::filesystem::path & dest_path)
protectedinherited

Definition at line 396 of file AthenaMPToolBase.cxx.

397{
398 if(std::filesystem::is_regular_file("PoolFileCatalog.xml.AthenaMP-saved"))
399 COPY_FILE_HACK("PoolFileCatalog.xml.AthenaMP-saved",dest_path.string()+"/PoolFileCatalog.xml");
400 return 0;
401}
#define COPY_FILE_HACK(_src, _dest)

◆ initialize()

StatusCode SharedWriterTool::initialize ( )
overridevirtual

Reimplemented from AthenaMPToolBase.

Definition at line 33 of file SharedWriterTool.cxx.

34{
35 ATH_MSG_DEBUG("In initialize");
36
38 m_cnvSvc = serviceLocator()->service("AthenaPoolSharedIOCnvSvc");
39 ATH_CHECK(m_cnvSvc.isValid());
40
41 return StatusCode::SUCCESS;
42}
#define ATH_CHECK
Evaluate an expression and check for errors.
virtual StatusCode initialize() override

◆ killChildren()

void AthenaMPToolBase::killChildren ( )
overridevirtualinherited

Definition at line 201 of file AthenaMPToolBase.cxx.

202{
203 for(const AthenaInterprocess::Process& child : m_processGroup->getChildren()) {
204 kill(child.getProcessID(),SIGKILL);
205 }
206}
AthenaInterprocess::ProcessGroup * m_processGroup

◆ operator()

virtual std::unique_ptr< ScheduledWork > AthenaInterprocess::IMessageDecoder::operator ( ) const &
pure virtualinherited

◆ operator=()

◆ redirectLog()

int AthenaMPToolBase::redirectLog ( const std::string & rundir,
bool addTimeStamp = true )
protectedinherited

Definition at line 269 of file AthenaMPToolBase.cxx.

270{
271 // Redirect both stdout and stderr to the same file AthenaMP.log
272 int dup2result1(0), dup2result2(0);
273
274 int newout = open(std::string(rundir+"/AthenaMP.log").c_str(),O_CREAT | O_RDWR, S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH);
275 if(newout==-1) {
276 ATH_MSG_ERROR("Unable to open log file in the run directory. " << fmterror(errno));
277 return -1;
278 }
279 dup2result1 = dup2(newout, STDOUT_FILENO);
280 dup2result2 = dup2(newout, STDERR_FILENO);
281 TEMP_FAILURE_RETRY(close(newout));
282 if(dup2result1==-1) {
283 ATH_MSG_ERROR("Unable to redirect standard output. " << fmterror(errno));
284 return -1;
285 }
286 if(dup2result2==-1) {
287 ATH_MSG_ERROR("Unable to redirect standard error. " << fmterror(errno));
288 return -1;
289 }
290
291 if(addTimeStamp) {
292 SmartIF<IProperty> propertyServer(msgSvc());
293 if(propertyServer==0) {
294 ATH_MSG_ERROR("Unable to cast message svc to IProperty");
295 return -1;
296 }
297
298 std::string propertyName("Format");
299 std::string oldFormat("");
300 StringProperty formatProp(propertyName,oldFormat);
301 StatusCode sc = propertyServer->getProperty(&formatProp);
302 if(sc.isFailure()) {
303 ATH_MSG_WARNING("Message Service does not have Format property");
304 }
305 else {
306 oldFormat = formatProp.value();
307 if(oldFormat.find("%t")==std::string::npos) {
308 // Add time stamps
309 std::string newFormat("%t " + oldFormat);
310 StringProperty newFormatProp(std::move(propertyName),newFormat);
311 ATH_CHECK(propertyServer->setProperty(newFormatProp), -1);
312 }
313 else {
314 ATH_MSG_DEBUG("MsgSvc format already contains timestamps. Nothing to be done");
315 }
316 }
317 }
318
319 return 0;
320}
static Double_t sc
::StatusCode StatusCode
StatusCode definition for legacy code.
msgSvc
Provide convenience handles for various services.
Definition StdJOSetup.py:36
@ open
Definition BinningType.h:40

◆ reopenFd()

int AthenaMPToolBase::reopenFd ( int fd,
const std::string & name )
privateinherited

Definition at line 419 of file AthenaMPToolBase.cxx.

