2 #include "artdaq/DAQrate/SharedMemoryEventManager.hh"
4 #include "artdaq-core/Core/StatisticsCollection.hh"
5 #include "artdaq-core/Utilities/TraceLock.hh"
7 #define TRACE_NAME (app_name + "_SharedMemoryEventManager").c_str()
10 #define TLVL_BUFLCK 41
12 #define build_key(seed) seed + ((GetPartitionNumber() + 1) << 16) + (getpid() & 0xFFFF)
14 std::mutex artdaq::SharedMemoryEventManager::sequence_id_mutex_;
15 std::mutex artdaq::SharedMemoryEventManager::subrun_event_map_mutex_;
22 : SharedMemoryManager(pset.get<uint32_t>(
"shared_memory_key", build_key(0xEE000000)),
23 pset.get<size_t>(
"buffer_count"),
24 pset.has_key(
"max_event_size_bytes") ? pset.get<size_t>(
"max_event_size_bytes") : pset.get<size_t>(
"expected_fragments_per_event") * pset.get<size_t>(
"max_fragment_size_bytes"),
25 pset.get<size_t>(
"stale_buffer_timeout_usec", pset.get<size_t>(
"event_queue_wait_time", 5) * 1000000),
26 !pset.get<bool>(
"broadcast_mode", false))
27 , num_art_processes_(pset.get<size_t>(
"art_analyzer_count", 1))
28 , num_fragments_per_event_(pset.get<size_t>(
"expected_fragments_per_event"))
29 , queue_size_(pset.get<size_t>(
"buffer_count"))
31 , max_subrun_event_map_length_(pset.get<size_t>(
"max_subrun_lookup_table_size", 100))
32 , max_event_list_length_(pset.get<size_t>(
"max_event_list_length", 100))
33 , update_run_ids_(pset.get<bool>(
"update_run_ids_on_new_fragment", true))
34 , use_sequence_id_for_event_number_(pset.get<bool>(
"use_sequence_id_for_event_number", true))
35 , overwrite_mode_(!pset.get<bool>(
"use_art", true) || pset.get<bool>(
"overwrite_mode", false) || pset.get<bool>(
"broadcast_mode", false))
36 , init_fragment_count_(pset.get<size_t>(
"init_fragment_count", pset.get<bool>(
"send_init_fragments", true) ? 1 : 0))
38 , buffer_writes_pending_()
39 , incomplete_event_report_interval_ms_(pset.get<int>(
"incomplete_event_report_interval_ms", -1))
40 , last_incomplete_event_report_time_(std::chrono::steady_clock::now())
41 , last_backpressure_report_time_(std::chrono::steady_clock::now())
42 , last_fragment_header_write_time_(std::chrono::steady_clock::now())
43 , broadcast_timeout_ms_(pset.get<int>(
"fragment_broadcast_timeout_ms", 3000))
45 , run_incomplete_event_count_(0)
46 , subrun_event_count_(0)
47 , subrun_incomplete_event_count_(0)
48 , oversize_fragment_count_(0)
49 , maximum_oversize_fragment_count_(pset.get<int>(
"maximum_oversize_fragment_count", 1))
52 , always_restart_art_(pset.get<bool>(
"restart_crashed_art_processes", true))
53 , manual_art_(pset.get<bool>(
"manual_art", false))
54 , current_art_pset_(art_pset)
55 , minimum_art_lifetime_s_(pset.get<double>(
"minimum_art_lifetime_s", 2.0))
56 , art_event_processing_time_us_(pset.get<size_t>(
"expected_art_event_processing_time_us", 1000000))
60 , broadcasts_(pset.get<uint32_t>(
"broadcast_shared_memory_key", build_key(0xBB000000)),
61 pset.get<size_t>(
"broadcast_buffer_count", 10),
62 pset.get<size_t>(
"broadcast_buffer_size", 0x100000),
63 pset.get<int>(
"expected_art_event_processing_time_us", 100000) * pset.get<size_t>(
"buffer_count"), false)
65 subrun_event_map_[0] = 1;
66 SetMinWriteSize(
sizeof(detail::RawEventHeader) +
sizeof(detail::RawFragmentHeader));
67 broadcasts_.SetMinWriteSize(
sizeof(detail::RawEventHeader) +
sizeof(detail::RawFragmentHeader));
69 if (pset.get<
bool>(
"use_art",
true) ==
false)
71 TLOG(TLVL_INFO) <<
"BEGIN SharedMemoryEventManager CONSTRUCTOR with use_art:false";
72 num_art_processes_ = 0;
76 TLOG(TLVL_INFO) <<
"BEGIN SharedMemoryEventManager CONSTRUCTOR with use_art:true";
77 TLOG(TLVL_TRACE) <<
"art_pset is " << art_pset.to_string();
79 current_art_config_file_ = std::make_shared<art_config_file>(art_pset );
81 if (overwrite_mode_ && num_art_processes_ > 0)
83 TLOG(TLVL_WARNING) <<
"Art is configured to run, but overwrite mode is enabled! Check your configuration if this in unintentional!";
85 else if (overwrite_mode_)
87 TLOG(TLVL_INFO) <<
"Overwrite Mode enabled, no configured art processes at startup";
90 for (
size_t ii = 0; ii < size(); ++ii)
92 buffer_writes_pending_[ii] = 0;
95 if (!IsValid())
throw cet::exception(app_name +
"_SharedMemoryEventManager") <<
"Unable to attach to Shared Memory!";
97 TLOG(TLVL_TRACE) <<
"Setting Writer rank to " << my_rank;
99 TLOG(TLVL_DEBUG) <<
"Writer Rank is " << GetRank();
107 TLOG(TLVL_TRACE) <<
"END CONSTRUCTOR";
112 TLOG(TLVL_TRACE) <<
"DESTRUCTOR";
113 if (running_) endOfData();
114 TLOG(TLVL_TRACE) <<
"Destructor END";
117 bool artdaq::SharedMemoryEventManager::AddFragment(detail::RawFragmentHeader frag,
void* dataPtr)
119 TLOG(TLVL_TRACE) <<
"AddFragment(Header, ptr) BEGIN frag.word_count=" << frag.word_count
120 <<
", sequence_id=" << frag.sequence_id;
121 auto buffer = getBufferForSequenceID_(frag.sequence_id,
true, frag.timestamp);
122 TLOG(TLVL_TRACE) <<
"Using buffer " << buffer <<
" for seqid=" << frag.sequence_id;
123 if (buffer == -1)
return false;
126 TLOG(TLVL_ERROR) <<
"Dropping event because data taking has already passed this event number: " << frag.sequence_id;
130 auto hdr = getEventHeader_(buffer);
133 hdr->run_id = run_id_;
135 hdr->subrun_id = GetSubrunForSequenceID(frag.sequence_id);
137 TLOG(TLVL_TRACE) <<
"AddFragment before Write calls";
138 Write(buffer, dataPtr, frag.word_count *
sizeof(RawDataType));
140 TLOG(TLVL_TRACE) <<
"Checking for complete event";
141 auto fragmentCount = GetFragmentCount(frag.sequence_id);
142 hdr->is_complete = fragmentCount == num_fragments_per_event_ && buffer_writes_pending_[buffer] == 0;
143 TLOG(TLVL_TRACE) <<
"hdr->is_complete=" << std::boolalpha << hdr->is_complete
144 <<
", fragmentCount=" << fragmentCount
145 <<
", num_fragments_per_event=" << num_fragments_per_event_
146 <<
", buffer_writes_pending_[buffer]=" << buffer_writes_pending_[buffer];
148 complete_buffer_(buffer);
