DataHandlingModel.hxx

Go to the documentation of this file.

// Declarations for DataHandlingModel
 
#include <folly/CancellationToken.h>
#include <folly/coro/BlockingWait.h>
#include <folly/coro/CurrentExecutor.h>
#include <folly/coro/Timeout.h>
#include <folly/futures/ThreadWheelTimekeeper.h>
 
#include <typeinfo>
 
namespace dunedaq {
namespace datahandlinglibs {
 
template<class RDT, class RHT, class LBT, class RPT, class IDT>
void
DataHandlingModel<RDT, RHT, LBT, RPT, IDT>::init(const appmodel::DataHandlerModule* mcfg)
{
  // Check if a callback is defined (TPs use IOManager Queues instead)
  m_raw_data_callback_conf = mcfg->get_raw_data_callback();
  if (m_raw_data_callback_conf != nullptr) {
    m_raw_data_callback_conf = mcfg->get_raw_data_callback();
    TLOG_DEBUG(TLVL_WORK_STEPS) << "DataHandlingModel operating in callback mode.";
  } else {
    TLOG_DEBUG(TLVL_WORK_STEPS) << "DataHandlingModel operating in message polling mode.";
  }
 
  try {
    for (auto input : mcfg->get_inputs()) {
      if (input->get_data_type() == "DataRequest") {
        m_data_request_receiver = get_iom_receiver<dfmessages::DataRequest>(input->UID());
      } else {
        m_raw_data_receiver_connection_name = input->UID();
 
        if (m_raw_data_callback_conf == nullptr) {
          m_raw_data_receiver = get_iom_receiver<IDT>(m_raw_data_receiver_connection_name);
          m_raw_receiver_timeout_ms = std::chrono::milliseconds(input->get_recv_timeout_ms());
        }
      }
    }
    for (auto output : mcfg->get_outputs()) {
      if (output->get_data_type() == "TimeSync") {
        m_generate_timesync = true;
        m_timesync_sender = get_iom_sender<dfmessages::TimeSync>(output->UID());
        m_timesync_connection_name = output->UID();
        break;
      }
    }
  } catch (const ers::Issue& excpt) {
    throw ResourceQueueError(ERS_HERE, "raw_input or frag_output", "DataHandlingModel", excpt);
  }
 
  // Raw input connection sensibility check
  if (m_raw_data_callback_conf == nullptr && m_raw_data_receiver == nullptr) {
    ers::error(ConfigurationError(ERS_HERE, m_sourceid, "No callback configuration, and receiver is unset!"));
  }
 
  // Instantiate functionalities
  m_error_registry.reset(new FrameErrorRegistry());
  m_error_registry->set_ers_metadata("DLH of SourceID[" + std::to_string(mcfg->get_source_id()) + "] ");
  m_latency_buffer_impl.reset(new LBT());
  m_raw_processor_impl.reset(new RPT(m_error_registry, mcfg->get_post_processing_enabled()));
  m_request_handler_impl.reset(new RHT(m_latency_buffer_impl, m_error_registry));
 
  register_node(mcfg->get_module_configuration()->get_latency_buffer()->UID(), m_latency_buffer_impl);
  register_node(mcfg->get_module_configuration()->get_data_processor()->UID(), m_raw_processor_impl);
  register_node(mcfg->get_module_configuration()->get_request_handler()->UID(), m_request_handler_impl);
 
  // m_request_handler_impl->init(args);
  // m_raw_processor_impl->init(args);
  m_request_handler_supports_cutoff_timestamp = m_request_handler_impl->supports_cutoff_timestamp();
  m_fake_trigger = false;
  m_raw_receiver_sleep_us = std::chrono::microseconds::zero();
  m_sourceid.id = mcfg->get_source_id();
  m_sourceid.subsystem = RDT::subsystem;
  m_processing_delay_ticks = mcfg->get_module_configuration()->get_post_processing_delay_ticks();
  m_post_processing_delay_min_wait = mcfg->get_module_configuration()->get_post_processing_delay_min_wait();
  m_post_processing_delay_max_wait = mcfg->get_module_configuration()->get_post_processing_delay_max_wait();
 
  if (m_processing_delay_ticks) {
    if constexpr (ExpectsOrder<LBT>) {
      ers::error(ConfigurationError(
        ERS_HERE,
        m_sourceid,
        "Queue buffers (FixedRateQueue, BinarySearchQueue) expect in-order data and must use "
        "post_processing_delay_ticks = 0."));
    }
  }
 
