2 #include "artdaq/DAQrate/SharedMemoryEventManager.hh"
6 #include "artdaq-core/Core/StatisticsCollection.hh"
7 #include "artdaq-core/Utilities/TraceLock.hh"
9 #define TRACE_NAME (app_name + "_SharedMemoryEventManager").c_str()
11 #define TLVL_BUFFER 40
12 #define TLVL_BUFLCK 41
14 #define build_key(seed) ((seed) + ((GetPartitionNumber() + 1) << 16) + (getpid() & 0xFFFF))
16 std::mutex artdaq::SharedMemoryEventManager::sequence_id_mutex_;
17 std::mutex artdaq::SharedMemoryEventManager::subrun_event_map_mutex_;
24 : SharedMemoryManager(pset.get<uint32_t>(
"shared_memory_key", build_key(0xEE000000)),
25 pset.get<size_t>(
"buffer_count"),
26 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"),
27 pset.get<size_t>(
"stale_buffer_timeout_usec", pset.get<size_t>(
"event_queue_wait_time", 5) * 1000000),
28 !pset.get<bool>(
"broadcast_mode", false))
29 , num_art_processes_(pset.get<size_t>(
"art_analyzer_count", 1))
30 , num_fragments_per_event_(pset.get<size_t>(
"expected_fragments_per_event"))
31 , queue_size_(pset.get<size_t>(
"buffer_count"))
33 , max_subrun_event_map_length_(pset.get<size_t>(
"max_subrun_lookup_table_size", 100))
34 , max_event_list_length_(pset.get<size_t>(
"max_event_list_length", 100))
35 , update_run_ids_(pset.get<bool>(
"update_run_ids_on_new_fragment", true))
36 , use_sequence_id_for_event_number_(pset.get<bool>(
"use_sequence_id_for_event_number", true))
37 , overwrite_mode_(!pset.get<bool>(
"use_art", true) || pset.get<bool>(
"overwrite_mode", false) || pset.get<bool>(
"broadcast_mode", false))
38 , init_fragment_count_(pset.get<size_t>(
"init_fragment_count", pset.get<bool>(
"send_init_fragments", true) ? 1 : 0))
40 , incomplete_event_report_interval_ms_(pset.get<int>(
"incomplete_event_report_interval_ms", -1))
41 , last_incomplete_event_report_time_(std::chrono::steady_clock::now())
42 , last_backpressure_report_time_(std::chrono::steady_clock::now())
43 , last_fragment_header_write_time_(std::chrono::steady_clock::now())
44 , broadcast_timeout_ms_(pset.get<int>(
"fragment_broadcast_timeout_ms", 3000))
46 , run_incomplete_event_count_(0)
47 , subrun_event_count_(0)
48 , subrun_incomplete_event_count_(0)
49 , oversize_fragment_count_(0)
50 , 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))
59 , broadcasts_(pset.get<uint32_t>(
"broadcast_shared_memory_key", build_key(0xBB000000)),
60 pset.get<size_t>(
"broadcast_buffer_count", 10),
61 pset.get<size_t>(
"broadcast_buffer_size", 0x100000),
62 pset.get<int>(
"expected_art_event_processing_time_us", 100000) * pset.get<size_t>(
"buffer_count"), false)
64 subrun_event_map_[0] = 1;
65 SetMinWriteSize(
sizeof(detail::RawEventHeader) +
sizeof(detail::RawFragmentHeader));
66 broadcasts_.SetMinWriteSize(
sizeof(detail::RawEventHeader) +
sizeof(detail::RawFragmentHeader));
68 if (!pset.get<
bool>(
"use_art",
true))
70 TLOG(TLVL_INFO) <<
"BEGIN SharedMemoryEventManager CONSTRUCTOR with use_art:false";
71 num_art_processes_ = 0;
75 TLOG(TLVL_INFO) <<
"BEGIN SharedMemoryEventManager CONSTRUCTOR with use_art:true";
76 TLOG(TLVL_TRACE) <<
"art_pset is " << art_pset.to_string();
78 current_art_config_file_ = std::make_shared<art_config_file>(art_pset );
80 if (overwrite_mode_ && num_art_processes_ > 0)
82 TLOG(TLVL_WARNING) <<
"Art is configured to run, but overwrite mode is enabled! Check your configuration if this in unintentional!";
84 else if (overwrite_mode_)
86 TLOG(TLVL_INFO) <<
"Overwrite Mode enabled, no configured art processes at startup";
89 for (
size_t ii = 0; ii < size(); ++ii)
91 buffer_writes_pending_[ii] = 0;
96 throw cet::exception(app_name +
"_SharedMemoryEventManager") <<
"Unable to attach to Shared Memory!";
99 TLOG(TLVL_TRACE) <<
"Setting Writer rank to " << my_rank;
101 TLOG(TLVL_DEBUG) <<
"Writer Rank is " << GetRank();
109 TLOG(TLVL_TRACE) <<
"END CONSTRUCTOR";
114 TLOG(TLVL_TRACE) <<
"DESTRUCTOR";
126 TLOG(TLVL_TRACE) <<
"Destructor END";
129 bool artdaq::SharedMemoryEventManager::AddFragment(detail::RawFragmentHeader frag,
void* dataPtr)
131 TLOG(TLVL_TRACE) <<
"AddFragment(Header, ptr) BEGIN frag.word_count=" << frag.word_count
132 <<
", sequence_id=" << frag.sequence_id;
133 auto buffer = getBufferForSequenceID_(frag.sequence_id,
true, frag.timestamp);
134 TLOG(TLVL_TRACE) <<
"Using buffer " << buffer <<
" for seqid=" << frag.sequence_id;
141 TLOG(TLVL_ERROR) <<
"Dropping event because data taking has already passed this event number: " << frag.sequence_id;
145 auto hdr = getEventHeader_(buffer);
148 hdr->run_id = run_id_;
150 hdr->subrun_id = GetSubrunForSequenceID(frag.sequence_id);
152 TLOG(TLVL_TRACE) <<
"AddFragment before Write calls";
153 Write(buffer, dataPtr, frag.word_count *
sizeof(RawDataType));
155 TLOG(TLVL_TRACE) <<
"Checking for complete event";
156 auto fragmentCount = GetFragmentCount(frag.sequence_id);
157 hdr->is_complete = fragmentCount == num_fragments_per_event_ && buffer_writes_pending_[buffer] == 0;
158 TLOG(TLVL_TRACE) <<
"hdr->is_complete=" << std::boolalpha << hdr->is_complete
159 <<
", fragmentCount=" << fragmentCount
160 <<
", num_fragments_per_event=" << num_fragments_per_event_
161 <<
", buffer_writes_pending_[buffer]=" << buffer_writes_pending_[buffer];
163 complete_buffer_(buffer);
166 requests_->SendRequest(
true);
169 TLOG(TLVL_TRACE) <<
"AddFragment END";
170 statsHelper_.addSample(FRAGMENTS_RECEIVED_STAT_KEY, frag.word_count *
sizeof(RawDataType));
174 bool artdaq::SharedMemoryEventManager::AddFragment(FragmentPtr frag,
size_t timeout_usec, FragmentPtr& outfrag)
176 TLOG(TLVL_TRACE) <<
"AddFragment(FragmentPtr) BEGIN";
177 auto hdr = *
reinterpret_cast<detail::RawFragmentHeader*
>(frag->headerAddress());
178 auto data = frag->headerAddress();
179 auto start = std::chrono::steady_clock::now();
181 while (!sts && TimeUtils::GetElapsedTimeMicroseconds(start) < timeout_usec)
183 sts = AddFragment(hdr, data);
191 outfrag = std::move(frag);
193 TLOG(TLVL_TRACE) <<
"AddFragment(FragmentPtr) RETURN " << std::boolalpha << sts;
199 TLOG(14) <<
"WriteFragmentHeader BEGIN";
200 auto buffer = getBufferForSequenceID_(frag.sequence_id,
true, frag.timestamp);
204 if (buffer == -1 && !dropIfNoBuffersAvailable)
206 std::unique_lock<std::mutex> bp_lk(sequence_id_mutex_);
207 if (TimeUtils::GetElapsedTime(last_backpressure_report_time_) > 1.0)
209 TLOG(TLVL_WARNING) << app_name <<
": Back-pressure condition: All Shared Memory buffers have been full for " << TimeUtils::GetElapsedTime(last_fragment_header_write_time_) <<
" s!";
210 last_backpressure_report_time_ = std::chrono::steady_clock::now();
216 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.";
220 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.";
222 dropped_data_[frag.fragment_id] = std::make_unique<Fragment>(frag.word_count - frag.num_words());
224 TLOG(TLVL_DEBUG + 3) <<
"Dropping fragment with sequence id " << frag.sequence_id <<
" and fragment id " << frag.fragment_id <<
" into "
225 <<
static_cast<void*
>(dropped_data_[frag.fragment_id]->dataBegin()) <<
" sz=" << dropped_data_[frag.fragment_id]->dataSizeBytes();
227 return dropped_data_[frag.fragment_id]->dataBegin();
230 last_backpressure_report_time_ = std::chrono::steady_clock::now();
231 last_fragment_header_write_time_ = std::chrono::steady_clock::now();
233 buffer_writes_pending_[buffer]++;
237 metricMan->sendMetric(
"Input Fragment Rate", 1,
"Fragments/s", 1, MetricMode::Rate);
240 TLOG(TLVL_BUFLCK) <<
"WriteFragmentHeader: obtaining buffer_mutexes lock for buffer " << buffer;
242 std::unique_lock<std::mutex> lk(buffer_mutexes_[buffer]);
244 TLOG(TLVL_BUFLCK) <<
"WriteFragmentHeader: obtained buffer_mutexes lock for buffer " << buffer;
247 auto hdrpos =
reinterpret_cast<RawDataType*
>(GetWritePos(buffer));
248 Write(buffer, &frag, frag.num_words() *
sizeof(RawDataType));
250 auto pos =
reinterpret_cast<RawDataType*
>(GetWritePos(buffer));
251 if (frag.word_count - frag.num_words() > 0)
253 auto sts = IncrementWritePos(buffer, (frag.word_count - frag.num_words()) *
sizeof(RawDataType));
257 reinterpret_cast<detail::RawFragmentHeader*
>(hdrpos)->word_count = frag.num_words();
258 reinterpret_cast<detail::RawFragmentHeader*
>(hdrpos)->type = Fragment::InvalidFragmentType;
259 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)";
260 dropped_data_[frag.fragment_id] = std::make_unique<Fragment>(frag.word_count - frag.num_words());
262 oversize_fragment_count_++;
264 if (maximum_oversize_fragment_count_ > 0 && oversize_fragment_count_ >= maximum_oversize_fragment_count_)
266 throw cet::exception(
"Too many over-size Fragments received! Please adjust max_event_size_bytes or max_fragment_size_bytes!");
269 TLOG(TLVL_DEBUG + 3) <<
"Dropping over-size fragment with sequence id " << frag.sequence_id <<
" and fragment id " << frag.fragment_id
270 <<
" into " <<
static_cast<void*
>(dropped_data_[frag.fragment_id]->dataBegin());
271 return dropped_data_[frag.fragment_id]->dataBegin();
274 TLOG(14) <<
"WriteFragmentHeader END";
280 TLOG(TLVL_TRACE) <<
"DoneWritingFragment BEGIN";
281 auto buffer = getBufferForSequenceID_(frag.sequence_id,
false, frag.timestamp);
284 Detach(
true,
"SharedMemoryEventManager",
285 "getBufferForSequenceID_ returned -1 in DoneWritingFragment. This indicates a possible mismatch between expected Fragment count and the actual number of Fragments received.");
287 if (buffer == -2) {
return; }
289 statsHelper_.addSample(FRAGMENTS_RECEIVED_STAT_KEY, frag.word_count *
sizeof(RawDataType));
291 TLOG(TLVL_BUFLCK) <<
"DoneWritingFragment: obtaining buffer_mutexes lock for buffer " << buffer;
293 std::unique_lock<std::mutex> lk(buffer_mutexes_[buffer]);
295 TLOG(TLVL_BUFLCK) <<
"DoneWritingFragment: obtained buffer_mutexes lock for buffer " << buffer;
299 TLOG(TLVL_DEBUG) <<
"DoneWritingFragment: Received Fragment with sequence ID " << frag.sequence_id <<
" and fragment id " << frag.fragment_id <<
" (type " <<
static_cast<int>(frag.type) <<
")";
300 auto hdr = getEventHeader_(buffer);
303 hdr->run_id = run_id_;
305 hdr->subrun_id = GetSubrunForSequenceID(frag.sequence_id);
307 TLOG(TLVL_TRACE) <<
"DoneWritingFragment: Updating buffer touch time";
310 buffer_writes_pending_[buffer]--;
311 if (buffer_writes_pending_[buffer] != 0)
313 TLOG(TLVL_TRACE) <<
"Done writing fragment, but there's another writer. Not doing bookkeeping steps.";
316 TLOG(TLVL_TRACE) <<
"Done writing fragment, and no other writer. Doing bookkeeping steps.";
317 auto frag_count = GetFragmentCount(frag.sequence_id);
318 hdr->is_complete = frag_count >= num_fragments_per_event_;
320 if (frag_count > num_fragments_per_event_)
322 TLOG(TLVL_WARNING) <<
"DoneWritingFragment: This Event has more Fragments ( " << frag_count <<
" ) than specified in configuration ( " << num_fragments_per_event_ <<
" )!"
323 <<
" This is probably due to a misconfiguration and is *not* a reliable mode!";
326 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_;
327 #if ART_SUPPORTS_DUPLICATE_EVENTS
328 if (!hdr->is_complete && released_incomplete_events_.count(frag.sequence_id))
330 hdr->is_complete = frag_count >= released_incomplete_events_[frag.sequence_id] && buffer_writes_pending_[buffer] == 0;
335 complete_buffer_(buffer);
338 requests_->SendRequest(
true);
340 TLOG(TLVL_TRACE) <<
"DoneWritingFragment END";
345 return GetFragmentCountInBuffer(getBufferForSequenceID_(seqID,
false), type);
354 ResetReadPos(buffer);
355 IncrementReadPos(buffer,
sizeof(detail::RawEventHeader));
359 while (MoreDataInBuffer(buffer))
361 auto fragHdr =
reinterpret_cast<artdaq::detail::RawFragmentHeader*
>(GetReadPos(buffer));
362 IncrementReadPos(buffer, fragHdr->word_count *
sizeof(RawDataType));
363 if (type != Fragment::InvalidFragmentType && fragHdr->type != type)
367 TLOG(TLVL_TRACE) <<
"Adding Fragment with size=" << fragHdr->word_count <<
" to Fragment count";
378 auto start_time = std::chrono::steady_clock::now();
380 TLOG(TLVL_INFO) <<
"Starting art process with config file " << config_file->getFileName();
386 char* filename =
new char[config_file->getFileName().length() + 1];
387 memcpy(filename, config_file->getFileName().c_str(), config_file->getFileName().length());
388 filename[config_file->getFileName().length()] =
'\0';
391 std::string debugArgS =
"--config-out=" + app_name +
"_art.out";
392 char* debugArg =
new char[debugArgS.length() + 1];
393 memcpy(debugArg, debugArgS.c_str(), debugArgS.length());
394 debugArg[debugArgS.length()] =
'\0';
396 std::vector<char*> args{
const_cast<char*
>(
"art"), const_cast<char*>(
"-c"), filename, debugArg, NULL};
398 std::vector<char*> args{
const_cast<char*
>(
"art"), const_cast<char*>(
"-c"), filename,
nullptr};
408 std::string envVarKey =
"ARTDAQ_PARTITION_NUMBER";
409 std::string envVarValue = std::to_string(GetPartitionNumber());
410 if (setenv(envVarKey.c_str(), envVarValue.c_str(), 1) != 0)
412 TLOG(TLVL_ERROR) <<
"Error setting environment variable \"" << envVarKey
413 <<
"\" in the environment of a child art process. "
414 <<
"This may result in incorrect TCP port number "
415 <<
"assignments or other issues, and data may "
416 <<
"not flow through the system correctly.";
418 envVarKey =
"ARTDAQ_APPLICATION_NAME";
419 envVarValue = app_name;
420 if (setenv(envVarKey.c_str(), envVarValue.c_str(), 1) != 0)
422 TLOG(TLVL_DEBUG) <<
"Error setting environment variable \"" << envVarKey
423 <<
"\" in the environment of a child art process. ";
425 envVarKey =
"ARTDAQ_RANK";
426 envVarValue = std::to_string(my_rank);
427 if (setenv(envVarKey.c_str(), envVarValue.c_str(), 1) != 0)
429 TLOG(TLVL_DEBUG) <<
"Error setting environment variable \"" << envVarKey
430 <<
"\" in the environment of a child art process. ";
433 execvp(
"art", &args[0]);
442 std::cout <<
"Please run the following command in a separate terminal:" << std::endl
443 <<
"art -c " << config_file->getFileName() << std::endl
444 <<
"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
445 <<
"Finally, return to this window and enter the pid: " << std::endl;
450 TLOG(TLVL_INFO) <<
"PID of new art process is " << pid;
452 std::unique_lock<std::mutex> lk(art_process_mutex_);
453 art_processes_.insert(pid);
456 auto sts = waitid(P_PID, pid, &status, WEXITED);
457 TLOG(TLVL_INFO) <<
"Removing PID " << pid <<
" from process list";
459 std::unique_lock<std::mutex> lk(art_process_mutex_);
460 art_processes_.erase(pid);
464 TLOG(TLVL_WARNING) <<
"Error occurred in waitid for art process " << pid <<
": " << errno <<
" (" << strerror(errno) <<
").";
466 else if (status.si_code == CLD_EXITED && status.si_status == 0)
468 TLOG(TLVL_INFO) <<
"art process " << pid <<
" exited normally, " << (restart_art_ ?
"restarting" :
"not restarting");
472 auto art_lifetime = TimeUtils::GetElapsedTime(start_time);
473 if (art_lifetime < minimum_art_lifetime_s_)
475 restart_art_ =
false;
478 auto exit_type =
"exited with status code";
479 switch (status.si_code)
483 exit_type =
"was killed with signal";
490 TLOG((restart_art_ ? TLVL_WARNING : TLVL_ERROR))
491 <<
"art process " << pid <<
" " << exit_type <<
" " << status.si_status
492 << (status.si_code == CLD_DUMPED ?
" (core dumped)" :
"")
493 <<
" after running for " << std::setprecision(2) << std::fixed << art_lifetime <<
" seconds, "
494 << (restart_art_ ?
"restarting" :
"not restarting");
496 }
while (restart_art_);
501 restart_art_ = always_restart_art_;
502 if (num_art_processes_ == 0)
506 for (
size_t ii = 0; ii < num_art_processes_; ++ii)
508 StartArtProcess(current_art_pset_);
514 static std::mutex start_art_mutex;
515 std::unique_lock<std::mutex> lk(start_art_mutex);
517 restart_art_ = always_restart_art_;
518 auto initialCount = GetAttachedCount();
519 auto startTime = std::chrono::steady_clock::now();
521 if (pset != current_art_pset_ || !current_art_config_file_)
523 current_art_pset_ = pset;
524 current_art_config_file_ = std::make_shared<art_config_file>(pset );
526 std::shared_ptr<std::atomic<pid_t>> pid(
new std::atomic<pid_t>(-1));
527 boost::thread thread([=] { RunArt(current_art_config_file_, pid); });
530 auto currentCount = GetAttachedCount() - initialCount;
531 while ((currentCount < 1 || *pid <= 0) && (TimeUtils::GetElapsedTime(startTime) < 5 || manual_art_))
534 currentCount = GetAttachedCount() - initialCount;
536 if ((currentCount < 1 || *pid <= 0) && manual_art_)
538 TLOG(TLVL_WARNING) <<
"Manually-started art process has not connected to shared memory or has bad PID: connected:" << currentCount <<
", PID:" << pid;
541 if (currentCount < 1 || *pid <= 0)
543 TLOG(TLVL_WARNING) <<
"art process has not started after 5s. Check art configuration!"
544 <<
" (pid=" << *pid <<
", attachedCount=" << currentCount <<
")";
548 TLOG(TLVL_INFO) << std::setw(4) << std::fixed <<
"art initialization took "
549 << TimeUtils::GetElapsedTime(startTime) <<
" seconds.";
556 restart_art_ =
false;
560 auto check_pids = [&](
bool print) {
561 std::unique_lock<std::mutex> lk(art_process_mutex_);
562 for (
auto pid = pids.begin(); pid != pids.end();)
568 TLOG(TLVL_WARNING) <<
"Removing an invalid PID (" << *pid
569 <<
") from the shutdown list.";
570 pid = pids.erase(pid);
572 else if (kill(*pid, 0) < 0)
574 pid = pids.erase(pid);
580 std::cout << *pid <<
" ";
586 auto count_pids = [&]() {
587 std::unique_lock<std::mutex> lk(art_process_mutex_);
591 if (count_pids() == 0)
593 TLOG(14) <<
"All art processes already exited, nothing to do.";
600 int graceful_wait_ms = art_event_processing_time_us_ * size() * 10 / 1000;
601 int gentle_wait_ms = art_event_processing_time_us_ * size() * 2 / 1000;
602 int int_wait_ms = art_event_processing_time_us_ * size() / 1000;
603 auto shutdown_start = std::chrono::steady_clock::now();
605 if (!overwrite_mode_)
607 TLOG(TLVL_TRACE) <<
"Waiting up to " << graceful_wait_ms <<
" ms for all art processes to exit gracefully";
608 for (
int ii = 0; ii < graceful_wait_ms; ++ii)
613 if (count_pids() == 0)
615 TLOG(TLVL_INFO) <<
"All art processes exited after " << TimeUtils::GetElapsedTimeMilliseconds(shutdown_start) <<
" ms.";
622 TLOG(TLVL_TRACE) <<
"Gently informing art processes that it is time to shut down";
623 std::unique_lock<std::mutex> lk(art_process_mutex_);
624 for (
auto pid : pids)
626 TLOG(TLVL_TRACE) <<
"Sending SIGQUIT to pid " << pid;
631 TLOG(TLVL_TRACE) <<
"Waiting up to " << gentle_wait_ms <<
" ms for all art processes to exit from SIGQUIT";
632 for (
int ii = 0; ii < gentle_wait_ms; ++ii)
637 if (count_pids() == 0)
639 TLOG(TLVL_INFO) <<
"All art processes exited after " << TimeUtils::GetElapsedTimeMilliseconds(shutdown_start) <<
" ms (SIGQUIT).";
645 TLOG(TLVL_TRACE) <<
"Insisting that the art processes shut down";
646 std::unique_lock<std::mutex> lk(art_process_mutex_);
647 for (
auto pid : pids)
653 TLOG(TLVL_TRACE) <<
"Waiting up to " << int_wait_ms <<
" ms for all art processes to exit from SIGINT";
654 for (
int ii = 0; ii < int_wait_ms; ++ii)
660 if (count_pids() == 0)
662 TLOG(TLVL_INFO) <<
"All art processes exited after " << TimeUtils::GetElapsedTimeMilliseconds(shutdown_start) <<
" ms (SIGINT).";
667 TLOG(TLVL_TRACE) <<
"Killing remaning art processes with extreme prejudice";
668 while (count_pids() > 0)
671 std::unique_lock<std::mutex> lk(art_process_mutex_);
672 kill(*pids.begin(), SIGKILL);
677 TLOG(TLVL_INFO) <<
"All art processes exited after " << TimeUtils::GetElapsedTimeMilliseconds(shutdown_start) <<
" ms (SIGKILL).";
681 std::cout <<
"Please shut down all art processes, then hit return/enter" << std::endl;
682 while (count_pids() > 0)
684 std::cout <<
"The following PIDs are running: ";
686 std::cout << std::endl;
695 TLOG(TLVL_DEBUG) <<
"ReconfigureArt BEGIN";
696 if (restart_art_ || !always_restart_art_)
700 for (
size_t ii = 0; ii < broadcasts_.size(); ++ii)
702 broadcasts_.MarkBufferEmpty(ii,
true);
706 newRun = run_id_ + 1;
709 if (art_pset != current_art_pset_ || !current_art_config_file_)
711 current_art_pset_ = art_pset;
712 current_art_config_file_ = std::make_shared<art_config_file>(art_pset );
715 if (n_art_processes != -1)
717 TLOG(TLVL_INFO) <<
"Setting number of art processes to " << n_art_processes;
718 num_art_processes_ = n_art_processes;
721 TLOG(TLVL_DEBUG) <<
"ReconfigureArt END";
727 init_fragments_.clear();
728 TLOG(TLVL_DEBUG) <<
"SharedMemoryEventManager::endOfData";
729 restart_art_ =
false;
731 size_t initialStoreSize = GetIncompleteEventCount();
732 TLOG(TLVL_DEBUG) <<
"endOfData: Flushing " << initialStoreSize
733 <<
" stale events from the SharedMemoryEventManager.";
734 int counter = initialStoreSize;
735 while (!active_buffers_.empty() && counter > 0)
737 complete_buffer_(*active_buffers_.begin());
740 TLOG(TLVL_DEBUG) <<
"endOfData: Done flushing, there are now " << GetIncompleteEventCount()
741 <<
" stale events in the SharedMemoryEventManager.";
743 TLOG(TLVL_DEBUG) <<
"Waiting for " << (ReadReadyCount() + (size() - WriteReadyCount(overwrite_mode_))) <<
" outstanding buffers...";
744 auto start = std::chrono::steady_clock::now();
745 auto lastReadCount = ReadReadyCount() + (size() - WriteReadyCount(overwrite_mode_));
746 auto end_of_data_wait_us = art_event_processing_time_us_ * (lastReadCount > 0 ? lastReadCount : 1);
748 auto outstanding_buffer_wait_time = art_event_processing_time_us_ > 100000 ? 100000 : art_event_processing_time_us_;
751 while (lastReadCount > 0 && (end_of_data_wait_us == 0 || TimeUtils::GetElapsedTimeMicroseconds(start) < end_of_data_wait_us) && get_art_process_count_() > 0)
753 auto temp = ReadReadyCount() + (size() - WriteReadyCount(overwrite_mode_));