149 if (requests_) requests_->SendRequest(
true);
151 TLOG(TLVL_TRACE) <<
"AddFragment END";
152 statsHelper_.addSample(FRAGMENTS_RECEIVED_STAT_KEY, frag.word_count *
sizeof(RawDataType));
156 bool artdaq::SharedMemoryEventManager::AddFragment(FragmentPtr frag,
size_t timeout_usec, FragmentPtr& outfrag)
158 TLOG(TLVL_TRACE) <<
"AddFragment(FragmentPtr) BEGIN";
159 auto hdr = *
reinterpret_cast<detail::RawFragmentHeader*
>(frag->headerAddress());
160 auto data = frag->headerAddress();
161 auto start = std::chrono::steady_clock::now();
163 while (!sts && TimeUtils::GetElapsedTimeMicroseconds(start) < timeout_usec)
165 sts = AddFragment(hdr, data);
166 if (!sts) usleep(1000);
170 outfrag = std::move(frag);
172 TLOG(TLVL_TRACE) <<
"AddFragment(FragmentPtr) RETURN " << std::boolalpha << sts;
178 TLOG(14) <<
"WriteFragmentHeader BEGIN";
179 auto buffer = getBufferForSequenceID_(frag.sequence_id,
true, frag.timestamp);
183 if (buffer == -1 && !dropIfNoBuffersAvailable)
185 std::unique_lock<std::mutex> bp_lk(sequence_id_mutex_);
186 if (TimeUtils::GetElapsedTime(last_backpressure_report_time_) > 1.0)
188 TLOG(TLVL_WARNING) << app_name <<
": Back-pressure condition: All Shared Memory buffers have been full for " << TimeUtils::GetElapsedTime(last_fragment_header_write_time_) <<
" s!";
189 last_backpressure_report_time_ = std::chrono::steady_clock::now();
195 TLOG(TLVL_ERROR) <<
"Dropping fragment with sequence id " << frag.sequence_id <<
" and fragment id " << frag.fragment_id <<
" because data taking has already passed this event.";
199 TLOG(TLVL_ERROR) <<
"Dropping fragment with sequence id " << frag.sequence_id <<
" and fragment id " << frag.fragment_id <<
" because there is no room in the queue and reliable mode is off.";
201 dropped_data_[frag.fragment_id].reset(
new Fragment(frag.word_count - frag.num_words()));
203 TLOG(6) <<
"Dropping fragment with sequence id " << frag.sequence_id <<
" and fragment id " << frag.fragment_id <<
" into " << (
void*)dropped_data_[frag.fragment_id]->dataBegin() <<
" sz=" << dropped_data_[frag.fragment_id]->dataSizeBytes();
204 return dropped_data_[frag.fragment_id]->dataBegin();
207 last_backpressure_report_time_ = std::chrono::steady_clock::now();
208 last_fragment_header_write_time_ = std::chrono::steady_clock::now();
210 buffer_writes_pending_[buffer]++;
214 metricMan->sendMetric(
"Input Fragment Rate", 1,
"Fragments/s", 1, MetricMode::Rate);
217 TLOG(TLVL_BUFLCK) <<
"WriteFragmentHeader: obtaining buffer_mutexes lock for buffer " << buffer;
219 std::unique_lock<std::mutex> lk(buffer_mutexes_[buffer]);
221 TLOG(TLVL_BUFLCK) <<
"WriteFragmentHeader: obtained buffer_mutexes lock for buffer " << buffer;
224 auto hdrpos =
reinterpret_cast<RawDataType*
>(GetWritePos(buffer));
225 Write(buffer, &frag, frag.num_words() *
sizeof(RawDataType));
227 auto pos =
reinterpret_cast<RawDataType*
>(GetWritePos(buffer));
228 if (frag.word_count - frag.num_words() > 0)
230 auto sts = IncrementWritePos(buffer, (frag.word_count - frag.num_words()) *
sizeof(RawDataType));
234 reinterpret_cast<detail::RawFragmentHeader*
>(hdrpos)->word_count = frag.num_words();
235 reinterpret_cast<detail::RawFragmentHeader*
>(hdrpos)->type = Fragment::InvalidFragmentType;
236 TLOG(TLVL_ERROR) <<
"Dropping over-size fragment with sequence id " << frag.sequence_id <<
" and fragment id " << frag.fragment_id <<
" because there is no room in the current buffer for this Fragment! (Keeping header)";
237 dropped_data_[frag.fragment_id].reset(
new Fragment(frag.word_count - frag.num_words()));
239 oversize_fragment_count_++;
241 if (maximum_oversize_fragment_count_ > 0 && oversize_fragment_count_ >= maximum_oversize_fragment_count_)
243 throw cet::exception(
"Too many over-size Fragments received! Please adjust max_event_size_bytes or max_fragment_size_bytes!");
246 TLOG(6) <<
"Dropping over-size fragment with sequence id " << frag.sequence_id <<
" and fragment id " << frag.fragment_id <<
" into " << (
void*)dropped_data_[frag.fragment_id]->dataBegin();
247 return dropped_data_[frag.fragment_id]->dataBegin();
250 TLOG(14) <<
"WriteFragmentHeader END";
256 TLOG(TLVL_TRACE) <<
"DoneWritingFragment BEGIN";
257 auto buffer = getBufferForSequenceID_(frag.sequence_id,
false, frag.timestamp);
258 if (buffer == -1) Detach(
true,
"SharedMemoryEventManager",
"getBufferForSequenceID_ returned -1 when it REALLY shouldn't have! Check program logic!");
259 if (buffer == -2) {
return; }
261 statsHelper_.addSample(FRAGMENTS_RECEIVED_STAT_KEY, frag.word_count *
sizeof(RawDataType));
263 TLOG(TLVL_BUFLCK) <<
"DoneWritingFragment: obtaining buffer_mutexes lock for buffer " << buffer;
265 std::unique_lock<std::mutex> lk(buffer_mutexes_[buffer]);
267 TLOG(TLVL_BUFLCK) <<
"DoneWritingFragment: obtained buffer_mutexes lock for buffer " << buffer;
271 TLOG(TLVL_DEBUG) <<
"DoneWritingFragment: Received Fragment with sequence ID " << frag.sequence_id <<
" and fragment id " << frag.fragment_id <<
" (type " << (int)frag.type <<
")";
272 auto hdr = getEventHeader_(buffer);
275 hdr->run_id = run_id_;
277 hdr->subrun_id = GetSubrunForSequenceID(frag.sequence_id);
279 TLOG(TLVL_TRACE) <<
"DoneWritingFragment: Updating buffer touch time";
282 buffer_writes_pending_[buffer]--;
283 if (buffer_writes_pending_[buffer] != 0)
285 TLOG(TLVL_TRACE) <<
"Done writing fragment, but there's another writer. Not doing bookkeeping steps.";
288 TLOG(TLVL_TRACE) <<
"Done writing fragment, and no other writer. Doing bookkeeping steps.";
289 auto frag_count = GetFragmentCount(frag.sequence_id);
290 hdr->is_complete = frag_count >= num_fragments_per_event_;
292 if (frag_count > num_fragments_per_event_)
294 TLOG(TLVL_WARNING) <<
"DoneWritingFragment: This Event has more Fragments ( " << frag_count <<
" ) than specified in configuration ( " << num_fragments_per_event_ <<
" )!"
295 <<
" This is probably due to a misconfiguration and is *not* a reliable mode!";
298 TLOG(TLVL_TRACE) <<
"DoneWritingFragment: Received Fragment with sequence ID " << frag.sequence_id <<
" and fragment id " << frag.fragment_id <<
", count/expected = " << frag_count <<
"/" << num_fragments_per_event_;
299 #if ART_SUPPORTS_DUPLICATE_EVENTS
300 if (!hdr->is_complete && released_incomplete_events_.count(frag.sequence_id))
302 hdr->is_complete = frag_count >= released_incomplete_events_[frag.sequence_id] && buffer_writes_pending_[buffer] == 0;
307 complete_buffer_(buffer);
308 if (requests_) requests_->SendRequest(
true);
309 TLOG(TLVL_TRACE) <<
"DoneWritingFragment END";
314 return GetFragmentCountInBuffer(getBufferForSequenceID_(seqID,
false), type);
319 if (buffer == -1)
return 0;
320 ResetReadPos(buffer);
321 IncrementReadPos(buffer,
sizeof(detail::RawEventHeader));
325 while (MoreDataInBuffer(buffer))
327 auto fragHdr =
reinterpret_cast<artdaq::detail::RawFragmentHeader*
>(GetReadPos(buffer));
328 IncrementReadPos(buffer, fragHdr->word_count *
sizeof(RawDataType));
329 if (type != Fragment::InvalidFragmentType && fragHdr->type != type)
continue;
330 TLOG(TLVL_TRACE) <<
"Adding Fragment with size=" << fragHdr->word_count <<
" to Fragment count";
341 auto start_time = std::chrono::steady_clock::now();
343 TLOG(TLVL_INFO) <<
"Starting art process with config file " << config_file->getFileName();
349 char* filename =
new char[config_file->getFileName().length() + 1];
350 strcpy(filename, config_file->getFileName().c_str());
353 std::string debugArgS =
"--config-out=" + app_name +
"_art.out";
354 char* debugArg =
new char[debugArgS.length() + 1];
355 strcpy(debugArg, debugArgS.c_str());
357 std::vector<char*> args{(
char*)
"art", (
char*)
"-c", filename, debugArg, NULL};
359 std::vector<char*> args{(
char*)
"art", (
char*)
"-c", filename, NULL};
369 std::string envVarKey =
"ARTDAQ_PARTITION_NUMBER";
370 std::string envVarValue = std::to_string(GetPartitionNumber());
371 if (setenv(envVarKey.c_str(), envVarValue.c_str(), 1) != 0)
373 TLOG(TLVL_ERROR) <<
"Error setting environment variable \"" << envVarKey
374 <<
"\" in the environment of a child art process. "
375 <<
"This may result in incorrect TCP port number "
376 <<
"assignments or other issues, and data may "
377 <<
"not flow through the system correctly.";
379 envVarKey =
"ARTDAQ_APPLICATION_NAME";
380 envVarValue = app_name;
381 if (setenv(envVarKey.c_str(), envVarValue.c_str(), 1) != 0)
383 TLOG(TLVL_DEBUG) <<
"Error setting environment variable \"" << envVarKey
384 <<
"\" in the environment of a child art process. ";
386 envVarKey =
"ARTDAQ_RANK";
387 envVarValue = std::to_string(my_rank);
388 if (setenv(envVarKey.c_str(), envVarValue.c_str(), 1) != 0)
390 TLOG(TLVL_DEBUG) <<
"Error setting environment variable \"" << envVarKey
391 <<
"\" in the environment of a child art process. ";
394 execvp(
"art", &args[0]);
403 std::cout <<
"Please run the following command in a separate terminal:" << std::endl
404 <<
"art -c " << config_file->getFileName() << std::endl
405 <<
"Then, in a third terminal, execute: \"ps aux|grep [a]rt -c " << config_file->getFileName() <<
"\" and note the PID of the art process." << std::endl
406 <<
"Finally, return to this window and enter the pid: " << std::endl;
411 TLOG(TLVL_INFO) <<
"PID of new art process is " << pid;
413 std::unique_lock<std::mutex> lk(art_process_mutex_);
414 art_processes_.insert(pid);
417 auto sts = waitid(P_PID, pid, &status, WEXITED);
418 TLOG(TLVL_INFO) <<
"Removing PID " << pid <<
" from process list";
420 std::unique_lock<std::mutex> lk(art_process_mutex_);
421 art_processes_.erase(pid);
425 TLOG(TLVL_WARNING) <<
"Error occurred in waitid for art process " << pid <<
": " << errno <<
" (" << strerror(errno) <<
").";
427 else if (status.si_code == CLD_EXITED && status.si_status == 0)
429 TLOG(TLVL_INFO) <<
"art process " << pid <<
" exited normally, " << (restart_art_ ?
"restarting" :
"not restarting");
433 auto art_lifetime = TimeUtils::GetElapsedTime(start_time);
434 if (art_lifetime < minimum_art_lifetime_s_) restart_art_ =
false;
436 auto exit_type =
"exited with status code";
437 switch (status.si_code)
441 exit_type =
"was killed with signal";
448 TLOG((restart_art_ ? TLVL_WARNING : TLVL_ERROR))
449 <<
"art process " << pid <<
" " << exit_type <<
" " << status.si_status
450 << (status.si_code == CLD_DUMPED ?
" (core dumped)" :
"")
451 <<
" after running for " << std::setprecision(2) << std::fixed << art_lifetime <<
" seconds, "
452 << (restart_art_ ?
"restarting" :
"not restarting");
454 }
while (restart_art_);
459 restart_art_ = always_restart_art_;
460 if (num_art_processes_ == 0)
return;
461 for (
size_t ii = 0; ii < num_art_processes_; ++ii)
463 StartArtProcess(current_art_pset_);
469 static std::mutex start_art_mutex;
470 std::unique_lock<std::mutex> lk(start_art_mutex);
472 restart_art_ = always_restart_art_;
473 auto initialCount = GetAttachedCount();
474 auto startTime = std::chrono::steady_clock::now();
476 if (pset != current_art_pset_ || !current_art_config_file_)
478 current_art_pset_ = pset;
479 current_art_config_file_ = std::make_shared<art_config_file>(pset );
481 std::shared_ptr<std::atomic<pid_t>> pid(
new std::atomic<pid_t>(-1));
482 boost::thread thread([&] { RunArt(current_art_config_file_, pid); });
485 auto currentCount = GetAttachedCount() - initialCount;
486 while ((currentCount < 1 || *pid <= 0) && (TimeUtils::GetElapsedTime(startTime) < 5 || manual_art_))
489 currentCount = GetAttachedCount() - initialCount;
491 if ((currentCount < 1 || *pid <= 0) && manual_art_)
493 TLOG(TLVL_WARNING) <<
"Manually-started art process has not connected to shared memory or has bad PID: connected:" << currentCount <<
", PID:" << pid;
496 else if (currentCount < 1 || *pid <= 0)
498 TLOG(TLVL_WARNING) <<
"art process has not started after 5s. Check art configuration!"
499 <<
" (pid=" << *pid <<
", attachedCount=" << currentCount <<
")";
504 TLOG(TLVL_INFO) << std::setw(4) << std::fixed <<
"art initialization took "
505 << TimeUtils::GetElapsedTime(startTime) <<
" seconds.";
513 restart_art_ =
false;
517 auto check_pids = [&](
bool print) {
518 std::unique_lock<std::mutex> lk(art_process_mutex_);
519 for (
auto pid = pids.begin(); pid != pids.end();)
525 TLOG(TLVL_WARNING) <<
"Removing an invalid PID (" << *pid
526 <<
") from the shutdown list.";
527 pid = pids.erase(pid);
529 else if (kill(*pid, 0) < 0)
531 pid = pids.erase(pid);
535 if (print) std::cout << *pid <<
" ";
540 auto count_pids = [&]() {
541 std::unique_lock<std::mutex> lk(art_process_mutex_);
545 if (count_pids() == 0)
547 TLOG(14) <<
"All art processes already exited, nothing to do.";
554 int graceful_wait_ms = art_event_processing_time_us_ * size() * 10 / 1000;
555 int gentle_wait_ms = art_event_processing_time_us_ * size() * 2 / 1000;
556 int int_wait_ms = art_event_processing_time_us_ * size() / 1000;
557 auto shutdown_start = std::chrono::steady_clock::now();
559 TLOG(TLVL_TRACE) <<
"Waiting up to " << graceful_wait_ms <<
" ms for all art processes to exit gracefully";
560 for (
int ii = 0; ii < graceful_wait_ms; ++ii)
565 if (count_pids() == 0)
567 TLOG(TLVL_TRACE) <<
"All art processes exited after " << TimeUtils::GetElapsedTimeMilliseconds(shutdown_start) <<
" ms.";
573 TLOG(TLVL_TRACE) <<
"Gently informing art processes that it is time to shut down";
574 std::unique_lock<std::mutex> lk(art_process_mutex_);
575 for (
auto pid : pids)
577 TLOG(TLVL_TRACE) <<
"Sending SIGQUIT to pid " << pid;
582 TLOG(TLVL_TRACE) <<
"Waiting up to " << gentle_wait_ms <<
" ms for all art processes to exit from SIGQUIT";
583 for (
int ii = 0; ii < gentle_wait_ms; ++ii)
588 if (count_pids() == 0)
590 TLOG(TLVL_TRACE) <<
"All art processes exited after " << TimeUtils::GetElapsedTimeMilliseconds(shutdown_start) <<
" ms.";
596 TLOG(TLVL_TRACE) <<
"Insisting that the art processes shut down";
597 std::unique_lock<std::mutex> lk(art_process_mutex_);
598 for (
auto pid : pids)
604 TLOG(TLVL_TRACE) <<
"Waiting up to " << int_wait_ms <<
" ms for all art processes to exit from SIGINT";
605 for (
int ii = 0; ii < int_wait_ms; ++ii)
611 if (count_pids() == 0)
613 TLOG(TLVL_TRACE) <<
"All art processes exited after " << TimeUtils::GetElapsedTimeMilliseconds(shutdown_start) <<
" ms.";
618 TLOG(TLVL_TRACE) <<
"Killing remaning art processes with extreme prejudice";
619 while (count_pids() > 0)
622 std::unique_lock<std::mutex> lk(art_process_mutex_);
623 kill(*pids.begin(), SIGKILL);
631 std::cout <<
"Please shut down all art processes, then hit return/enter" << std::endl;
632 while (count_pids() > 0)
634 std::cout <<
"The following PIDs are running: ";
636 std::cout << std::endl;
645 TLOG(TLVL_DEBUG) <<
"ReconfigureArt BEGIN";
646 if (restart_art_ || !always_restart_art_)
650 for (
size_t ii = 0; ii < broadcasts_.size(); ++ii)
652 broadcasts_.MarkBufferEmpty(ii,
true);
654 if (newRun == 0) newRun = run_id_ + 1;
656 if (art_pset != current_art_pset_ || !current_art_config_file_)
658 current_art_pset_ = art_pset;
659 current_art_config_file_ = std::make_shared<art_config_file>(art_pset );
662 if (n_art_processes != -1)
664 TLOG(TLVL_INFO) <<
"Setting number of art processes to " << n_art_processes;
665 num_art_processes_ = n_art_processes;
668 TLOG(TLVL_DEBUG) <<
"ReconfigureArt END";
674 init_fragments_.clear();
675 TLOG(TLVL_DEBUG) <<
"SharedMemoryEventManager::endOfData";
676 restart_art_ =
false;
678 size_t initialStoreSize = GetIncompleteEventCount();
679 TLOG(TLVL_DEBUG) <<
"endOfData: Flushing " << initialStoreSize
680 <<
" stale events from the SharedMemoryEventManager.";
681 int counter = initialStoreSize;
682 while (active_buffers_.size() > 0 && counter > 0)
684 complete_buffer_(*active_buffers_.begin());
687 TLOG(TLVL_DEBUG) <<
"endOfData: Done flushing, there are now " << GetIncompleteEventCount()
688 <<
" stale events in the SharedMemoryEventManager.";
690 TLOG(TLVL_DEBUG) <<
"Waiting for " << (ReadReadyCount() + (size() - WriteReadyCount(overwrite_mode_))) <<
" outstanding buffers...";
691 auto start = std::chrono::steady_clock::now();
692 auto lastReadCount = ReadReadyCount() + (size() - WriteReadyCount(overwrite_mode_));
693 auto end_of_data_wait_us = art_event_processing_time_us_ * (lastReadCount > 0 ? lastReadCount : 1);
695 auto outstanding_buffer_wait_time = art_event_processing_time_us_ > 100000 ? 100000 : art_event_processing_time_us_;
698 while (lastReadCount > 0 && (end_of_data_wait_us == 0 || TimeUtils::GetElapsedTimeMicroseconds(start) < end_of_data_wait_us) && get_art_process_count_() > 0)
700 auto temp = ReadReadyCount() + (size() - WriteReadyCount(overwrite_mode_));
701 if (temp != lastReadCount)
703 TLOG(TLVL_TRACE) <<
"Waiting for " << temp <<
" outstanding buffers...";
704 lastReadCount = temp;
705 start = std::chrono::steady_clock::now();
707 if (lastReadCount > 0)
709 TRACE(19,
"About to sleep %lu us - lastReadCount=%lu size=%lu end_of_data_wait_us=%lu", outstanding_buffer_wait_time, lastReadCount, size(), end_of_data_wait_us);
710 usleep(outstanding_buffer_wait_time);
714 TLOG(TLVL_DEBUG) <<
"endOfData: After wait for outstanding buffers. Still outstanding: " << lastReadCount <<
", time waited: "
715 << TimeUtils::GetElapsedTime(start) <<
" s / " << (end_of_data_wait_us / 1000000.0) <<
" s, art process count: " << get_art_process_count_();
717 TLOG(TLVL_DEBUG) <<
"endOfData: Broadcasting EndOfData Fragment";
718 FragmentPtrs broadcast;
719 broadcast.emplace_back(Fragment::eodFrag(GetBufferCount()));
720 bool success = broadcastFragments_(broadcast);
723 TLOG(TLVL_DEBUG) <<
"endOfData: Clearing buffers to make room for EndOfData Fragment";
724 for (
size_t ii = 0; ii < broadcasts_.size(); ++ii)
726 broadcasts_.MarkBufferEmpty(ii,
true);
728 broadcastFragments_(broadcast);
730 auto endOfDataProcessingStart = std::chrono::steady_clock::now();
731 while (get_art_process_count_() > 0)
733 TLOG(TLVL_DEBUG) <<
"There are " << get_art_process_count_() <<
" art processes remaining. Proceeding to shutdown.";
735 ShutdownArtProcesses(art_processes_);
737 TLOG(TLVL_DEBUG) <<
"It took " << TimeUtils::GetElapsedTime(endOfDataProcessingStart) <<
" s for all art processes to close after sending EndOfData Fragment";
739 ResetAttachedCount();
741 TLOG(TLVL_DEBUG) <<
"endOfData: Clearing buffers";
742 for (
size_t ii = 0; ii < size(); ++ii)
744 MarkBufferEmpty(ii,
true);
752 released_events_.clear();
753 released_incomplete_events_.clear();
755 TLOG(TLVL_DEBUG) <<
"endOfData: Shutting down RequestSender";
756 requests_.reset(
nullptr);
758 TLOG(TLVL_DEBUG) <<
"endOfData END";
759 TLOG(TLVL_INFO) <<
"EndOfData Complete. There were " << GetLastSeenBufferID() <<
" buffers processed.";
766 init_fragments_.clear();
767 statsHelper_.resetStatistics();
768 TLOG(TLVL_TRACE) <<
"startRun: Clearing broadcast buffers";
769 for (
size_t ii = 0; ii < broadcasts_.size(); ++ii)
771 broadcasts_.MarkBufferEmpty(ii,
true);
773 released_events_.clear();
774 released_incomplete_events_.clear();
778 std::unique_lock<std::mutex> lk(subrun_event_map_mutex_);
779 subrun_event_map_.clear();
780 subrun_event_map_[0] = 1;
782 run_event_count_ = 0;
783 run_incomplete_event_count_ = 0;
787 requests_->SetRunNumber(static_cast<uint32_t>(run_id_));
788 requests_->SendRoutingToken(queue_size_, run_id_);
790 TLOG(TLVL_DEBUG) <<
"Starting run " << run_id_
791 <<
", max queue size = "
794 << GetLockedBufferCount();
797 metricMan->sendMetric(
"Run Number", static_cast<unsigned long>(run_id_),
"Run", 1, MetricMode::LastPoint);
803 TLOG(TLVL_INFO) <<
"Ending run " << run_id_;
804 FragmentPtr endOfRunFrag(
new Fragment(static_cast<size_t>(ceil(
sizeof(my_rank) /
805 static_cast<double>(
sizeof(Fragment::value_type))))));
807 TLOG(TLVL_DEBUG) <<
"Broadcasting EndOfRun Fragment";
808 endOfRunFrag->setSystemType(Fragment::EndOfRunFragmentType);
809 *endOfRunFrag->dataBegin() = my_rank;
810 FragmentPtrs broadcast;
811 broadcast.emplace_back(std::move(endOfRunFrag));
812 broadcastFragments_(broadcast);
814 TLOG(TLVL_INFO) <<
"Run " << run_id_ <<
" has ended. There were " << run_event_count_ <<
" events in this run.";
815 run_event_count_ = 0;
816 run_incomplete_event_count_ = 0;
817 oversize_fragment_count_ = 0;
819 std::unique_lock<std::mutex> lk(subrun_event_map_mutex_);
820 subrun_event_map_.clear();
821 subrun_event_map_[0] = 1;
829 if (boundary == 0 || boundary == Fragment::InvalidSequenceID)
return;
831 std::unique_lock<std::mutex> lk(subrun_event_map_mutex_);
833 TLOG(TLVL_INFO) <<
"Will roll over to subrun " << subrun <<
" when I reach Sequence ID " << boundary;
834 subrun_event_map_[boundary] = subrun;
835 while (subrun_event_map_.size() > max_subrun_event_map_length_)
837 subrun_event_map_.erase(subrun_event_map_.begin());
843 Fragment::sequence_id_t seqID = 0;
846 std::unique_lock<std::mutex> lk(subrun_event_map_mutex_);
847 for (
auto& it : subrun_event_map_)
849 if (it.first >= seqID) seqID = it.first + 1;
850 if (it.second >= subrun) subrun = it.second + 1;
853 rolloverSubrun(seqID, subrun);
860 metricMan->sendMetric(
"Incomplete Event Count", GetIncompleteEventCount(),
"events", 1, MetricMode::LastPoint);
861 metricMan->sendMetric(
"Pending Event Count", GetPendingEventCount(),
"events", 1, MetricMode::LastPoint);
864 if (incomplete_event_report_interval_ms_ > 0 && GetLockedBufferCount())
866 if (TimeUtils::GetElapsedTimeMilliseconds(last_incomplete_event_report_time_) < static_cast<size_t>(incomplete_event_report_interval_ms_))
869 last_incomplete_event_report_time_ = std::chrono::steady_clock::now();
870 std::ostringstream oss;
871 oss <<
"Incomplete Events (" << num_fragments_per_event_ <<
"): ";
872 for (
auto& ev : active_buffers_)
874 auto hdr = getEventHeader_(ev);
875 oss << hdr->sequence_id <<
" (" << GetFragmentCount(hdr->sequence_id) <<
"), ";
877 TLOG(TLVL_DEBUG) << oss.str();
881 bool artdaq::SharedMemoryEventManager::broadcastFragments_(FragmentPtrs& frags)
883 if (frags.size() == 0)
885 TLOG(TLVL_ERROR) <<
"Requested broadcast but no Fragments given!";
888 TLOG(TLVL_DEBUG) <<
"Broadcasting Fragments with seqID=" << frags.front()->sequenceID()
889 <<
", type " << detail::RawFragmentHeader::SystemTypeToString(frags.front()->type())
890 <<
", size=" << frags.front()->sizeBytes() <<
"B.";
891 auto buffer = broadcasts_.GetBufferForWriting(
false);
892 TLOG(TLVL_DEBUG) <<
"broadcastFragments_: after getting buffer 1st buffer=" << buffer;
893 auto start_time = std::chrono::steady_clock::now();
894 while (buffer == -1 && TimeUtils::GetElapsedTimeMilliseconds(start_time) < static_cast<size_t>(broadcast_timeout_ms_))
897 buffer = broadcasts_.GetBufferForWriting(
false);
899 TLOG(TLVL_DEBUG) <<
"broadcastFragments_: after getting buffer w/timeout, buffer=" << buffer <<
", elapsed time=" << TimeUtils::GetElapsedTime(start_time) <<
" s.";
902 TLOG(TLVL_ERROR) <<
"Broadcast of fragment type " << frags.front()->typeString() <<
" failed due to timeout waiting for buffer!";
906 TLOG(TLVL_DEBUG) <<
"broadcastFragments_: Filling in RawEventHeader";
907 auto hdr =
reinterpret_cast<detail::RawEventHeader*
>(broadcasts_.GetBufferStart(buffer));
908 hdr->run_id = run_id_;
909 hdr->subrun_id = GetSubrunForSequenceID(frags.front()->sequenceID());
910 hdr->sequence_id = frags.front()->sequenceID();
911 hdr->is_complete =
true;
912 broadcasts_.IncrementWritePos(buffer,
sizeof(detail::RawEventHeader));
914 for (
auto& frag : frags)
916 TLOG(TLVL_DEBUG) <<
"broadcastFragments_ before Write calls";
917 if (frag->sequenceID() != hdr->sequence_id || frag->type() != frags.front()->type())
919 TLOG(TLVL_WARNING) <<
"Not sending fragment because its SequenceID or Type disagrees with leading Fragment";
922 broadcasts_.Write(buffer, frag->headerAddress(), frag->size() *
sizeof(RawDataType));
925 TLOG(TLVL_DEBUG) <<
"broadcastFragments_ Marking buffer full";
926 broadcasts_.MarkBufferFull(buffer, -1);
927 TLOG(TLVL_DEBUG) <<
"broadcastFragment_s Complete";
931 artdaq::detail::RawEventHeader* artdaq::SharedMemoryEventManager::getEventHeader_(
int buffer)
933 return reinterpret_cast<detail::RawEventHeader*
>(GetBufferStart(buffer));
938 std::unique_lock<std::mutex> lk(subrun_event_map_mutex_);
940 TLOG(TLVL_TRACE) <<
"GetSubrunForSequenceID BEGIN map size = " << subrun_event_map_.size();
941 auto it = subrun_event_map_.begin();
944 while (it->first <= seqID && it != subrun_event_map_.end())
946 TLOG(TLVL_TRACE) <<
"Map has sequence ID " << it->first <<
", subrun " << it->second <<
" (looking for <= " << seqID <<
")";
951 TLOG(TLVL_DEBUG) <<
"GetSubrunForSequenceID returning subrun " << subrun <<
" for sequence ID " << seqID;
955 int artdaq::SharedMemoryEventManager::getBufferForSequenceID_(Fragment::sequence_id_t seqID,
bool create_new, Fragment::timestamp_t timestamp)
957 TLOG(14) <<
"getBufferForSequenceID " << seqID <<
" BEGIN";
958 std::unique_lock<std::mutex> lk(sequence_id_mutex_);
960 TLOG(14) <<
"getBufferForSequenceID obtained sequence_id_mutex for seqid=" << seqID;
962 auto buffers = GetBuffersOwnedByManager();
963 for (
auto& buf : buffers)
965 auto hdr = getEventHeader_(buf);
966 if (hdr->sequence_id == seqID)
968 TLOG(14) <<
"getBufferForSequenceID " << seqID <<
" returning " << buf;
973 #if !ART_SUPPORTS_DUPLICATE_EVENTS
974 if (released_incomplete_events_.count(seqID))
976 TLOG(TLVL_ERROR) <<
"Event " << seqID <<
" has already been marked \"Incomplete\" and sent to art!";
979 if (released_events_.count(seqID))
981 TLOG(TLVL_ERROR) <<
"Event " << seqID <<
" has already been completed and released to art! Check configuration for inconsistent Fragment count per event!";
986 if (!create_new)
return -1;
988 check_pending_buffers_(lk);
989 int new_buffer = GetBufferForWriting(
false);
991 if (new_buffer == -1)
993 new_buffer = GetBufferForWriting(overwrite_mode_);
996 if (new_buffer == -1)
return -1;
997 TLOG(TLVL_BUFLCK) <<
"getBufferForSequenceID_: obtaining buffer_mutexes lock for buffer " << new_buffer;
998 std::unique_lock<std::mutex> buffer_lk(buffer_mutexes_[new_buffer]);
999 TLOG(TLVL_BUFLCK) <<
"getBufferForSequenceID_: obtained buffer_mutexes lock for buffer " << new_buffer;
1001 auto hdr = getEventHeader_(new_buffer);
1002 hdr->is_complete =
false;
1003 hdr->run_id = run_id_;
1004 hdr->subrun_id = GetSubrunForSequenceID(seqID);
1005 hdr->event_id = use_sequence_id_for_event_number_ ?
static_cast<uint32_t
>(seqID) : static_cast<uint32_t>(timestamp);
1006 hdr->sequence_id = seqID;
1007 buffer_writes_pending_[new_buffer] = 0;
1008 IncrementWritePos(new_buffer,
sizeof(detail::RawEventHeader));
1009 SetMFIteration(
"Sequence ID " + std::to_string(seqID));
1011 TLOG(TLVL_BUFFER) <<
"getBufferForSequenceID placing " << new_buffer <<
" to active.";
1012 active_buffers_.insert(new_buffer);
1013 TLOG(TLVL_BUFFER) <<
"Buffer occupancy now (total,full,reading,empty,pending,active)=("
1015 << ReadReadyCount() <<
","
1016 << WriteReadyCount(
true) - WriteReadyCount(
false) - ReadReadyCount() <<
","
1017 << WriteReadyCount(
false) <<
","
1018 << pending_buffers_.size() <<
","
1019 << active_buffers_.size() <<
")";
1023 if (timestamp != Fragment::InvalidTimestamp)
1025 requests_->AddRequest(seqID, timestamp);
1031 requests_->SendRequest();
1034 TLOG(14) <<
"getBufferForSequenceID " << seqID <<
" returning newly initialized buffer " << new_buffer;
1038 bool artdaq::SharedMemoryEventManager::hasFragments_(
int buffer)
1040 if (buffer == -1)
return true;
1041 if (!CheckBuffer(buffer, BufferSemaphoreFlags::Writing))
1045 ResetReadPos(buffer);
1046 IncrementReadPos(buffer,
sizeof(detail::RawEventHeader));
1047 return MoreDataInBuffer(buffer);
1050 void artdaq::SharedMemoryEventManager::complete_buffer_(
int buffer)
1052 auto hdr = getEventHeader_(buffer);
1053 if (hdr->is_complete)
1055 TLOG(TLVL_DEBUG) <<
"complete_buffer_: This fragment completes event " << hdr->sequence_id <<
".";
1058 TLOG(TLVL_BUFFER) <<
"complete_buffer_ moving " << buffer <<
" from active to pending.";
1060 TLOG(TLVL_BUFLCK) <<
"complete_buffer_: obtaining sequence_id_mutex lock for seqid=" << hdr->sequence_id;
1061 std::unique_lock<std::mutex> lk(sequence_id_mutex_);
1062 TLOG(TLVL_BUFLCK) <<
"complete_buffer_: obtained sequence_id_mutex lock for seqid=" << hdr->sequence_id;
1063 active_buffers_.erase(buffer);
1064 pending_buffers_.insert(buffer);
1065 released_events_.insert(hdr->sequence_id);
1066 while (released_events_.size() > max_event_list_length_)
1068 released_events_.erase(released_events_.begin());
1071 TLOG(TLVL_BUFFER) <<
"Buffer occupancy now (total,full,reading,empty,pending,active)=("
1073 << ReadReadyCount() <<
","
1074 << WriteReadyCount(
true) - WriteReadyCount(
false) - ReadReadyCount() <<
","
1075 << WriteReadyCount(
false) <<
","
1076 << pending_buffers_.size() <<
","
1077 << active_buffers_.size() <<
")";
1081 requests_->RemoveRequest(hdr->sequence_id);
1084 CheckPendingBuffers();
1087 bool artdaq::SharedMemoryEventManager::bufferComparator(
int bufA,
int bufB)
1089 return getEventHeader_(bufA)->sequence_id < getEventHeader_(bufB)->sequence_id;
1094 TLOG(TLVL_BUFLCK) <<
"CheckPendingBuffers: Obtaining sequence_id_mutex_";
1095 std::unique_lock<std::mutex> lk(sequence_id_mutex_);
1096 TLOG(TLVL_BUFLCK) <<
"CheckPendingBuffers: Obtained sequence_id_mutex_";
1097 check_pending_buffers_(lk);
1100 void artdaq::SharedMemoryEventManager::check_pending_buffers_(std::unique_lock<std::mutex>
const& lock)
1102 TLOG(TLVL_TRACE) <<
"check_pending_buffers_ BEGIN Locked=" << std::boolalpha << lock.owns_lock();
1104 auto buffers = GetBuffersOwnedByManager();
1105 for (
auto buf : buffers)
1107 if (ResetBuffer(buf) && !pending_buffers_.count(buf))
1109 TLOG(15) <<
"check_pending_buffers_ Incomplete buffer detected, buf=" << buf <<
" active_bufers_.count(buf)=" << active_buffers_.count(buf) <<
" buffer_writes_pending_[buf]=" << buffer_writes_pending_[buf].load();
1110 auto hdr = getEventHeader_(buf);
1111 if (active_buffers_.count(buf) && (buffer_writes_pending_[buf].load() == 0 || !running_))
1115 requests_->RemoveRequest(hdr->sequence_id);
1117 TLOG(TLVL_BUFFER) <<
"check_pending_buffers_ moving buffer " << buf <<
" from active to pending";
1118 active_buffers_.erase(buf);
1119 pending_buffers_.insert(buf);
1120 TLOG(TLVL_BUFFER) <<
"Buffer occupancy now (total,full,reading,empty,pending,active)=("
1122 << ReadReadyCount() <<
","
1123 << WriteReadyCount(
true) - WriteReadyCount(
false) - ReadReadyCount() <<
","
1124 << WriteReadyCount(
false) <<
","
1125 << pending_buffers_.size() <<
","
1126 << active_buffers_.size() <<
")";
1128 run_incomplete_event_count_++;
1129 if (metricMan) metricMan->sendMetric(
"Incomplete Event Rate", 1,
"events/s", 3, MetricMode::Rate);
1130 if (!released_incomplete_events_.count(hdr->sequence_id))
1132 released_incomplete_events_[hdr->sequence_id] = num_fragments_per_event_ - GetFragmentCountInBuffer(buf);
1136 released_incomplete_events_[hdr->sequence_id] -= GetFragmentCountInBuffer(buf);
1138 TLOG(TLVL_WARNING) <<
"Active event " << hdr->sequence_id <<
" is stale. Scheduling release of incomplete event (missing " << released_incomplete_events_[hdr->sequence_id] <<
" Fragments) to art.";
1143 std::list<int> sorted_buffers(pending_buffers_.begin(), pending_buffers_.end());
1144 sorted_buffers.sort([
this](
int a,
int b) {
return bufferComparator(a, b); });
1147 double eventSize = 0;
1148 for (
auto buf : sorted_buffers)
1150 auto hdr = getEventHeader_(buf);
1151 auto thisEventSize = BufferDataSize(buf);
1153 TLOG(TLVL_DEBUG) <<
"Releasing event " << std::to_string(hdr->sequence_id) <<
" in buffer " << buf <<
" to art, "
1154 <<
"event_size=" << thisEventSize <<
", buffer_size=" << BufferSize();
1155 statsHelper_.addSample(EVENTS_RELEASED_STAT_KEY, thisEventSize);
1157 TLOG(TLVL_BUFFER) <<
"check_pending_buffers_ removing buffer " << buf <<
" moving from pending to full";
1158 MarkBufferFull(buf);
1161 eventSize += thisEventSize;
1162 pending_buffers_.erase(buf);
1163 TLOG(TLVL_BUFFER) <<
"Buffer occupancy now (total,full,reading,empty,pending,active)=("
1165 << ReadReadyCount() <<
","
1166 << WriteReadyCount(
true) - WriteReadyCount(
false) - ReadReadyCount() <<
","
1167 << WriteReadyCount(
false) <<
","
1168 << pending_buffers_.size() <<
","
1169 << active_buffers_.size() <<
")";
1172 if (requests_ && requests_->RoutingTokenSendsEnabled())
1174 TLOG(TLVL_TRACE) <<
"Sent tokens: " << requests_->GetSentTokenCount() <<
", Event count: " << run_event_count_;
1175 auto outstanding_tokens = requests_->GetSentTokenCount() - run_event_count_;
1176 auto available_buffers = WriteReadyCount(overwrite_mode_);
1178 TLOG(TLVL_TRACE) <<
"check_pending_buffers_: outstanding_tokens: " << outstanding_tokens <<
", available_buffers: " << available_buffers
1179 <<
", tokens_to_send: " << available_buffers - outstanding_tokens;
1181 if (available_buffers > outstanding_tokens)
1183 auto tokens_to_send = available_buffers - outstanding_tokens;
1185 while (tokens_to_send > 0)
1187 TLOG(35) <<
"check_pending_buffers_: Sending a Routing Token";
1188 requests_->SendRoutingToken(1, run_id_);
1194 if (statsHelper_.readyToReport())
1196 std::string statString = buildStatisticsString_();
1197 TLOG(TLVL_INFO) << statString;
1202 TLOG(TLVL_TRACE) <<
"check_pending_buffers_: Sending Metrics";
1203 metricMan->sendMetric(
"Event Rate", counter,
"Events", 1, MetricMode::Rate);
1204 metricMan->sendMetric(
"Data Rate", eventSize,
"Bytes", 1, MetricMode::Rate);
1207 metricMan->sendMetric(
"Average Event Size", eventSize / counter,
"Bytes", 1, MetricMode::Average);
1210 metricMan->sendMetric(
"Events Released to art this run", run_event_count_,
"Events", 1, MetricMode::LastPoint);
1211 metricMan->sendMetric(
"Incomplete Events Released to art this run", run_incomplete_event_count_,
"Events", 1, MetricMode::LastPoint);
1212 if (requests_) metricMan->sendMetric(
"Tokens sent", requests_->GetSentTokenCount(),
"Tokens", 2, MetricMode::LastPoint);
1214 auto bufferReport = GetBufferReport();
1215 int full = 0, empty = 0, writing = 0, reading = 0;
1216 for (
auto& buf : bufferReport)
1220 case BufferSemaphoreFlags::Full:
1223 case BufferSemaphoreFlags::Empty:
1226 case BufferSemaphoreFlags::Writing:
1229 case BufferSemaphoreFlags::Reading:
1234 auto total = size();
1235 TLOG(15) <<
"Buffer usage: full=" << full <<
", empty=" << empty <<
", writing=" << writing <<
", reading=" << reading <<
", total=" << total;
1237 metricMan->sendMetric(
"Shared Memory Full Buffers", full,
"buffers", 2, MetricMode::LastPoint);
1238 metricMan->sendMetric(
"Shared Memory Available Buffers", empty,
"buffers", 2, MetricMode::LastPoint);
1239 metricMan->sendMetric(
"Shared Memory Pending Buffers", writing,
"buffers", 2, MetricMode::LastPoint);
1240 metricMan->sendMetric(
"Shared Memory Reading Buffers", reading,
"buffers", 2, MetricMode::LastPoint);
1243 metricMan->sendMetric(
"Shared Memory Full %", full * 100 / static_cast<double>(total),
"%", 2, MetricMode::LastPoint);
1244 metricMan->sendMetric(
"Shared Memory Available %", empty * 100 / static_cast<double>(total),
"%", 2, MetricMode::LastPoint);
1247 TLOG(TLVL_TRACE) <<
"check_pending_buffers_ END";
1250 void artdaq::SharedMemoryEventManager::send_init_frags_()
1252 if (init_fragments_.size() >= init_fragment_count_ && init_fragment_count_ > 0)
1254 TLOG(TLVL_INFO) <<
"Broadcasting init fragment to all art subprocesses...";
1257 std::string fileName =
"receiveInitMessage_" + std::to_string(my_rank) +
".bin";
1258 std::fstream ostream(fileName.c_str(), std::ios::out | std::ios::binary);
1259 ostream.write(reinterpret_cast<char*>(init_fragment_->dataBeginBytes()), init_fragment_->dataSizeBytes());
1263 broadcastFragments_(init_fragments_);
1264 TLOG(TLVL_TRACE) <<
"Init Fragment sent";
1266 else if (init_fragment_count_ > 0)
1268 TLOG(TLVL_WARNING) <<
"Cannot send init fragments because I haven't yet received them! Set send_init_fragments to false or init_fragment_count to 0 if this process does not receive serialized art events to avoid potentially lengthy timeouts!";
1273 artdaq::FragmentPtrs begin_run_fragments_;
1274 begin_run_fragments_.emplace_back(
new artdaq::Fragment());
1275 begin_run_fragments_.back()->setSystemType(artdaq::Fragment::InitFragmentType);
1276 broadcastFragments_(begin_run_fragments_);
1282 init_fragments_.push_back(std::move(frag));
1288 TLOG(TLVL_DEBUG) <<
"UpdateArtConfiguration BEGIN";
1289 if (art_pset != current_art_pset_ || !current_art_config_file_)
1291 current_art_pset_ = art_pset;
1292 current_art_config_file_ = std::make_shared<art_config_file>(art_pset );
1294 TLOG(TLVL_DEBUG) <<
"UpdateArtConfiguration END";
1297 std::string artdaq::SharedMemoryEventManager::buildStatisticsString_()
const
1299 std::ostringstream oss;
1300 oss << app_name <<
" statistics:" << std::endl;
1302 artdaq::MonitoredQuantityPtr mqPtr =
1303 artdaq::StatisticsCollection::getInstance().getMonitoredQuantity(EVENTS_RELEASED_STAT_KEY);
1304 if (mqPtr.get() != 0)
1306 artdaq::MonitoredQuantityStats stats;
1307 mqPtr->getStats(stats);
1308 oss <<
" Event statistics: " << stats.recentSampleCount <<
" events released at " << stats.recentSampleRate
1309 <<
" events/sec, effective data rate = "
1310 << (stats.recentValueRate / 1024.0 / 1024.0)
1311 <<
" MB/sec, monitor window = " << stats.recentDuration
1312 <<
" sec, min::max event size = " << (stats.recentValueMin / 1024.0 / 1024.0)
1313 <<
"::" << (stats.recentValueMax / 1024.0 / 1024.0) <<
" MB" << std::endl;
1314 if (stats.recentSampleRate > 0.0)
1316 oss <<
" Average time per event: ";
1317 oss <<
" elapsed time = " << (1.0 / stats.recentSampleRate) <<
" sec" << std::endl;
1321 mqPtr = artdaq::StatisticsCollection::getInstance().getMonitoredQuantity(FRAGMENTS_RECEIVED_STAT_KEY);
1322 if (mqPtr.get() != 0)
1324 artdaq::MonitoredQuantityStats stats;
1325 mqPtr->getStats(stats);
1326 oss <<
" Fragment statistics: " << stats.recentSampleCount <<
" fragments received at " << stats.recentSampleRate
1327 <<
" fragments/sec, effective data rate = "
1328 << (stats.recentValueRate / 1024.0 / 1024.0)
1329 <<
" MB/sec, monitor window = " << stats.recentDuration
1330 <<
" sec, min::max fragment size = " << (stats.recentValueMin / 1024.0 / 1024.0)
1331 <<
"::" << (stats.recentValueMax / 1024.0 / 1024.0) <<
" MB" << std::endl;
1334 oss <<
" Event counts: Run -- " << run_event_count_ <<
" Total, " << run_incomplete_event_count_ <<
" Incomplete."
1335 <<
" Subrun -- " << subrun_event_count_ <<
" Total, " << subrun_incomplete_event_count_ <<
" Incomplete. "
void RunArt(std::shared_ptr< art_config_file > config_file, std::shared_ptr< std::atomic< pid_t >> pid_out)
Run an art instance, recording the return codes and restarting it until the end flag is raised...
void addMonitoredQuantityName(std::string const &statKey)
Add a MonitoredQuantity name to the list.
void AddInitFragment(FragmentPtr &frag)
Set the stored Init fragment, if one has not yet been set already.
void ShutdownArtProcesses(std::set< pid_t > &pids)
Shutdown a set of art processes.
virtual ~SharedMemoryEventManager()
SharedMemoryEventManager Destructor.
Fragment::sequence_id_t sequence_id_t
Copy Fragment::sequence_id_t into local scope.
void ReconfigureArt(fhicl::ParameterSet art_pset, run_id_t newRun=0, int n_art_processes=-1)
Restart all art processes, using the given fhicl code to configure the new art processes.
The RequestSender contains methods used to send data requests and Routing tokens. ...
pid_t StartArtProcess(fhicl::ParameterSet pset)
Start one art process.
RawDataType * WriteFragmentHeader(detail::RawFragmentHeader frag, bool dropIfNoBuffersAvailable=false)
Get a pointer to a reserved memory area for the given Fragment header.
RawEvent::run_id_t run_id_t
Copy RawEvent::run_id_t into local scope.
size_t GetFragmentCount(Fragment::sequence_id_t seqID, Fragment::type_t type=Fragment::InvalidFragmentType)
Get the count of Fragments of a given type in an event.
void UpdateArtConfiguration(fhicl::ParameterSet art_pset)
Updates the internally-stored copy of the art configuration.
void StartArt()
Start all the art processes.
subrun_id_t GetSubrunForSequenceID(Fragment::sequence_id_t seqID)
Get the subrun number that the given Sequence ID would be assigned to.
void rolloverSubrun()
Add a subrun transition immediately after the highest currently define sequence ID.
void sendMetrics()
Send metrics to the MetricManager, if one has been instantiated in the application.
static const std::string FRAGMENTS_RECEIVED_STAT_KEY
Key for Fragments Received MonitoredQuantity.
bool createCollectors(fhicl::ParameterSet const &pset, int defaultReportIntervalFragments, double defaultReportIntervalSeconds, double defaultMonitorWindow, std::string const &primaryStatKeyName)
Create MonitoredQuantity objects for all names registered with the StatisticsHelper.
SharedMemoryEventManager(fhicl::ParameterSet pset, fhicl::ParameterSet art_pset)
SharedMemoryEventManager Constructor.
bool endRun()
Send an EndOfRunFragment to the art thread.
void DoneWritingFragment(detail::RawFragmentHeader frag)
Used to indicate that the given Fragment is now completely in the buffer. Will check for buffer compl...
static const std::string EVENTS_RELEASED_STAT_KEY
Key for the Events Released MonitoredQuantity.
bool endOfData()
Indicate that the end of input has been reached to the art processes.
RawEvent::subrun_id_t subrun_id_t
Copy RawEvent::subrun_id_t into local scope.
void startRun(run_id_t runID)
Start a Run.
size_t GetFragmentCountInBuffer(int buffer, Fragment::type_t type=Fragment::InvalidFragmentType)
Get the count of Fragments of a given type in a buffer.
void CheckPendingBuffers()
Check for buffers which are ready to be marked incomplete and released to art and issue tokens for an...