  // Configure implementations:
  m_raw_processor_impl->conf(mcfg);
  // Configure the latency buffer before the request handler so the request handler can check for alignment
  // restrictions
  try {
    m_latency_buffer_impl->conf(mcfg->get_module_configuration()->get_latency_buffer());
  } catch (const std::bad_alloc& be) {
    ers::error(ConfigurationError(ERS_HERE, m_sourceid, "Latency Buffer can't be allocated with size!"));
  }
  m_request_handler_impl->conf(mcfg);
}
 
template<class RDT, class RHT, class LBT, class RPT, class IDT>
void
DataHandlingModel<RDT, RHT, LBT, RPT, IDT>::conf(const appfwk::DAQModule::CommandData_t& /*args*/)
{
  // Register callbacks if operating in that mode.
  if (m_raw_data_callback_conf != nullptr) {
    // Configure and register consume callback
    m_consume_callback =
      std::bind(&DataHandlingModel<RDT, RHT, LBT, RPT, IDT>::consume_callback, this, std::placeholders::_1);
 
    // Register callback
    auto dmcbr = DataMoveCallbackRegistry::get();
    dmcbr->register_callback<IDT>(m_raw_data_callback_conf, m_consume_callback);
  }
 
  // Configure threads:
  m_consumer_thread.set_name("consumer", m_sourceid.id);
  if (m_generate_timesync) {
    m_timesync_thread.set_name("timesync", m_sourceid.id);
  }
  if (m_processing_delay_ticks) {
    m_postprocess_scheduler_thread.set_name("pprocsched", m_sourceid.id);
    m_timekeeper = std::make_unique<folly::ThreadWheelTimekeeper>();
  }
}
 
template<class RDT, class RHT, class LBT, class RPT, class IDT>
void
DataHandlingModel<RDT, RHT, LBT, RPT, IDT>::start(const appfwk::DAQModule::CommandData_t& args)
{
  // Reset opmon variables
  m_sum_payloads = 0;
  m_num_payloads = 0;
  m_sum_requests = 0;
  m_num_requests = 0;
  m_num_lb_insert_failures = 0;
  m_stats_packet_count = 0;
  m_rawq_timeout_count = 0;
  m_num_post_processing_delay_max_waits = 0;
 
  m_t0 = std::chrono::high_resolution_clock::now();
 
  m_run_number = args.value<dunedaq::daqdataformats::run_number_t>("run", 1);
 
  TLOG_DEBUG(TLVL_WORK_STEPS) << "Starting threads...";
  m_raw_processor_impl->start(args);
  m_request_handler_impl->start(args);
  if (m_raw_data_callback_conf == nullptr) {
    m_consumer_thread.set_work(&DataHandlingModel<RDT, RHT, LBT, RPT, IDT>::run_consume, this);
  }
  if (m_generate_timesync) {
    m_timesync_thread.set_work(&DataHandlingModel<RDT, RHT, LBT, RPT, IDT>::run_timesync, this);
  }
  if (m_processing_delay_ticks) {
    m_postprocess_scheduler_thread.set_work(&DataHandlingModel<RDT, RHT, LBT, RPT, IDT>::run_postprocess_scheduler,
                                            this);
  }
  // Register callback to receive and dispatch data requests
  m_data_request_receiver->add_callback(
    std::bind(&DataHandlingModel<RDT, RHT, LBT, RPT, IDT>::dispatch_requests, this, std::placeholders::_1));
}
 
template<class RDT, class RHT, class LBT, class RPT, class IDT>
void
DataHandlingModel<RDT, RHT, LBT, RPT, IDT>::stop(const appfwk::DAQModule::CommandData_t& args)
{
  TLOG_DEBUG(TLVL_WORK_STEPS) << "Stoppping threads...";
 
  // Stop receiving data requests as first thing
  m_data_request_receiver->remove_callback();
  // Stop the other threads
  m_request_handler_impl->stop(args);
  if (m_generate_timesync) {
    while (!m_timesync_thread.get_readiness()) {
      std::this_thread::sleep_for(std::chrono::milliseconds(10));
    }
  }
  if (m_raw_data_callback_conf == nullptr) {
    while (!m_consumer_thread.get_readiness()) {
      std::this_thread::sleep_for(std::chrono::milliseconds(10));
    }
  }
  if (m_processing_delay_ticks) {
    m_baton.post(); // In case the coroutine is still waiting when the consumer has stopped
    while (!m_postprocess_scheduler_thread.get_readiness()) {
      std::this_thread::sleep_for(std::chrono::milliseconds(10));
    }
  }
  TLOG_DEBUG(TLVL_WORK_STEPS) << "Flushing latency buffer with occupancy: " << m_latency_buffer_impl->occupancy();
  m_latency_buffer_impl->flush();
  m_raw_processor_impl->stop(args);
  m_raw_processor_impl->reset_last_daq_time();
}
 
template<class RDT, class RHT, class LBT, class RPT, class IDT>
void
DataHandlingModel<RDT, RHT, LBT, RPT, IDT>::generate_opmon_data()
{
  opmon::DataHandlerInfo ri;
  ri.set_sum_payloads(m_sum_payloads.load());
  ri.set_num_payloads(m_num_payloads.exchange(0));
 
  ri.set_num_data_input_timeouts(m_rawq_timeout_count.exchange(0));
 
  auto now = std::chrono::high_resolution_clock::now();
  int new_packets = m_stats_packet_count.exchange(0);
  double seconds = std::chrono::duration_cast<std::chrono::microseconds>(now - m_t0).count() / 1000000.;
  m_t0 = now;
 
  // 08-May-2025, KAB: added a message to warn users when latency buffer inserts are failing.
  int local_num_lb_insert_failures = m_num_lb_insert_failures.exchange(0);
  if (local_num_lb_insert_failures != 0) {
    ers::warning(
      NonZeroLatencyBufferInsertFailures(ERS_HERE, m_sourceid, local_num_lb_insert_failures, ri.num_payloads()));
  }
 
  ri.set_rate_payloads_consumed(new_packets / seconds / 1000.);
  ri.set_num_lb_insert_failures(local_num_lb_insert_failures);
  ri.set_sum_requests(m_sum_requests.load());
  ri.set_num_requests(m_num_requests.exchange(0));
  ri.set_num_post_processing_delay_max_waits(m_num_post_processing_delay_max_waits.exchange(0));
  ri.set_last_daq_timestamp(m_raw_processor_impl->get_last_daq_time());
  ri.set_newest_timestamp(m_raw_processor_impl->get_last_daq_time());
  ri.set_oldest_timestamp(m_request_handler_impl->get_oldest_time());
 
  this->publish(std::move(ri));
}
 
template<class RDT, class RHT, class LBT, class RPT, class IDT>
void
DataHandlingModel<RDT, RHT, LBT, RPT, IDT>::transform_and_process(IDT&& payload)
{
  if constexpr (std::is_same_v<IDT, RDT>) {
    process_item(std::move(payload));
  } else {
    auto transformed = transform_payload(payload);
    for (auto& i : transformed) {
      process_item(std::move(i));
    }
  }
}
 
template<class RDT, class RHT, class LBT, class RPT, class IDT>
void
DataHandlingModel<RDT, RHT, LBT, RPT, IDT>::consume_callback(IDT&& payload)
{
  transform_and_process(std::move(payload));
}
 
template<class RDT, class RHT, class LBT, class RPT, class IDT>
void
DataHandlingModel<RDT, RHT, LBT, RPT, IDT>::process_item(RDT&& payload)
{
  m_raw_processor_impl->preprocess_item(&payload);
  if (m_request_handler_supports_cutoff_timestamp) {
    int64_t diff1 = payload.get_timestamp() - m_request_handler_impl->get_cutoff_timestamp();
    if (diff1 <= 0) {
      // m_request_handler_impl->increment_tardy_tp_count();
      ers::warning(DataPacketArrivedTooLate(ERS_HERE,
                                            m_sourceid,
                                            m_run_number,
                                            payload.get_timestamp(),
                                            m_request_handler_impl->get_cutoff_timestamp(),
                                            diff1,
                                            (static_cast<double>(diff1) / 62500.0)));
    }
  }
 
  const RDT* written = nullptr;
  if constexpr (ExpectsOrder<LBT>) {
    if (!m_latency_buffer_impl->write(std::move(payload))) {
      // TLOG_DEBUG(TLVL_TAKE_NOTE) << "***ERROR: Latency buffer insert failed! (Payload timestamp=" << payload.get_timestamp() << ")";
      m_num_lb_insert_failures++;
      return;
    }
    written = m_latency_buffer_impl->back();
  } else {
    const auto [returned, result] = m_latency_buffer_impl->write_and_return(std::move(payload));
    if (!result) {
      // TLOG_DEBUG(TLVL_TAKE_NOTE) << "***ERROR: Latency buffer insert failed! (Payload timestamp=" << payload.get_timestamp() << ")";
      m_num_lb_insert_failures++;
      return;
    }
    written = returned;
  }
 
  if (m_processing_delay_ticks == 0) {
    m_raw_processor_impl->postprocess_item(written);
    ++m_num_payloads;
    ++m_sum_payloads;
    ++m_stats_packet_count;
  } else {
    m_baton.post();
  }
}
 
template<class RDT, class RHT, class LBT, class RPT, class IDT>
void
DataHandlingModel<RDT, RHT, LBT, RPT, IDT>::run_postprocess_scheduler()
{
  folly::coro::blockingWait(postprocess_schedule());
}
 
template<class RDT, class RHT, class LBT, class RPT, class IDT>
void
DataHandlingModel<RDT, RHT, LBT, RPT, IDT>::run_consume()
{
 
  TLOG_DEBUG(TLVL_WORK_STEPS) << "Consumer thread started...";
  m_rawq_timeout_count = 0;
  m_num_payloads = 0;
  m_sum_payloads = 0;
  m_stats_packet_count = 0;
  m_num_post_processing_delay_max_waits = 0;
 
  while (m_run_marker.load()) {
    // Try to acquire data
 
    auto opt_payload = m_raw_data_receiver->try_receive(m_raw_receiver_timeout_ms);
 
    if (opt_payload) {
      IDT& payload = opt_payload.value();
      transform_and_process(std::move(payload));
    } else {
      ++m_rawq_timeout_count;
      // Protection against a zero sleep becoming a yield
      if (m_raw_receiver_sleep_us != std::chrono::microseconds::zero())
        std::this_thread::sleep_for(m_raw_receiver_sleep_us);
    }
  }
  TLOG_DEBUG(TLVL_WORK_STEPS) << "Consumer thread joins... ";
}
 
template<class RDT, class RHT, class LBT, class RPT, class IDT>
folly::coro::Task<void>
DataHandlingModel<RDT, RHT, LBT, RPT, IDT>::postprocess_schedule()
{
 
  TLOG_DEBUG(TLVL_WORK_STEPS) << "Postprocess schedule coroutine started...";
  TLOG() << "***** Starting post-process coroutine with timout " << m_post_processing_delay_max_wait << " *****";
 
  PostprocessScheduleAlgorithm sched_algo{ *m_latency_buffer_impl,
                                           *m_raw_processor_impl,
                                           m_processing_delay_ticks,
                                           m_post_processing_delay_min_wait,
                                           m_post_processing_delay_max_wait };
 
  const auto wait_data = [this]() -> folly::coro::Task<void> {
    // folly::coro::timeout cancels the task on timeout.
    // Baton is not cancellable, so we attach a callback to resume the coroutine.
    auto token = co_await folly::coro::co_current_cancellation_token;
    folly::CancellationCallback cb(token, [this] { m_baton.post(); });
    co_await m_baton; // Wait data
  };
 
  while (m_run_marker.load()) {
    bool timeout = false;
 
    if (m_post_processing_delay_max_wait > 0) {
      try {
        co_await folly::coro::timeout(
          wait_data(), std::chrono::milliseconds{ m_post_processing_delay_max_wait }, m_timekeeper.get());
 
      } catch (const folly::FutureTimeout&) {
        timeout = true;
        ++m_num_post_processing_delay_max_waits;
      }
    } else {
      co_await m_baton;
    }
 
    m_baton.reset();
 
    if (auto processed = sched_algo.run(timeout); processed > 0) {
      m_num_payloads += processed;
      m_sum_payloads += processed;
      m_stats_packet_count += processed;
    }
  }
}
 
template<class RDT, class RHT, class LBT, class RPT, class IDT>
void
DataHandlingModel<RDT, RHT, LBT, RPT, IDT>::run_timesync()
{
  TLOG_DEBUG(TLVL_WORK_STEPS) << "TimeSync thread started...";
  m_num_requests = 0;
  m_sum_requests = 0;
  uint64_t msg_seqno = 0;
  timestamp_t prev_timestamp = 0;
  auto once_per_run = true;
  size_t zero_timestamp_count = 0;
  size_t duplicate_timestamp_count = 0;
  size_t total_timestamp_count = 0;
  while (m_run_marker.load()) {
    try {
      auto timesyncmsg = dfmessages::TimeSync(m_raw_processor_impl->get_last_daq_time());
      ++total_timestamp_count;
      // daq_time is zero for the first received timesync, and may
      // be the same as the previous daq_time if the data has
      // stopped flowing. In both cases we don't send the TimeSync
      if (timesyncmsg.daq_time != 0 && timesyncmsg.daq_time != prev_timestamp) {
        prev_timestamp = timesyncmsg.daq_time;
        timesyncmsg.run_number = m_run_number;
        timesyncmsg.sequence_number = ++msg_seqno;
        timesyncmsg.source_id = m_sourceid.id;
        TLOG_DEBUG(TLVL_TIME_SYNCS) << "New timesync: daq=" << timesyncmsg.daq_time
                                    << " wall=" << timesyncmsg.system_time << " run=" << timesyncmsg.run_number
                                    << " seqno=" << timesyncmsg.sequence_number
                                    << " source_id=" << timesyncmsg.source_id;
        try {
          dfmessages::TimeSync timesyncmsg_copy(timesyncmsg);
          m_timesync_sender->send(std::move(timesyncmsg_copy), std::chrono::milliseconds(500));
        } catch (ers::Issue& excpt) {
          ers::warning(TimeSyncTransmissionFailed(ERS_HERE, m_sourceid, m_timesync_connection_name, excpt));
        }
 
        if (m_fake_trigger) {
          dfmessages::DataRequest dr;
          ++m_current_fake_trigger_id;
          dr.trigger_number = m_current_fake_trigger_id;
          dr.trigger_timestamp = timesyncmsg.daq_time > 500 * us ? timesyncmsg.daq_time - 500 * us : 0;
          auto width = 300000;
          uint offset = 100;
          dr.request_information.window_begin = dr.trigger_timestamp > offset ? dr.trigger_timestamp - offset : 0;
          dr.request_information.window_end = dr.request_information.window_begin + width;
          dr.request_information.component = m_sourceid;
          dr.data_destination = "data_fragments_q";
          TLOG_DEBUG(TLVL_WORK_STEPS) << "Issuing fake trigger based on timesync. "
                                      << " ts=" << dr.trigger_timestamp
                                      << " window_begin=" << dr.request_information.window_begin
                                      << " window_end=" << dr.request_information.window_end;
          m_request_handler_impl->issue_request(dr);
 
          ++m_num_requests;
          ++m_sum_requests;
        }
      } else {
        if (timesyncmsg.daq_time == 0) {
          ++zero_timestamp_count;
        }
        if (timesyncmsg.daq_time == prev_timestamp) {
          ++duplicate_timestamp_count;
        }
        if (once_per_run) {
          TLOG() << "Timesync with DAQ time 0 won't be sent out as it's an invalid sync.";
          once_per_run = false;
        }
      }
    } catch (const iomanager::TimeoutExpired& excpt) {
      // ++m_timesyncqueue_timeout;
    }
    // Split up the 100ms sleep into 10 sleeps of 10ms, so we respond to "stop" quicker
    for (size_t i = 0; i < 10; ++i) {
      std::this_thread::sleep_for(std::chrono::milliseconds(10));
      if (!m_run_marker.load()) {
        break;
      }
    }
  }
  once_per_run = true;
  TLOG_DEBUG(TLVL_WORK_STEPS) << "TimeSync thread joins... (timestamp count, zero/same/total  = "
                              << zero_timestamp_count << "/" << duplicate_timestamp_count << "/"
                              << total_timestamp_count << ")";
}
 
template<class RDT, class RHT, class LBT, class RPT, class IDT>
void
DataHandlingModel<RDT, RHT, LBT, RPT, IDT>::dispatch_requests(dfmessages::DataRequest& data_request)
{
  if (data_request.request_information.component != m_sourceid) {
    ers::error(RequestSourceIDMismatch(ERS_HERE, m_sourceid, data_request.request_information.component));
    return;
  }
  TLOG_DEBUG(TLVL_QUEUE_POP) << "Received DataRequest"
                             << " for trig/seq_number " << data_request.trigger_number << "."
                             << data_request.sequence_number << ", runno " << data_request.run_number
                             << ", trig timestamp " << data_request.trigger_timestamp
                             << ", SourceID: " << data_request.request_information.component << ", window begin/end "
                             << data_request.request_information.window_begin << "/"
                             << data_request.request_information.window_end
                             << ", dest: " << data_request.data_destination;
  m_request_handler_impl->issue_request(data_request);
  ++m_num_requests;
  ++m_sum_requests;
}
 
} // namespace datahandlinglibs
} // namespace dunedaq