420{
421 ATH_MSG_DEBUG("Attempting to reopen descriptor for " << name);
422 int old_openflags = fcntl(fd,F_GETFL,0);
423 switch(old_openflags & O_ACCMODE) {
424 case O_RDONLY: {
425 ATH_MSG_DEBUG("The File Access Mode is RDONLY");
426 break;
427 }
428 case O_WRONLY: {
429 ATH_MSG_DEBUG("The File Access Mode is WRONLY");
430 break;
431 }
432 case O_RDWR: {
433 ATH_MSG_DEBUG("The File Access Mode is RDWR");
434 break;
435 }
436 }
437
438 int old_descflags = fcntl(fd,F_GETFD,0);
439 off_t oldpos = lseek(fd,0,SEEK_CUR);
440 if(oldpos==-1) {
441 if(errno==ESPIPE) {
442 ATH_MSG_WARNING("Dealing with PIPE. Skipping ... (FIXME!)");
443 }
444 else {
445 ATH_MSG_ERROR("When re-opening file descriptors lseek failed on " << name << ". " << fmterror(errno));
446 return -1;
447 }
448 }
449 else {
450 Io::Fd newfd = open(name.c_str(),old_openflags);
451 if(newfd==-1) {
452 ATH_MSG_ERROR("When re-opening file descriptors unable to open " << name << " for reading. " << fmterror(errno));
453 return -1;
454 }
455 if(lseek(newfd,oldpos,SEEK_SET)==-1){
456 ATH_MSG_ERROR("When re-opening file descriptors lseek failed on the newly opened " << name << ". " << fmterror(errno));
457 TEMP_FAILURE_RETRY(close(newfd));
458 return -1;
459 }
460 TEMP_FAILURE_RETRY(close(fd));
461 if(dup2(newfd,fd)==-1) {
462 ATH_MSG_ERROR("When re-opening file descriptors unable to duplicate descriptor for " << name << ". " << fmterror(errno));
463 TEMP_FAILURE_RETRY(close(newfd));
464 return -1;
465 }
466 if(fcntl(fd,F_SETFD,old_descflags)==-1) {
467 ATH_MSG_ERROR("When re-opening file descriptors unable to set descriptor flags for " << name << ". " << fmterror(errno));
468 TEMP_FAILURE_RETRY(close(newfd));
469 return -1;
470 }
471 TEMP_FAILURE_RETRY(close(newfd));
472 }
473 return 0;
474}

◆ reopenFds()

int AthenaMPToolBase::reopenFds ( )
protectedinherited

Definition at line 340 of file AthenaMPToolBase.cxx.

341{
342 // Reopen file descriptors.
343 // First go over all open files, which have been registered with the FileMgr
344 // Then also check the FdsRegistry, in case it contains some files not registered with the FileMgr
345 std::set<int> fdLog;
346
347 // Query the FileMgr contents
348 std::vector<const Io::FileAttr*> filemgrFiles;
349 std::vector<const Io::FileAttr*>::const_iterator itFile;
350 unsigned filenum = m_fileMgr->getFiles(filemgrFiles); // Get attributes for open files only. We don't care about closed ones at this point
351 if(filenum!=filemgrFiles.size())
352 ATH_MSG_WARNING("getFiles returned " << filenum << " while vector size is " << filemgrFiles.size());
353
354 for(itFile=filemgrFiles.begin();itFile!=filemgrFiles.end();++itFile) {
355 ATH_MSG_DEBUG("* " << **itFile);
356 const std::string& filename = (**itFile).name();
357 Io::Fd fd = (**itFile).fd();
358
359 if(fd==-1) {
360 // It is legal to have fd=-1 for remote inputs
361 // On the other hand, these inputs should not remain open after fork. The issue being tracked at ATEAM-434.
362 // So, this hopefully is a temporary patch
363 ATH_MSG_WARNING("FD=-1 detected on an open file retrieved from FileMgr. Skip FD reopening. File name: " << filename);
364 continue;
365 }
366
367 if(reopenFd(fd,filename))
368 return -1;
369
370 fdLog.insert(fd);
371 }
372
373 // Check the FdsRegistry
374 for(const AthenaInterprocess::FdsRegistryEntry& regEntry : *m_fdsRegistry) {
375 if(fdLog.find(regEntry.fd)!=fdLog.end()) {
376 ATH_MSG_DEBUG("The file from FdsRegistry " << regEntry.name << " was registered with FileMgr. Skip reopening");
377 }
378 else {
379 ATH_MSG_WARNING("The file " << regEntry.name << " has not been registered with the FileMgr!");
380
381 if(regEntry.fd==-1) {
382 // Same protection as the one above
383 ATH_MSG_WARNING("FD=-1 detected on an open file retrieved from FD Registry. Skip FD reopening. File name: " << regEntry.name);
384 continue;
385 }
386
387 if(reopenFd(regEntry.fd,regEntry.name))
388 return -1;
389
390 fdLog.insert(regEntry.fd);
391 }
392 }
393 return 0;
394}
int reopenFd(int fd, const std::string &name)
std::shared_ptr< AthenaInterprocess::FdsRegistry > m_fdsRegistry
ServiceHandle< IFileMgr > m_fileMgr

