1 #include "artdaq/DAQrate/TransferTest.hh"
3 #include "artdaq-core/Data/Fragment.hh"
4 #include "artdaq/DAQrate/FragmentReceiverManager.hh"
5 #include "artdaq/DAQrate/DataSenderManager.hh"
7 #define TRACE_NAME "TransferTest"
8 #include "artdaq/DAQdata/Globals.hh"
10 #include "fhiclcpp/make_ParameterSet.h"
15 : senders_(psi.get<int>(
"num_senders"))
16 , receivers_(psi.get<int>(
"num_receivers"))
17 , sending_threads_(psi.get<int>(
"sending_threads", 1))
18 , sends_each_sender_(psi.get<int>(
"sends_per_sender"))
19 , receives_each_receiver_(senders_ * sending_threads_ * sends_each_sender_ / receivers_)
20 , buffer_count_(psi.get<int>(
"buffer_count", 10))
21 , error_count_max_(psi.get<int>(
"max_errors_before_abort", 3))
22 , fragment_size_(psi.get<size_t>(
"fragment_size", 0x100000))
24 , validate_mode_(psi.get<bool>(
"validate_data_mode", false))
25 , partition_number_(psi.get<int>(
"partition_number", rand() % 0x7F))
27 TLOG(10) <<
"CONSTRUCTOR";
28 metricMan = &metricMan_;
30 if (fragment_size_ < artdaq::detail::RawFragmentHeader::num_words() *
sizeof(artdaq::RawDataType))
32 fragment_size_ = artdaq::detail::RawFragmentHeader::num_words() *
sizeof(artdaq::RawDataType);
35 fhicl::ParameterSet metric_pset;
39 metric_pset = psi.get<fhicl::ParameterSet>(
"metrics");
45 std::string name =
"TransferTest" + std::to_string(my_rank);
46 metricMan_.initialize(metric_pset, name);
47 metricMan_.do_start();
51 std::string type(psi.get<std::string>(
"transfer_plugin_type",
"Shmem"));
55 if (senders_ * sends_each_sender_ % receivers_ != 0)
57 TLOG(TLVL_TRACE) <<
"Adding sends so that sends_each_sender * num_sending_ranks is a multiple of num_receiving_ranks" << std::endl;
58 while (senders_ * sends_each_sender_ % receivers_ != 0)
62 receives_each_receiver_ = senders_ * sends_each_sender_ / receivers_;
63 TLOG(TLVL_TRACE) <<
"sends_each_sender is now " << sends_each_sender_ << std::endl;
64 psi.put_or_replace(
"sends_per_sender", sends_each_sender_);
68 std::string hostmap =
"";
69 if (psi.has_key(
"hostmap"))
71 hostmap =
" host_map: @local::hostmap";
75 ss << psi.to_string() << std::endl;
78 for (
int ii = 0; ii < senders_; ++ii)
80 ss <<
"s" << ii <<
": { transferPluginType: " << type <<
" source_rank: " << ii <<
" max_fragment_size_words : " << fragment_size_ <<
" buffer_count : " << buffer_count_ <<
" partition_number : " << partition_number_ << hostmap <<
" }" << std::endl;
82 ss <<
"}" << std::endl <<
" destinations: {";
83 for (
int jj = senders_; jj < senders_ + receivers_; ++jj)
85 ss <<
"d" << jj <<
": { transferPluginType: " << type <<
" destination_rank: " << jj <<
" max_fragment_size_words : " << fragment_size_ <<
" buffer_count : " << buffer_count_ <<
" partition_number : " << partition_number_ << hostmap <<
" }" << std::endl;
87 ss <<
"}" << std::endl;
89 make_ParameterSet(ss.str(), ps_);
92 TLOG(TLVL_DEBUG) <<
"Going to configure with ParameterSet: " << ps_.to_string() << std::endl;
97 TLOG(TLVL_INFO) <<
"runTest BEGIN: " << (my_rank < senders_ ?
"sending" :
"receiving");
98 start_time_ = std::chrono::steady_clock::now();
99 std::pair<size_t, double> result;
100 if (my_rank >= senders_ + receivers_)
return 0;
101 if (my_rank < senders_)
103 std::vector<std::future<std::pair<size_t, double>>> results_futures(sending_threads_);
104 for (
int ii = 0; ii < sending_threads_; ++ii)
106 results_futures[ii] = std::async(std::bind(&TransferTest::do_sending,
this, ii));
108 for (
auto& future : results_futures)
112 auto thisresult = future.get();
113 result.first += thisresult.first;
114 result.second += thisresult.second;
120 result = do_receiving();
122 auto duration = std::chrono::duration_cast<artdaq::TimeUtils::seconds>(std::chrono::steady_clock::now() - start_time_).count();
123 TLOG(TLVL_INFO) << (my_rank < senders_ ?
"Sent " :
"Received ") << result.first <<
" bytes in " << duration <<
" seconds ( " << formatBytes(result.first / duration) <<
"/s )." << std::endl;
124 TLOG(TLVL_INFO) <<
"Rate of " << (my_rank < senders_ ?
