1 #include "artdaq/DAQdata/Globals.hh"
2 #define TRACE_NAME "TransferWrapper"
4 #include "artdaq-core/Data/Fragment.hh"
5 #include "artdaq-core/Utilities/ExceptionHandler.hh"
6 #include "artdaq/ArtModules/detail/TransferWrapper.hh"
7 #include "artdaq/DAQdata/NetMonHeader.hh"
8 #include "artdaq/ExternalComms/MakeCommanderPlugin.hh"
9 #include "artdaq/TransferPlugins/MakeTransferPlugin.hh"
11 #include "cetlib/BasicPluginFactory.h"
12 #include "cetlib_except/exception.h"
13 #include "fhiclcpp/ParameterSet.h"
30 void signal_handler(
int signal)
32 gSignalStatus = signal;
36 : timeoutInUsecs_(pset.get<std::size_t>(
"timeoutInUsecs", 100000))
40 , dispatcherHost_(pset.get<std::string>(
"dispatcherHost",
"localhost"))
41 , dispatcherPort_(pset.get<std::string>(
"dispatcherPort",
"5266"))
42 , serverUrl_(pset.get<std::string>(
"server_url",
"http://" + dispatcherHost_ +
":" + dispatcherPort_ +
"/RPC2"))
43 , maxEventsBeforeInit_(pset.get<std::size_t>(
"maxEventsBeforeInit", 5))
44 , allowedFragmentTypes_(pset.get<std::vector<int>>(
"allowedFragmentTypes", {226, 227, 229}))
45 , runningStateTimeout_(pset.get<
double>(
"dispatcherConnectTimeout", 0))
46 , runningStateInterval_us_(pset.get<
size_t>(
"dispatcherConnectRetryInterval_us", 1000000))
47 , quitOnFragmentIntegrityProblem_(pset.get<
bool>(
"quitOnFragmentIntegrityProblem",
true))
48 , multi_run_mode_(pset.get<
bool>(
"allowMultipleRuns",
false))
49 , monitorRegistered_(
false)
51 std::signal(SIGINT, signal_handler);
57 metricMan->initialize(pset.get<fhicl::ParameterSet>(
"metrics", fhicl::ParameterSet()),
"Online Monitor");
58 metricMan->do_start();
63 ExceptionHandler(ExceptionHandlerRethrow::no,
"TransferWrapper: could not configure metrics");
67 if (runningStateInterval_us_ < 1000)
69 TLOG(TLVL_WARNING) <<
"Invalid value " << runningStateInterval_us_ <<
" us detected for dispatcherConnectRetryInterval_us. Setting to 1000 us";
70 runningStateInterval_us_ = 1000;
72 if (runningStateInterval_us_ > 30000000)
74 TLOG(TLVL_WARNING) <<
"Invalid value " << runningStateInterval_us_ <<
" us detected for dispatcherConnectRetryInterval_us. Setting to 30,000,000 us";
75 runningStateInterval_us_ = 30000000;
78 fhicl::ParameterSet new_pset(pset);
79 if (!new_pset.has_key(
"server_url"))
81 new_pset.put<std::string>(
"server_url", serverUrl_);
90 artdaq::FragmentPtrs fragmentPtrs;
91 bool receivedFragment =
false;
92 static bool initialized =
false;
93 static size_t fragments_received = 0;
95 while (gSignalStatus == 0)
97 receivedFragment =
false;
98 auto fragmentPtr = std::make_unique<artdaq::Fragment>();
100 while (!receivedFragment)
102 if (gSignalStatus != 0)
104 TLOG(TLVL_INFO) <<
"Ctrl-C appears to have been hit";
108 if (!monitorRegistered_)
111 if (!monitorRegistered_)
119 auto result = transfer_->receiveFragment(*fragmentPtr, timeoutInUsecs_);
123 receivedFragment =
true;
124 fragments_received++;
126 static size_t cntr = 0;
127 auto mod = ++cntr % 10;
141 TLOG(TLVL_INFO) <<
"Received " << cntr << suffix <<
" event, "
142 <<
"seqID == " << fragmentPtr->sequenceID()
143 <<
", type == " << fragmentPtr->typeString();
148 TLOG(TLVL_ERROR) <<
"Transfer Plugin disconnected or other unrecoverable error. Shutting down.";
157 TLOG(TLVL_WARNING) <<
"Timeout occurred in call to transfer_->receiveFragmentFrom; will try again"