754 if (temp != lastReadCount)
756 TLOG(TLVL_TRACE) <<
"Waiting for " << temp <<
" outstanding buffers...";
757 lastReadCount = temp;
758 start = std::chrono::steady_clock::now();
760 if (lastReadCount > 0)
762 TLOG(19) <<
"About to sleep " << outstanding_buffer_wait_time <<
" us - lastReadCount=" << lastReadCount <<
" size=" << size() <<
" end_of_data_wait_us=" << end_of_data_wait_us;
763 usleep(outstanding_buffer_wait_time);
767 TLOG(TLVL_DEBUG) <<
"endOfData: After wait for outstanding buffers. Still outstanding: " << lastReadCount <<
", time waited: "
768 << TimeUtils::GetElapsedTime(start) <<
" s / " << (end_of_data_wait_us / 1000000.0) <<
" s, art process count: " << get_art_process_count_();
770 TLOG(TLVL_DEBUG) <<
"endOfData: Broadcasting EndOfData Fragment";
771 FragmentPtrs broadcast;
772 broadcast.emplace_back(Fragment::eodFrag(GetBufferCount()));
773 bool success = broadcastFragments_(broadcast);
776 TLOG(TLVL_DEBUG) <<
"endOfData: Clearing buffers to make room for EndOfData Fragment";
777 for (
size_t ii = 0; ii < broadcasts_.size(); ++ii)
779 broadcasts_.MarkBufferEmpty(ii,
true);
781 broadcastFragments_(broadcast);
783 auto endOfDataProcessingStart = std::chrono::steady_clock::now();
784 while (get_art_process_count_() > 0)
786 TLOG(TLVL_DEBUG) <<
"There are " << get_art_process_count_() <<
" art processes remaining. Proceeding to shutdown.";
788 ShutdownArtProcesses(art_processes_);
790 TLOG(TLVL_DEBUG) <<
"It took " << TimeUtils::GetElapsedTime(endOfDataProcessingStart) <<
" s for all art processes to close after sending EndOfData Fragment";
792 ResetAttachedCount();
794 TLOG(TLVL_DEBUG) <<
"endOfData: Clearing buffers";
795 for (
size_t ii = 0; ii < size(); ++ii)
797 MarkBufferEmpty(ii,
true);
805 released_events_.clear();
806 released_incomplete_events_.clear();
808 TLOG(TLVL_DEBUG) <<
"endOfData END";
809 TLOG(TLVL_INFO) <<
"EndOfData Complete. There were " << GetLastSeenBufferID() <<
" buffers processed.";
816 init_fragments_.clear();
817 statsHelper_.resetStatistics();
818 TLOG(TLVL_TRACE) <<
"startRun: Clearing broadcast buffers";
819 for (
size_t ii = 0; ii < broadcasts_.size(); ++ii)
821 broadcasts_.MarkBufferEmpty(ii,
true);
823 released_events_.clear();
824 released_incomplete_events_.clear();
828 std::unique_lock<std::mutex> lk(subrun_event_map_mutex_);
829 subrun_event_map_.clear();
830 subrun_event_map_[0] = 1;
832 run_event_count_ = 0;
833 run_incomplete_event_count_ = 0;
834 requests_ = std::make_unique<RequestSender>(data_pset_);
837 requests_->SetRunNumber(static_cast<uint32_t>(run_id_));
838 requests_->SendRoutingToken(queue_size_, run_id_);
840 TLOG(TLVL_DEBUG) <<
"Starting run " << run_id_
841 <<
", max queue size = "
844 << GetLockedBufferCount();
847 metricMan->sendMetric(
"Run Number", static_cast<uint64_t>(run_id_),
"Run", 1, MetricMode::LastPoint | MetricMode::Persist);
853 TLOG(TLVL_INFO) <<
"Ending run " << run_id_;
854 FragmentPtr endOfRunFrag(
new Fragment(static_cast<size_t>(ceil(
sizeof(my_rank) /
855 static_cast<double>(
sizeof(Fragment::value_type))))));
857 TLOG(TLVL_DEBUG) <<
"Shutting down RequestSender";
858 requests_.reset(
nullptr);
860 TLOG(TLVL_DEBUG) <<
"Broadcasting EndOfRun Fragment";
861 endOfRunFrag->setSystemType(Fragment::EndOfRunFragmentType);
862 *endOfRunFrag->dataBegin() = my_rank;
863 FragmentPtrs broadcast;
864 broadcast.emplace_back(std::move(endOfRunFrag));
865 broadcastFragments_(broadcast);
867 TLOG(TLVL_INFO) <<
"Run " << run_id_ <<
" has ended. There were " << run_event_count_ <<
" events in this run.";
868 run_event_count_ = 0;
869 run_incomplete_event_count_ = 0;
870 oversize_fragment_count_ = 0;
872 std::unique_lock<std::mutex> lk(subrun_event_map_mutex_);
873 subrun_event_map_.clear();
874 subrun_event_map_[0] = 1;
882 if (boundary == 0 || boundary == Fragment::InvalidSequenceID)
887 std::unique_lock<std::mutex> lk(subrun_event_map_mutex_);
889 TLOG(TLVL_INFO) <<
"Will roll over to subrun " << subrun <<
" when I reach Sequence ID " << boundary;
890 subrun_event_map_[boundary] = subrun;
891 while (subrun_event_map_.size() > max_subrun_event_map_length_)
893 subrun_event_map_.erase(subrun_event_map_.begin());
899 Fragment::sequence_id_t seqID = 0;
902 std::unique_lock<std::mutex> lk(subrun_event_map_mutex_);
903 for (
auto& it : subrun_event_map_)
905 if (it.first >= seqID)
907 seqID = it.first + 1;
909 if (it.second >= subrun)
911 subrun = it.second + 1;
915 rolloverSubrun(seqID, subrun);
922 metricMan->sendMetric(
"Incomplete Event Count", GetIncompleteEventCount(),
"events", 1, MetricMode::LastPoint);
923 metricMan->sendMetric(
"Pending Event Count", GetPendingEventCount(),
"events", 1, MetricMode::LastPoint);
926 if (incomplete_event_report_interval_ms_ > 0 && (GetLockedBufferCount() != 0u))
928 if (TimeUtils::GetElapsedTimeMilliseconds(last_incomplete_event_report_time_) < static_cast<size_t>(incomplete_event_report_interval_ms_))