"sending" :
"receiving") <<
": " << formatBytes(result.first / result.second) <<
"/s." << std::endl;
125 metricMan_.do_stop();
126 metricMan_.shutdown();
127 TLOG(11) <<
"runTest DONE";
131 std::pair<size_t, double> artdaq::TransferTest::do_sending(
int index)
133 TLOG(7) <<
"do_sending entered RawFragmentHeader::num_words()=" << artdaq::detail::RawFragmentHeader::num_words();
135 size_t totalSize = 0;
136 double totalTime = 0;
139 unsigned data_size_wrds = (fragment_size_ /
sizeof(artdaq::RawDataType)) - artdaq::detail::RawFragmentHeader::num_words();
140 artdaq::Fragment frag(data_size_wrds);
144 artdaq::RawDataType gen_seed = 0;
146 std::generate_n(frag.dataBegin(), data_size_wrds, [&]() {
return ++gen_seed; });
147 for (
size_t ii = 0; ii < frag.dataSize(); ++ii)
149 if (*(frag.dataBegin() + ii) != ii + 1)
151 TLOG(TLVL_ERROR) <<
"Data corruption detected! (" << (*(frag.dataBegin() + ii)) <<
" != " << (ii + 1) <<
") Aborting!";
157 int metric_send_interval = sends_each_sender_ / 1000 > 1 ? sends_each_sender_ / 1000 : 1;
158 auto init_time_metric = 0.0;
159 auto send_time_metric = 0.0;
160 auto after_time_metric = 0.0;
161 auto send_size_metric = 0.0;
162 auto error_count = 0;
164 for (
int ii = 0; ii < sends_each_sender_; ++ii)
166 auto loop_start = std::chrono::steady_clock::now();
167 TLOG(7) <<
"sender rank " << my_rank <<
" #" << ii <<
" resized bytes=" << frag.sizeBytes();
168 totalSize += frag.sizeBytes();
171 frag.setSequenceID(ii * sending_threads_ + index);
172 frag.setFragmentID(my_rank);
173 frag.setSystemType(artdaq::Fragment::DataFragmentType);
180 auto send_start = std::chrono::steady_clock::now();
181 auto stspair = sender.sendFragment(std::move(frag));
182 TLOG(TLVL_DEBUG) <<
"Sender " << my_rank <<
" sending fragment " << ii;
183 auto after_send = std::chrono::steady_clock::now();
184 TLOG(TLVL_TRACE) <<
"Sender " << my_rank <<
" sent fragment " << ii;
190 if (error_count >= error_count_max_)
192 TLOG(TLVL_ERROR) <<
"Too many errors sending fragments! Aborting... (sent=" << ii <<
"/" << sends_each_sender_ <<
")";
193 exit(sends_each_sender_ - ii);
197 frag = artdaq::Fragment(data_size_wrds);
200 artdaq::RawDataType gen_seed = ii + 1;
202 std::generate_n(frag.dataBegin(), data_size_wrds, [&]() {
return ++gen_seed; });
203 for (
size_t jj = 0; jj < frag.dataSize(); ++jj)
205 if (*(frag.dataBegin() + jj) != (ii + 1) + jj + 1)
207 TLOG(TLVL_ERROR) <<
"Input Data corruption detected! (" << *(frag.dataBegin() + jj) <<
" != " << ii + jj + 2 <<
" at position " << ii <<
") Aborting!";
212 TLOG(9) <<
"sender rank " << my_rank <<
" frag replaced";
214 auto total_send_time = std::chrono::duration_cast<artdaq::TimeUtils::seconds>(after_send - send_start).count();
215 totalTime += total_send_time;
216 send_time_metric += total_send_time;
217 send_size_metric += data_size_wrds *
sizeof(artdaq::RawDataType);
218 after_time_metric += std::chrono::duration_cast<artdaq::TimeUtils::seconds>(std::chrono::steady_clock::now() - after_send).count();
219 init_time_metric += std::chrono::duration_cast<artdaq::TimeUtils::seconds>(send_start - loop_start).count();
221 if (metricMan && ii % metric_send_interval == 0)
223 metricMan->sendMetric(
"send_init_time", init_time_metric,
"seconds", 3, MetricMode::Accumulate);
224 metricMan->sendMetric(
"total_send_time", send_time_metric,
"seconds", 3, MetricMode::Accumulate);
225 metricMan->sendMetric(
"after_send_time", after_time_metric,
"seconds", 3, MetricMode::Accumulate);
226 metricMan->sendMetric(
"send_rate", send_size_metric / send_time_metric,
"B/s", 3, MetricMode::Average);
227 init_time_metric = 0.0;
228 send_time_metric = 0.0;
229 after_time_metric = 0.0;
230 send_size_metric = 0.0;
235 return std::make_pair(totalSize, totalTime);
238 std::pair<size_t, double> artdaq::TransferTest::do_receiving()
240 TLOG(7) <<
"do_receiving entered";
243 receiver.start_threads();
244 int counter = receives_each_receiver_;
245 size_t totalSize = 0;
246 double totalTime = 0;
248 int activeSenders = senders_ * sending_threads_;
249 auto end_loop = std::chrono::steady_clock::now();
251 auto recv_size_metric = 0.0;
252 auto recv_time_metric = 0.0;
253 auto input_wait_metric = 0.0;
254 auto init_wait_metric = 0.0;
255 int metric_send_interval = receives_each_receiver_ / 1000 > 1 ? receives_each_receiver_ : 1;
258 while ((activeSenders > 0 || counter > receives_each_receiver_ / 10) && counter > 0)
260 auto start_loop = std::chrono::steady_clock::now();
261 TLOG(7) <<
"do_receiving: Counter is " << counter <<
", calling recvFragment";
263 auto before_receive = std::chrono::steady_clock::now();
265 auto ignoreFragPtr = receiver.recvFragment(senderSlot);
266 auto after_receive = std::chrono::steady_clock::now();
267 init_wait_metric += std::chrono::duration_cast<artdaq::TimeUtils::seconds>(before_receive - start_loop).count();
271 if (ignoreFragPtr->type() == artdaq::Fragment::EndOfDataFragmentType)
273 TLOG(TLVL_INFO) <<
"Receiver " << my_rank <<
" received EndOfData Fragment from Sender " << senderSlot;
275 TLOG(TLVL_DEBUG) <<
"Active Senders is now " << activeSenders;
277 else if (ignoreFragPtr->type() != artdaq::Fragment::DataFragmentType)
279 TLOG(TLVL_WARNING) <<
"Receiver " << my_rank <<
" received Fragment with System type " << artdaq::detail::RawFragmentHeader::SystemTypeToString(ignoreFragPtr->type()) <<
" (Unexpected!)";
285 start_time_ = std::chrono::steady_clock::now();
289 TLOG(TLVL_INFO) <<
"Receiver " << my_rank <<
" received fragment " << receives_each_receiver_ - counter
290 <<
" with seqID " << ignoreFragPtr->sequenceID() <<
" from Sender " << senderSlot <<
" (Expecting " << counter <<
" more)";
291 thisSize = ignoreFragPtr->size() *
sizeof(artdaq::RawDataType);
292 totalSize += thisSize;
295 for (
size_t ii = 0; ii < ignoreFragPtr->dataSize(); ++ii)
297 if (*(ignoreFragPtr->dataBegin() + ii) != ignoreFragPtr->sequenceID() + ii + 1)
299 TLOG(TLVL_ERROR) <<
"Output Data corruption detected! (" << *(ignoreFragPtr->dataBegin() + ii) <<
" != " << (ignoreFragPtr->sequenceID() + ii + 1) <<
" at position " << ii <<
") Aborting!";
305 input_wait_metric += std::chrono::duration_cast<artdaq::TimeUtils::seconds>(after_receive - end_loop).count();
309 TLOG(TLVL_ERROR) <<
"Receiver " << my_rank <<
" detected fatal protocol error! Reducing active sender count by one!" << std::endl;
311 TLOG(TLVL_DEBUG) <<
"Active Senders is now " << activeSenders;
313 TLOG(7) <<
"do_receiving: Recv Loop end, counter is " << counter;
316 auto total_recv_time = std::chrono::duration_cast<artdaq::TimeUtils::seconds>(after_receive - before_receive).count();
317 recv_time_metric += total_recv_time;
318 totalTime += total_recv_time;
319 recv_size_metric += thisSize;
321 if (metricMan && counter % metric_send_interval == 0)
323 metricMan->sendMetric(
"input_wait", input_wait_metric,
"seconds", 3, MetricMode::Accumulate);
324 metricMan->sendMetric(
"recv_init_time", init_wait_metric,
"seconds", 3, MetricMode::Accumulate);
325 metricMan->sendMetric(
"total_recv_time", recv_time_metric,
"seconds", 3, MetricMode::Accumulate);
326 metricMan->sendMetric(
"recv_rate", recv_size_metric / recv_time_metric,
"B/s", 3, MetricMode::Average);
328 input_wait_metric = 0.0;
329 init_wait_metric = 0.0;
330 recv_time_metric = 0.0;
331 recv_size_metric = 0.0;
333 end_loop = std::chrono::steady_clock::now();
338 TLOG(TLVL_ERROR) <<
"Did not receive all expected Fragments! Missing " << counter <<
" Fragments!";
342 return std::make_pair(totalSize, totalTime);
int runTest()
Run the test as configured.
Sends Fragment objects using TransferInterface plugins. Uses Routing Tables if confgiured, otherwise will Round-Robin Fragments to the destinations.
Receives Fragment objects from one or more DataSenderManager instances using TransferInterface plugin...
Value that is to be returned when a Transfer plugin determines that no more data will be arriving...
The send operation completed successfully.
TransferTest(fhicl::ParameterSet psi)
TransferTest Constructor.
For code clarity, things checking for successful receive should check retval >= RECV_SUCCESS.
Value to be returned upon receive timeout.