◆ reportSubprocessStatuses()

void AthenaMPToolBase::reportSubprocessStatuses ( )
overridevirtualinherited

Reimplemented in EvtRangeProcessor, SharedEvtQueueConsumer, and SharedHiveEvtQueueConsumer.

Definition at line 111 of file AthenaMPToolBase.cxx.

112{
113 ATH_MSG_INFO("Statuses of sub-processes");
114 const std::vector<AthenaInterprocess::ProcessStatus>& statuses = m_processGroup->getStatuses();
115 for(size_t i=0; i<statuses.size(); ++i)
116 ATH_MSG_INFO("*** Process PID=" << statuses[i].pid << ". Status " << ((statuses[i].exitcode)?"FAILURE":"SUCCESS"));
117}
list statuses

◆ setMaxEvt()

virtual void AthenaMPToolBase::setMaxEvt ( int maxEvt)
inlineoverridevirtualinherited

Definition at line 44 of file AthenaMPToolBase.h.

44{m_maxEvt=maxEvt;}
int m_maxEvt
Maximum number of events assigned to the job.

◆ setMPRunStop()

virtual void AthenaMPToolBase::setMPRunStop ( const AthenaInterprocess::IMPRunStop * runStop)
inlineoverridevirtualinherited

Definition at line 45 of file AthenaMPToolBase.h.

45{m_mpRunStop=runStop;}
const AthenaInterprocess::IMPRunStop * m_mpRunStop

◆ setRandString()

void AthenaMPToolBase::setRandString ( const std::string & randStr)
overridevirtualinherited

Definition at line 196 of file AthenaMPToolBase.cxx.

197{
198 m_randStr = randStr;
199}

◆ subProcessLogs()

void SharedWriterTool::subProcessLogs ( std::vector< std::string > & filenames)
overridevirtual

Definition at line 121 of file SharedWriterTool.cxx.

122{
123 filenames.clear();
124 std::filesystem::path writer_rundir(m_subprocTopDir);
125 writer_rundir/= std::filesystem::path(m_subprocDirPrefix);
126 filenames.push_back(writer_rundir.string()+std::string("/AthenaMP.log"));
127}
list filenames
Definition grepfile.py:34

◆ updateIoReg()

int AthenaMPToolBase::updateIoReg ( const std::string & rundir)
protectedinherited

Definition at line 322 of file AthenaMPToolBase.cxx.

323{
324 ATH_CHECK(m_ioMgr.retrieve(), -1);
325
326 // update the IoRegistry for the new workdir - make sure we use absolute path
327 std::filesystem::path abs_rundir = std::filesystem::absolute(rundir);
328 ATH_CHECK(m_ioMgr->io_update_all(abs_rundir.string()), -1);
329
330 return 0;
331}

◆ useFdsRegistry()

void AthenaMPToolBase::useFdsRegistry ( std::shared_ptr< AthenaInterprocess::FdsRegistry > registry)
overridevirtualinherited

Definition at line 191 of file AthenaMPToolBase.cxx.

192{
193 m_fdsRegistry = std::move(registry);
194}

◆ waitForSignal()

void AthenaMPToolBase::waitForSignal ( )
protectedinherited

Definition at line 403 of file AthenaMPToolBase.cxx.

404{
405 ATH_MSG_INFO("Bootstrap worker PID " << getpid() << " - waiting for SIGUSR1");
406 sigset_t mask, oldmask;
407
409
410 sigemptyset (&mask);
411 sigaddset (&mask, SIGUSR1);
412
413 sigprocmask (SIG_BLOCK, &mask, &oldmask);
415 sigsuspend (&oldmask);
416 sigprocmask (SIG_UNBLOCK, &mask, NULL);
417}
#define sigemptyset(x)
Definition SealSignal.h:82
#define sigaddset(x, y)
Definition SealSignal.h:84
int sigset_t
Definition SealSignal.h:80
std::atomic< bool > sig_done

Member Data Documentation

◆ m_appMgr

ServiceHandle<IAppMgrUI> AthenaMPToolBase::m_appMgr
protectedinherited

Definition at line 95 of file AthenaMPToolBase.h.