158 <<
", status = " << result;
163 ExceptionHandler(ExceptionHandlerRethrow::yes,
164 "Problem receiving data in TransferWrapper::receiveMessage");
168 if (fragmentPtr->type() == artdaq::Fragment::EndOfDataFragmentType)
184 checkIntegrity(*fragmentPtr);
186 if (initialized || fragmentPtr->type() == artdaq::Fragment::InitFragmentType)
189 fragmentPtrs.push_back(std::move(fragmentPtr));
193 if (fragments_received > maxEventsBeforeInit_)
195 throw cet::exception(
"TransferWrapper") <<
"First " << maxEventsBeforeInit_ <<
" events received did not include the \"Init\" event containing necessary info for art; exiting...";
204 std::unordered_map<artdaq::Fragment::type_t, std::unique_ptr<artdaq::Fragments>> output;
206 auto ptrs = receiveMessage();
207 for (
auto& ptr : ptrs)
209 auto fragType = ptr->type();
210 auto fragPtr = ptr.release();
213 if (output.count(fragType) == 0u)
215 output[fragType] = std::make_unique<artdaq::Fragments>();
218 output[fragType]->emplace_back(std::move(*fragPtr));
224 void artdaq::TransferWrapper::checkIntegrity(
const artdaq::Fragment& fragment)
const
226 const size_t artdaqheader = artdaq::detail::RawFragmentHeader::num_words() *
227 sizeof(artdaq::detail::RawFragmentHeader::RawDataType);
228 const auto payload =
static_cast<size_t>(fragment.dataEndBytes() - fragment.dataBeginBytes());
230 const size_t totalsize = fragment.sizeBytes();
232 const auto type =
static_cast<size_t>(fragment.type());
234 if (totalsize != artdaqheader + metadata + payload)
236 std::stringstream errmsg;
237 errmsg <<
"Error: artdaq fragment of type " << fragment.typeString() <<
", sequence ID " << fragment.sequenceID() <<
" has internally inconsistent measures of its size, signalling data corruption: in bytes,"
238 <<
" total size = " << totalsize <<
", artdaq fragment header = " << artdaqheader <<
", metadata = " << metadata <<
", payload = " << payload;
240 TLOG(TLVL_ERROR) << errmsg.str();
242 if (quitOnFragmentIntegrityProblem_)
244 throw cet::exception(
"TransferWrapper") << errmsg.str();
250 auto findloc = std::find(allowedFragmentTypes_.begin(), allowedFragmentTypes_.end(),
static_cast<int>(type));
252 if (findloc == allowedFragmentTypes_.end())
254 std::stringstream errmsg;
255 errmsg <<
"Error: artdaq fragment appears to have type "
256 << type <<
", not found in the allowed fragment types list";
258 TLOG(TLVL_ERROR) << errmsg.str();
259 if (quitOnFragmentIntegrityProblem_)
261 throw cet::exception(
"TransferWrapper") << errmsg.str();
268 void artdaq::TransferWrapper::registerMonitor()
272 transfer_.reset(
nullptr);
277 ExceptionHandler(ExceptionHandlerRethrow::yes,
278 "TransferWrapper: failure in call to MakeTransferPlugin");
281 auto start = std::chrono::steady_clock::now();
282 auto sts = getDispatcherStatus();
283 while (sts !=
"Running" && (runningStateTimeout_ == 0 || TimeUtils::GetElapsedTime(start) < runningStateTimeout_))
285 TLOG(TLVL_DEBUG) <<
"Dispatcher state: " << sts;
286 if (gSignalStatus != 0)
288 TLOG(TLVL_INFO) <<
"Ctrl-C appears to have been hit";
291 TLOG(TLVL_INFO) <<
"Waited " << std::fixed << std::setprecision(2) << TimeUtils::GetElapsedTime(start) <<
" s / " << runningStateTimeout_ <<
" s for Dispatcher to enter the Running state (state=" << sts <<
")";
292 usleep(runningStateInterval_us_);
293 sts = getDispatcherStatus();
295 if (sts !=
"Running")
300 auto dispatcherConfig = pset_.get<fhicl::ParameterSet>(
"dispatcher_config");
306 TLOG(TLVL_INFO) <<
"Attempting to register this monitor (\"" << transfer_->uniqueLabel()
307 <<
"\") with the dispatcher aggregator";
309 auto status = commander_->send_register_monitor(dispatcherConfig.to_string());
311 TLOG(TLVL_INFO) <<
"Response from dispatcher is \"" << status <<
"\"";
313 if (status ==
"Success")
315 monitorRegistered_ =
true;
319 TLOG(TLVL_WARNING) <<
"Error in TransferWrapper: attempt to register with dispatcher did not result in the \"Success\" response";
326 void artdaq::TransferWrapper::unregisterMonitor()
328 if (!monitorRegistered_)
330 TLOG(TLVL_WARNING) <<
"The function to unregister the monitor was called, but the monitor doesn't appear to be registered";
334 std::string sts = getDispatcherStatus();
341 if (sts !=
"Running" && sts !=
"Ready")
343 TLOG(TLVL_WARNING) <<
"The Dispatcher is not in the Running or Ready state, will not attempt to unregister";
350 TLOG(TLVL_INFO) <<
"Requesting that this monitor (" << transfer_->uniqueLabel()
351 <<
") be unregistered from the dispatcher aggregator";
353 auto status = commander_->send_unregister_monitor(transfer_->uniqueLabel());
355 TLOG(TLVL_INFO) <<
"Response from dispatcher is \"" << status <<
"\"";
357 if (status ==
"Success")
361 if (status ==
"busy")
365 TLOG(TLVL_WARNING) <<
"The Dispatcher returned status " << status <<
" when attempting to unregister this monitor!";
371 monitorRegistered_ =
false;
374 std::string artdaq::TransferWrapper::getDispatcherStatus()
378 return commander_->send_status();
380 catch (std::exception
const& ex)
382 TLOG(TLVL_WARNING) <<
"An exception was thrown trying to collect the Dispatcher's status. Most likely cause is the application is no longer running.";
389 if (monitorRegistered_)
397 ExceptionHandler(ExceptionHandlerRethrow::no,
398 "An exception occurred when trying to unregister monitor during TransferWrapper's destruction");
Commandable is the base class for all artdaq components which implement the artdaq state machine...
TransferWrapper(const fhicl::ParameterSet &pset)
TransferWrapper Constructor.
static void CleanUpGlobals()
Clean up statically-allocated Manager class instances.
virtual ~TransferWrapper()
TransferWrapper Destructor.
std::unique_ptr< artdaq::TransferInterface > MakeTransferPlugin(const fhicl::ParameterSet &pset, const std::string &plugin_label, TransferInterface::Role role)
Load a TransferInterface plugin.
artdaq::FragmentPtrs receiveMessage()
Receive a Fragment from the TransferInterface, and send it to art.
This TransferInterface is a Receiver.
volatile std::sig_atomic_t gSignalStatus
Stores singal from signal handler.
std::unique_ptr< artdaq::CommanderInterface > MakeCommanderPlugin(const fhicl::ParameterSet &commander_pset, artdaq::Commandable &commandable)
Load a CommanderInterface plugin.
Value that is to be returned when a Transfer plugin determines that no more data will be arriving...
For code clarity, things checking for successful receive should check retval >= NO_RANK_INFO.
std::unordered_map< artdaq::Fragment::type_t, std::unique_ptr< artdaq::Fragments > > receiveMessages()
Receive all messsages for an event from ArtdaqSharedMemoryService.