933 last_incomplete_event_report_time_ = std::chrono::steady_clock::now();
934 std::ostringstream oss;
935 oss <<
"Incomplete Events (" << num_fragments_per_event_ <<
"): ";
936 for (
auto& ev : active_buffers_)
938 auto hdr = getEventHeader_(ev);
939 oss << hdr->sequence_id <<
" (" << GetFragmentCount(hdr->sequence_id) <<
"), ";
941 TLOG(TLVL_DEBUG) << oss.str();
945 bool artdaq::SharedMemoryEventManager::broadcastFragments_(FragmentPtrs& frags)
949 TLOG(TLVL_ERROR) <<
"Requested broadcast but no Fragments given!";
952 TLOG(TLVL_DEBUG) <<
"Broadcasting Fragments with seqID=" << frags.front()->sequenceID()
953 <<
", type " << detail::RawFragmentHeader::SystemTypeToString(frags.front()->type())
954 <<
", size=" << frags.front()->sizeBytes() <<
"B.";
955 auto buffer = broadcasts_.GetBufferForWriting(
false);
956 TLOG(TLVL_DEBUG) <<
"broadcastFragments_: after getting buffer 1st buffer=" << buffer;
957 auto start_time = std::chrono::steady_clock::now();
958 while (buffer == -1 && TimeUtils::GetElapsedTimeMilliseconds(start_time) < static_cast<size_t>(broadcast_timeout_ms_))
961 buffer = broadcasts_.GetBufferForWriting(
false);
963 TLOG(TLVL_DEBUG) <<
"broadcastFragments_: after getting buffer w/timeout, buffer=" << buffer <<
", elapsed time=" << TimeUtils::GetElapsedTime(start_time) <<
" s.";
966 TLOG(TLVL_ERROR) <<
"Broadcast of fragment type " << frags.front()->typeString() <<
" failed due to timeout waiting for buffer!";
970 TLOG(TLVL_DEBUG) <<
"broadcastFragments_: Filling in RawEventHeader";
971 auto hdr =
reinterpret_cast<detail::RawEventHeader*
>(broadcasts_.GetBufferStart(buffer));
972 hdr->run_id = run_id_;
973 hdr->subrun_id = GetSubrunForSequenceID(frags.front()->sequenceID());
974 hdr->sequence_id = frags.front()->sequenceID();
975 hdr->is_complete =
true;
976 broadcasts_.IncrementWritePos(buffer,
sizeof(detail::RawEventHeader));
978 for (
auto& frag : frags)
980 TLOG(TLVL_DEBUG) <<
"broadcastFragments_ before Write calls";
981 if (frag->sequenceID() != hdr->sequence_id || frag->type() != frags.front()->type())
983 TLOG(TLVL_WARNING) <<
"Not sending fragment because its SequenceID or Type disagrees with leading Fragment";
986 broadcasts_.Write(buffer, frag->headerAddress(), frag->size() *
sizeof(RawDataType));
989 TLOG(TLVL_DEBUG) <<
"broadcastFragments_ Marking buffer full";
990 broadcasts_.MarkBufferFull(buffer, -1);
991 TLOG(TLVL_DEBUG) <<
"broadcastFragment_s Complete";
995 artdaq::detail::RawEventHeader* artdaq::SharedMemoryEventManager::getEventHeader_(
int buffer)
997 return reinterpret_cast<detail::RawEventHeader*
>(GetBufferStart(buffer));
1002 std::unique_lock<std::mutex> lk(subrun_event_map_mutex_);
1004 TLOG(TLVL_TRACE) <<
"GetSubrunForSequenceID BEGIN map size = " << subrun_event_map_.size();
1005 auto it = subrun_event_map_.begin();
1008 while (it->first <= seqID && it != subrun_event_map_.end())
1010 TLOG(TLVL_TRACE) <<
"Map has sequence ID " << it->first <<
", subrun " << it->second <<
" (looking for <= " << seqID <<
")";
1011 subrun = it->second;
1015 TLOG(TLVL_DEBUG) <<
"GetSubrunForSequenceID returning subrun " << subrun <<
" for sequence ID " << seqID;
1019 int artdaq::SharedMemoryEventManager::getBufferForSequenceID_(Fragment::sequence_id_t seqID,
bool create_new, Fragment::timestamp_t timestamp)
1021 TLOG(14) <<
"getBufferForSequenceID " << seqID <<
" BEGIN";
1022 std::unique_lock<std::mutex> lk(sequence_id_mutex_);
1024 TLOG(14) <<
"getBufferForSequenceID obtained sequence_id_mutex for seqid=" << seqID;
1026 auto buffers = GetBuffersOwnedByManager();
1027 for (
auto& buf : buffers)
1029 auto hdr = getEventHeader_(buf);
1030 if (hdr->sequence_id == seqID)
1032 TLOG(14) <<
"getBufferForSequenceID " << seqID <<
" returning " << buf;
1037 #if !ART_SUPPORTS_DUPLICATE_EVENTS
1038 if (released_incomplete_events_.count(seqID) != 0u)
1040 TLOG(TLVL_ERROR) <<
"Event " << seqID <<
" has already been marked \"Incomplete\" and sent to art!";
1043 if (released_events_.count(seqID) != 0u)
1045 TLOG(TLVL_ERROR) <<
"Event " << seqID <<
" has already been completed and released to art! Check configuration for inconsistent Fragment count per event!";
1055 check_pending_buffers_(lk);
1056 int new_buffer = GetBufferForWriting(
false);
1058 if (new_buffer == -1)
1060 new_buffer = GetBufferForWriting(overwrite_mode_);
1063 if (new_buffer == -1)
1067 TLOG(TLVL_BUFLCK) <<
"getBufferForSequenceID_: obtaining buffer_mutexes lock for buffer " << new_buffer;
1068 std::unique_lock<std::mutex> buffer_lk(buffer_mutexes_[new_buffer]);
1069 TLOG(TLVL_BUFLCK) <<
"getBufferForSequenceID_: obtained buffer_mutexes lock for buffer " << new_buffer;
1071 auto hdr = getEventHeader_(new_buffer);
1072 hdr->is_complete =
false;
1073 hdr->run_id = run_id_;
1074 hdr->subrun_id = GetSubrunForSequenceID(seqID);
1075 hdr->event_id = use_sequence_id_for_event_number_ ?