◆ m_cnvSvc

SmartIF<IConversionSvc> SharedWriterTool::m_cnvSvc
private

Definition at line 52 of file SharedWriterTool.h.

◆ m_debug

Gaudi::Property<bool> SharedWriterTool::m_debug
private
Initial value:
{
this, "Debug", false,
"Are we running in debug mode? The default is false"}

Definition at line 45 of file SharedWriterTool.h.

45 {
46 this, "Debug", false,
47 "Are we running in debug mode? The default is false"};

◆ m_evtProcessor

ServiceHandle<IEventProcessor> AthenaMPToolBase::m_evtProcessor
protectedinherited

Definition at line 94 of file AthenaMPToolBase.h.

◆ m_evtSelector

SmartIF<IEvtSelector> AthenaMPToolBase::m_evtSelector
protectedinherited

Definition at line 98 of file AthenaMPToolBase.h.

◆ m_evtSelName

std::string AthenaMPToolBase::m_evtSelName
protectedinherited

Name of the event selector.

Definition at line 89 of file AthenaMPToolBase.h.

◆ m_fdsRegistry

std::shared_ptr<AthenaInterprocess::FdsRegistry> AthenaMPToolBase::m_fdsRegistry
protectedinherited

Definition at line 100 of file AthenaMPToolBase.h.

◆ m_fileMgr

ServiceHandle<IFileMgr> AthenaMPToolBase::m_fileMgr
protectedinherited

Definition at line 96 of file AthenaMPToolBase.h.

◆ m_fileMgrLog

std::string AthenaMPToolBase::m_fileMgrLog
protectedinherited

Definition at line 99 of file AthenaMPToolBase.h.

◆ m_ioMgr

ServiceHandle<IIoComponentMgr> AthenaMPToolBase::m_ioMgr
protectedinherited

Definition at line 97 of file AthenaMPToolBase.h.

◆ m_isPileup

Gaudi::Property<bool> AthenaMPToolBase::m_isPileup {this, "IsPileup", false, "Flag for configuring PileUpEventLoopMgr"}
protectedinherited

Definition at line 103 of file AthenaMPToolBase.h.

103{this, "IsPileup", false, "Flag for configuring PileUpEventLoopMgr"};

◆ m_maxEvt

int AthenaMPToolBase::m_maxEvt {-1}
protectedinherited

Maximum number of events assigned to the job.

Definition at line 86 of file AthenaMPToolBase.h.

86{-1};

◆ m_mpRunStop

const AthenaInterprocess::IMPRunStop* AthenaMPToolBase::m_mpRunStop {nullptr}
protectedinherited

Definition at line 92 of file AthenaMPToolBase.h.

92{nullptr};

◆ m_nMotherProcess

Gaudi::Property<bool> SharedWriterTool::m_nMotherProcess
private
Initial value:
{
this, "MotherProcess", false,
"Expect mother process to write event data. The default is false."}

Definition at line 42 of file SharedWriterTool.h.

42 {
43 this, "MotherProcess", false,
44 "Expect mother process to write event data. The default is false."};

◆ m_nprocs

int AthenaMPToolBase::m_nprocs {-1}
protectedinherited

Number of workers spawned by the master process.

Definition at line 85 of file AthenaMPToolBase.h.

85{-1};

◆ m_processGroup

AthenaInterprocess::ProcessGroup* AthenaMPToolBase::m_processGroup {nullptr}
protectedinherited

Definition at line 91 of file AthenaMPToolBase.h.

91{nullptr};

◆ m_randStr

std::string AthenaMPToolBase::m_randStr
protectedinherited

Definition at line 101 of file AthenaMPToolBase.h.

◆ m_rankId

int SharedWriterTool::m_rankId
private

Definition at line 49 of file SharedWriterTool.h.

◆ m_sharedRankQueue

AthenaInterprocess::SharedQueue* SharedWriterTool::m_sharedRankQueue
private

Definition at line 51 of file SharedWriterTool.h.

◆ m_subprocDirPrefix

std::string AthenaMPToolBase::m_subprocDirPrefix
protectedinherited

For ex. "worker__".

Definition at line 88 of file AthenaMPToolBase.h.

◆ m_subprocTopDir

std::string AthenaMPToolBase::m_subprocTopDir
protectedinherited

Top run directory for subprocesses.

Definition at line 87 of file AthenaMPToolBase.h.


The documentation for this class was generated from the following files: