1 #define TRACE_NAME (app_name + "_FragmentReceiverManager").c_str()
2 #include "artdaq/DAQdata/Globals.hh"
6 #include "artdaq/DAQrate/FragmentReceiverManager.hh"
7 #include "artdaq/TransferPlugins/MakeTransferPlugin.hh"
8 #include "cetlib_except/exception.h"
11 : stop_requested_(false)
14 , source_metric_data_()
15 , source_metric_send_time_()
21 , suppress_noisy_senders_(pset.get<bool>(
"auto_suppression_enabled", true))
22 , suppression_threshold_(pset.get<size_t>(
"max_receive_difference", 50))
23 , receive_timeout_(pset.get<size_t>(
"receive_timeout_usec", 100000))
26 TLOG(TLVL_DEBUG) <<
"Constructor";
27 auto enabled_srcs = pset.get<std::vector<int>>(
"enabled_sources", std::vector<int>());
28 auto enabled_srcs_empty = enabled_srcs.size() == 0;
29 if (enabled_srcs_empty)
31 TLOG(TLVL_INFO) <<
"enabled_sources not specified, assuming all sources enabled.";
35 for (
auto& s : enabled_srcs)
37 enabled_sources_[s] =
true;
41 auto srcs = pset.get<fhicl::ParameterSet>(
"sources", fhicl::ParameterSet());
42 for (
auto& s : srcs.get_pset_names())
48 auto source_rank = transfer->source_rank();
49 if (enabled_srcs_empty) enabled_sources_[source_rank] =
true;
50 else if (!enabled_sources_.count(source_rank)) enabled_sources_[source_rank] =
false;
51 running_sources_[source_rank] =
false;
52 source_plugins_[source_rank] = std::move(transfer);
53 fragment_store_[source_rank];
54 source_metric_send_time_[source_rank] = std::chrono::steady_clock::now();
55 source_metric_data_[source_rank] = std::pair<size_t, double>();
57 catch (cet::exception ex)
59 TLOG(TLVL_WARNING) <<
"cet::exception caught while setting up source " << s <<
": " << ex.what();
61 catch (std::exception ex)
63 TLOG(TLVL_WARNING) <<
"std::exception caught while setting up source " << s <<
": " << ex.what();
67 TLOG(TLVL_WARNING) <<
"Non-cet exception caught while setting up source " << s <<
".";
70 if (srcs.get_pset_names().size() == 0)
72 TLOG(TLVL_ERROR) <<
"No sources configured!";
78 TLOG(TLVL_DEBUG) <<
"Destructor";
79 TLOG(5) <<
"~FragmentReceiverManager: BEGIN: Setting stop_requested to true, frags=" << count() <<
", bytes=" << byteCount();
80 stop_requested_ =
true;
82 TLOG(5) <<
"~FragmentReceiverManager: Notifying all threads";
83 output_cv_.notify_all();
85 TLOG(5) <<
"~FragmentReceiverManager: Joining all threads";
86 for (
auto& s : source_threads_)
88 auto& thread = s.second;
89 if (thread.joinable()) thread.join();
91 TLOG(5) <<
"~FragmentReceiverManager: DONE";
94 bool artdaq::FragmentReceiverManager::fragments_ready_()
const
96 for (
auto& it : fragment_store_)
98 if (!enabled_sources_.count(it.first))
continue;
99 if (!it.second.empty()) {
return true; }
104 int artdaq::FragmentReceiverManager::get_next_source_()
const
107 std::set<int> ready_sources;
108 for (
auto& it : fragment_store_)
110 if (!enabled_sources_.count(it.first))
continue;
111 if (!it.second.empty()) {
112 ready_sources.insert(it.first);
116 if (ready_sources.size()) {
117 auto iter = ready_sources.find(last_source_);
118 if (iter == ready_sources.end() || ++iter == ready_sources.end()) {
119 TLOG(10) <<
"get_next_source returning " << *ready_sources.begin();
120 last_source_ = *ready_sources.begin();
121 return *ready_sources.begin();
124 TLOG(10) <<
"get_next_source returning " << *iter;
125 last_source_ = *iter;
129 TLOG(10) <<
"get_next_source returning -1";
135 for (
auto& source : source_plugins_)
137 auto& rank = source.first;
138 if (enabled_sources_.count(rank))
140 running_sources_[rank] =
true;
142 source_threads_[rank] = boost::thread(&FragmentReceiverManager::runReceiver_,
this, rank);
144 catch (
const boost::exception& e)
146 TLOG(TLVL_ERROR) <<
"Caught boost::exception starting Receiver " << rank <<
" thread: " << boost::diagnostic_information(e) <<
", errno=" << errno;
147 std::cerr <<
"Caught boost::exception starting Receiver " << rank <<
" thread: " << boost::diagnostic_information(e) <<
", errno=" << errno << std::endl;
156 TLOG(5) <<
"recvFragment entered tmo=" << timeout_usec <<
" us";
158 if (timeout_usec == 0) timeout_usec = 1000000;
160 auto ready = fragments_ready_();
162 auto wait_amount = timeout_usec / 1000 > 1000 ? timeout_usec / 1000 : 1000;
163 TLOG(5) <<
"recvFragment fragment_ready_=" << ready <<
" before wait";
164 while (!ready && waited < timeout_usec)
167 std::unique_lock<std::mutex> lck(input_cv_mutex_);
168 input_cv_.wait_for(lck, std::chrono::microseconds(wait_amount));
170 waited += wait_amount;
171 ready = fragments_ready_();
172 if (running_sources().size() == 0)
break;
174 TLOG(5) <<
"recvFragment fragment_ready_=" << ready <<
" after waited=" << waited;
177 TLOG(5) <<
"recvFragment: No fragments ready, returning empty";
179 return std::unique_ptr<Fragment>{};
182 int current_source = get_next_source_();
183 FragmentPtr current_fragment = fragment_store_[current_source].front();
184 output_cv_.notify_all();
185 rank = current_source;
187 if (current_fragment !=
nullptr)
188 TLOG(5) <<
"recvFragment: Done rank=" << rank <<
", fragment size=" << std::to_string(current_fragment->size()) <<
" words, seqId=" << current_fragment->sequenceID();
189 return current_fragment;
194 std::set<int> output;
195 for (
auto& src : running_sources_)
197 if (src.second) output.insert(src.first);
204 std::set<int> output;
205 for (
auto& src : enabled_sources_)
207 if (src.second) output.insert(src.first);
212 void artdaq::FragmentReceiverManager::runReceiver_(
int source_rank)
214 while (!stop_requested_ && enabled_sources_.count(source_rank))
216 TLOG(16) <<
"runReceiver_ " << source_rank <<
": Begin loop";
217 auto is_suppressed = suppress_noisy_senders_ && recv_seq_count_.slotCount(source_rank) > suppression_threshold_ + recv_seq_count_.minCount();
218 while (!stop_requested_ && is_suppressed)
220 TLOG(6) <<
"runReceiver_: Suppressing receiver rank " << source_rank;
221 if (!is_suppressed) input_cv_.notify_all();
224 std::unique_lock<std::mutex> lck(output_cv_mutex_);
225 output_cv_.wait_for(lck, std::chrono::seconds(1));
227 is_suppressed = suppress_noisy_senders_ && recv_seq_count_.slotCount(source_rank) > suppression_threshold_ + recv_seq_count_.minCount();
231 running_sources_[source_rank] =
false;
235 if (fragment_store_[source_rank].GetEndOfData() <= recv_frag_count_.slotCount(source_rank) && !source_plugins_[source_rank]->isRunning())
237 TLOG(TLVL_DEBUG) <<
"runReceiver_: EndOfData conditions satisfied, ending receive loop";
238 running_sources_[source_rank] =
false;
242 auto start_time = std::chrono::steady_clock::now();
243 TLOG(16) <<
"runReceiver_: Calling receiveFragment";
244 auto fragment = std::unique_ptr<Fragment>(
new Fragment());
246 auto ret = source_plugins_[source_rank]->receiveFragment(*fragment, receive_timeout_);
247 TLOG(16) <<
"runReceiver_: Done with receiveFragment, ret=" << ret <<
" (should be " << source_rank <<
")";
248 if (ret != source_rank)
continue;
250 artdaq::detail::RawFragmentHeader hdr;
251 auto ret1 = source_plugins_[source_rank]->receiveFragmentHeader(hdr, receive_timeout_);
252 TLOG(16) <<
"runReceiver_: Done with receiveFragmentHeader, ret1=" << ret1 <<
" (should be " << source_rank <<
")";
254 if (ret1 != source_rank)
continue;
256 fragment->resize(hdr.word_count - hdr.num_words());
257 memcpy(fragment->headerAddress(), &hdr, hdr.num_words() *
sizeof(artdaq::RawDataType));
258 auto ret2 = source_plugins_[source_rank]->receiveFragmentData(fragment->headerAddress() + hdr.num_words(), hdr.word_count - hdr.num_words());
261 TLOG(TLVL_ERROR) <<
"ReceiveFragmentHeader returned " << ret1 <<
", but ReceiveFragmentData returned " << ret2;
267 if (fragment->type() == artdaq::Fragment::EndOfDataFragmentType)
269 TLOG(TLVL_TRACE) <<
"runReceiver_: EndOfData Fragment received!";
270 fragment_store_[source_rank].SetEndOfData(*reinterpret_cast<size_t*>(fragment->dataBegin()));
272 else if (fragment->type() == artdaq::Fragment::DataFragmentType || fragment->type() == artdaq::Fragment::ContainerFragmentType || fragment->isUserFragmentType(fragment->type()))
274 TLOG(TLVL_TRACE) <<
"runReceiver_: Data Fragment received!";
275 recv_frag_count_.incSlot(source_rank);
276 recv_frag_size_.incSlot(source_rank, fragment->size() *
sizeof(RawDataType));
277 recv_seq_count_.setSlot(source_rank, fragment->sequenceID());
284 auto delta_t = std::chrono::duration_cast<std::chrono::duration<double, std::ratio<1>>>(std::chrono::steady_clock::now() - start_time).count();
285 source_metric_data_[source_rank].first += fragment->size() *
sizeof(RawDataType);
286 source_metric_data_[source_rank].second += delta_t;
288 if (metricMan && TimeUtils::GetElapsedTime(source_metric_send_time_[source_rank]) > 1)
290 TLOG(6) <<
"runReceiver_: Sending receive stats";
291 metricMan->sendMetric(
"Data Receive Time From Rank " + std::to_string(source_rank), source_metric_data_[source_rank].second,
"s", 1, MetricMode::Accumulate);
292 metricMan->sendMetric(
"Data Receive Size From Rank " + std::to_string(source_rank), static_cast<unsigned long>(source_metric_data_[source_rank].first),
"B", 1, MetricMode::Accumulate);
293 metricMan->sendMetric(
"Data Receive Rate From Rank " + std::to_string(source_rank), source_metric_data_[source_rank].first / source_metric_data_[source_rank].second,
"B/s", 1, MetricMode::Average);
295 source_metric_send_time_[source_rank] = std::chrono::steady_clock::now();
296 source_metric_data_[source_rank].first = 0;
297 source_metric_data_[source_rank].second = 0.0;
301 fragment_store_[source_rank].emplace_back(std::move(fragment));
302 TLOG(TLVL_TRACE) <<
"runReceiver_: There are now " << fragment_store_[source_rank].size() <<
" Fragments stored from this source";
303 input_cv_.notify_all();
307 running_sources_[source_rank] =
false;
void start_threads()
Start receiver threads for all enabled sources.
std::set< int > enabled_sources() const
Get the list of enabled sources.
virtual ~FragmentReceiverManager()
FragmentReceiverManager Destructor.
This TransferInterface is a Receiver.
FragmentReceiverManager(const fhicl::ParameterSet &ps)
FragmentReceiverManager Constructor.
std::unique_ptr< artdaq::TransferInterface > MakeTransferPlugin(const fhicl::ParameterSet &pset, std::string plugin_label, TransferInterface::Role role)
Load a TransferInterface plugin.
FragmentPtr recvFragment(int &rank, size_t timeout_usec=0)
Receive a Fragment.
std::set< int > running_sources() const
Get the list of sources which are still receiving data.
Value to be returned upon receive timeout.