static_cast<uint32_t
>(seqID) : static_cast<uint32_t>(timestamp);
1076 hdr->sequence_id = seqID;
1077 hdr->timestamp = timestamp;
1078 buffer_writes_pending_[new_buffer] = 0;
1079 IncrementWritePos(new_buffer,
sizeof(detail::RawEventHeader));
1080 SetMFIteration(
"Sequence ID " + std::to_string(seqID));
1082 TLOG(TLVL_BUFFER) <<
"getBufferForSequenceID placing " << new_buffer <<
" to active.";
1083 active_buffers_.insert(new_buffer);
1084 TLOG(TLVL_BUFFER) <<
"Buffer occupancy now (total,full,reading,empty,pending,active)=("
1086 << ReadReadyCount() <<
","
1087 << WriteReadyCount(
true) - WriteReadyCount(
false) - ReadReadyCount() <<
","
1088 << WriteReadyCount(
false) <<
","
1089 << pending_buffers_.size() <<
","
1090 << active_buffers_.size() <<
")";
1094 if (timestamp != Fragment::InvalidTimestamp)
1096 requests_->AddRequest(seqID, timestamp);
1102 requests_->SendRequest();
1105 TLOG(14) <<
"getBufferForSequenceID " << seqID <<
" returning newly initialized buffer " << new_buffer;
1109 bool artdaq::SharedMemoryEventManager::hasFragments_(
int buffer)
1115 if (!CheckBuffer(buffer, BufferSemaphoreFlags::Writing))
1119 ResetReadPos(buffer);
1120 IncrementReadPos(buffer,
sizeof(detail::RawEventHeader));
1121 return MoreDataInBuffer(buffer);
1124 void artdaq::SharedMemoryEventManager::complete_buffer_(
int buffer)
1126 auto hdr = getEventHeader_(buffer);
1127 if (hdr->is_complete)
1129 TLOG(TLVL_DEBUG) <<
"complete_buffer_: This fragment completes event " << hdr->sequence_id <<
".";
1132 TLOG(TLVL_BUFFER) <<
"complete_buffer_ moving " << buffer <<
" from active to pending.";
1134 TLOG(TLVL_BUFLCK) <<
"complete_buffer_: obtaining sequence_id_mutex lock for seqid=" << hdr->sequence_id;
1135 std::unique_lock<std::mutex> lk(sequence_id_mutex_);
1136 TLOG(TLVL_BUFLCK) <<
"complete_buffer_: obtained sequence_id_mutex lock for seqid=" << hdr->sequence_id;
1137 active_buffers_.erase(buffer);
1138 pending_buffers_.insert(buffer);
1139 released_events_.insert(hdr->sequence_id);
1140 while (released_events_.size() > max_event_list_length_)
1142 released_events_.erase(released_events_.begin());
1145 TLOG(TLVL_BUFFER) <<
"Buffer occupancy now (total,full,reading,empty,pending,active)=("
1147 << ReadReadyCount() <<
","
1148 << WriteReadyCount(
true) - WriteReadyCount(
false) - ReadReadyCount() <<
","
1149 << WriteReadyCount(
false) <<
","
1150 << pending_buffers_.size() <<
","
1151 << active_buffers_.size() <<
")";
1155 requests_->RemoveRequest(hdr->sequence_id);
1158 CheckPendingBuffers();
1161 bool artdaq::SharedMemoryEventManager::bufferComparator(
int bufA,
int bufB)
1163 return getEventHeader_(bufA)->sequence_id < getEventHeader_(bufB)->sequence_id;
1168 TLOG(TLVL_BUFLCK) <<
"CheckPendingBuffers: Obtaining sequence_id_mutex_";
1169 std::unique_lock<std::mutex> lk(sequence_id_mutex_);
1170 TLOG(TLVL_BUFLCK) <<
"CheckPendingBuffers: Obtained sequence_id_mutex_";
1171 check_pending_buffers_(lk);
1174 void artdaq::SharedMemoryEventManager::check_pending_buffers_(std::unique_lock<std::mutex>
const& lock)
1176 TLOG(14) <<
"check_pending_buffers_ BEGIN Locked=" << std::boolalpha << lock.owns_lock();
1178 auto buffers = GetBuffersOwnedByManager();
1179 for (
auto buf : buffers)
1181 if (ResetBuffer(buf) && (pending_buffers_.count(buf) == 0u))
1183 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();
1184 auto hdr = getEventHeader_(buf);
1185 if ((active_buffers_.count(buf) != 0u) && (buffer_writes_pending_[buf].load() == 0 || !running_))
1189 requests_->RemoveRequest(hdr->sequence_id);
1191 TLOG(TLVL_BUFFER) <<
"check_pending_buffers_ moving buffer " << buf <<
" from active to pending";
1192 active_buffers_.erase(buf);
1193 pending_buffers_.insert(buf);
1194 TLOG(TLVL_BUFFER) <<
"Buffer occupancy now (total,full,reading,empty,pending,active)=("
1196 << ReadReadyCount() <<
","
1197 << WriteReadyCount(
true) - WriteReadyCount(
false) - ReadReadyCount() <<
","
1198 << WriteReadyCount(
false) <<
","
1199 << pending_buffers_.size() <<
","
1200 << active_buffers_.size() <<
")";
1202 run_incomplete_event_count_++;
1205 metricMan->sendMetric(
"Incomplete Event Rate", 1,
"events/s", 3, MetricMode::Rate);
1207 if (released_incomplete_events_.count(hdr->sequence_id) == 0u)
1209 released_incomplete_events_[hdr->sequence_id] = num_fragments_per_event_ - GetFragmentCountInBuffer(buf);
1213 released_incomplete_events_[hdr->sequence_id] -= GetFragmentCountInBuffer(buf);
1215 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.";
1220 std::list<int> sorted_buffers(pending_buffers_.begin(), pending_buffers_.end());
1221 sorted_buffers.sort([
this](
int a,
int b) {
return bufferComparator(a, b); });
1224 double eventSize = 0;
1225 for (
auto buf : sorted_buffers)
1227 auto hdr = getEventHeader_(buf);
1228 auto thisEventSize = BufferDataSize(buf);
1230 TLOG(TLVL_DEBUG) <<
"Releasing event " << std::to_string(hdr->sequence_id) <<
" in buffer " << buf <<
" to art, "
1231 <<
"event_size=" << thisEventSize <<
", buffer_size=" << BufferSize();
1232 statsHelper_.addSample(EVENTS_RELEASED_STAT_KEY, thisEventSize);
1234 TLOG(TLVL_BUFFER) <<
"check_pending_buffers_ removing buffer " << buf <<
" moving from pending to full";
1235 MarkBufferFull(buf);
1238 eventSize += thisEventSize;
1239 pending_buffers_.erase(buf);
1240 TLOG(TLVL_BUFFER) <<
"Buffer occupancy now (total,full,reading,empty,pending,active)=("
1242 << ReadReadyCount() <<
","
1243 << WriteReadyCount(
true) - WriteReadyCount(
false) - ReadReadyCount() <<
","
1244 << WriteReadyCount(
false) <<
","
1245 << pending_buffers_.size() <<
","
1246 << active_buffers_.size() <<
")";
1249 if (requests_ && requests_->RoutingTokenSendsEnabled())
1251 TLOG(TLVL_TRACE) <<
"Sent tokens: " << requests_->GetSentTokenCount() <<
", Event count: " << run_event_count_;
1252 auto outstanding_tokens = requests_->GetSentTokenCount() - run_event_count_;
1253 auto available_buffers = WriteReadyCount(overwrite_mode_);
1255 TLOG(TLVL_TRACE) <<
"check_pending_buffers_: outstanding_tokens: " << outstanding_tokens <<
", available_buffers: " << available_buffers
1256 <<
", tokens_to_send: " << available_buffers - outstanding_tokens;
1258 if (available_buffers > outstanding_tokens)
1260 auto tokens_to_send = available_buffers - outstanding_tokens;
1262 while (tokens_to_send > 0)
1264 TLOG(35) <<
"check_pending_buffers_: Sending a Routing Token";
1265 requests_->SendRoutingToken(1, run_id_);
1271 if (statsHelper_.readyToReport())
1273 std::string statString = buildStatisticsString_();
1274 TLOG(TLVL_INFO) << statString;
1279 TLOG(14) <<
"check_pending_buffers_: Sending Metrics";
1280 metricMan->sendMetric(
"Event Rate", counter,
"Events", 1, MetricMode::Rate);
1281 metricMan->sendMetric(
"Data Rate", eventSize,
"Bytes", 1, MetricMode::Rate);
1284 metricMan->sendMetric(
"Average Event Size", eventSize / counter,
"Bytes", 1, MetricMode::Average);
1287 metricMan->sendMetric(
"Events Released to art this run", run_event_count_,
"Events", 1, MetricMode::LastPoint);
1288 metricMan->sendMetric(
"Incomplete Events Released to art this run", run_incomplete_event_count_,
"Events", 1, MetricMode::LastPoint);
1291 metricMan->sendMetric(
"Tokens sent", requests_->GetSentTokenCount(),
"Tokens", 2, MetricMode::LastPoint);
1294 auto bufferReport = GetBufferReport();
1295 int full = 0, empty = 0, writing = 0, reading = 0;
1296 for (
auto& buf : bufferReport)
1300 case BufferSemaphoreFlags::Full:
1303 case BufferSemaphoreFlags::Empty:
1306 case BufferSemaphoreFlags::Writing:
1309 case BufferSemaphoreFlags::Reading:
1314 auto total = size();
1315 TLOG(15) <<
"Buffer usage: full=" << full <<
", empty=" << empty <<
", writing=" << writing <<
", reading=" << reading <<
", total=" << total;
1317 metricMan->sendMetric(
"Shared Memory Full Buffers", full,
"buffers", 2, MetricMode::LastPoint);
1318 metricMan->sendMetric(
"Shared Memory Available Buffers", empty,
"buffers", 2, MetricMode::LastPoint);
1319 metricMan->sendMetric(
"Shared Memory Pending Buffers", writing,
"buffers", 2, MetricMode::LastPoint);
1320 metricMan->sendMetric(
"Shared Memory Reading Buffers", reading,
"buffers", 2, MetricMode::LastPoint);
1323 metricMan->sendMetric(
"Shared Memory Full %", full * 100 / static_cast<double>(total),
"%", 2, MetricMode::LastPoint);
1324 metricMan->sendMetric(
"Shared Memory Available %", empty * 100 / static_cast<double>(total),
"%", 2, MetricMode::LastPoint);
1327 TLOG(14) <<
"check_pending_buffers_ END";
1330 void artdaq::SharedMemoryEventManager::send_init_frags_()
1332 if (init_fragments_.size() >= init_fragment_count_ && init_fragment_count_ > 0)
1334 TLOG(TLVL_INFO) <<
"Broadcasting init fragment to all art subprocesses...";
1337 std::string fileName =
"receiveInitMessage_" + std::to_string(my_rank) +
".bin";
1338 std::fstream ostream(fileName.c_str(), std::ios::out | std::ios::binary);
1339 ostream.write(reinterpret_cast<char*>(init_fragment_->dataBeginBytes()), init_fragment_->dataSizeBytes());
1343 broadcastFragments_(init_fragments_);
1344 TLOG(TLVL_TRACE) <<
"Init Fragment sent";
1346 else if (init_fragment_count_ > 0)
1348 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!";
1353 artdaq::FragmentPtrs begin_run_fragments_;
1354 begin_run_fragments_.emplace_back(
new artdaq::Fragment());
1355 begin_run_fragments_.back()->setSystemType(artdaq::Fragment::InitFragmentType);
1356 broadcastFragments_(begin_run_fragments_);
1362 init_fragments_.push_back(std::move(frag));
1368 TLOG(TLVL_DEBUG) <<
"UpdateArtConfiguration BEGIN";
1369 if (art_pset != current_art_pset_ || !current_art_config_file_)
1371 current_art_pset_ = art_pset;
1372 current_art_config_file_ = std::make_shared<art_config_file>(art_pset );
1374 TLOG(TLVL_DEBUG) <<
"UpdateArtConfiguration END";
1377 std::string artdaq::SharedMemoryEventManager::buildStatisticsString_()
const
1379 std::ostringstream oss;
1380 oss << app_name <<
" statistics:" << std::endl;
1382 artdaq::MonitoredQuantityPtr mqPtr =
1383 artdaq::StatisticsCollection::getInstance().getMonitoredQuantity(EVENTS_RELEASED_STAT_KEY);
1384 if (mqPtr.get() !=
nullptr)
1386 artdaq::MonitoredQuantityStats stats;
1387 mqPtr->getStats(stats);
1388 oss <<
" Event statistics: " << stats.recentSampleCount <<
" events released at " << stats.recentSampleRate
1389 <<
" events/sec, effective data rate = "
1390 << (stats.recentValueRate / 1024.0 / 1024.0)
1391 <<
" MB/sec, monitor window = " << stats.recentDuration
1392 <<
" sec, min::max event size = " << (stats.recentValueMin / 1024.0 / 1024.0)
1393 <<
"::" << (stats.recentValueMax / 1024.0 / 1024.0) <<
" MB" << std::endl;
1394 if (stats.recentSampleRate > 0.0)
1396 oss <<
" Average time per event: ";
1397 oss <<
" elapsed time = " << (1.0 / stats.recentSampleRate) <<
" sec" << std::endl;
1401 mqPtr = artdaq::StatisticsCollection::getInstance().getMonitoredQuantity(FRAGMENTS_RECEIVED_STAT_KEY);
1402 if (mqPtr.get() !=
nullptr)
1404 artdaq::MonitoredQuantityStats stats;
1405 mqPtr->getStats(stats);
1406 oss <<
" Fragment statistics: " << stats.recentSampleCount <<
" fragments received at " << stats.recentSampleRate
1407 <<
" fragments/sec, effective data rate = "
1408 << (stats.recentValueRate / 1024.0 / 1024.0)
1409 <<
" MB/sec, monitor window = " << stats.recentDuration
1410 <<
" sec, min::max fragment size = " << (stats.recentValueMin / 1024.0 / 1024.0)
1411 <<
"::" << (stats.recentValueMax / 1024.0 / 1024.0) <<
" MB" << std::endl;
1414 oss <<
" Event counts: Run -- " << run_event_count_ <<
" Total, " << run_incomplete_event_count_ <<
" Incomplete."
1415 <<
" Subrun -- " << subrun_event_count_ <<
" Total, " << subrun_incomplete_event_count_ <<
" Incomplete. "
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.
void RunArt(const std::shared_ptr< art_config_file > &config_file, const 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...
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.
SharedMemoryEventManager(const fhicl::ParameterSet &pset, fhicl::ParameterSet art_pset)
SharedMemoryEventManager Constructor.
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.
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...