Line data    Source code

       1              : /**
       2              :  * @file OpMonManager.cpp
       3              :  *
       4              :  * This is part of the DUNE DAQ Application Framework, copyright 2020.
       5              :  * Licensing/copyright details are in the COPYING file that you should have
       6              :  * received with this code.
       7              :  */
       8              : 
       9              : #include <chrono>
      10              : 
      11              : #include "logging/Logging.hpp"
      12              : #include <opmonlib/OpMonManager.hpp>
      13              : 
      14              : using namespace dunedaq::opmonlib;
      15              : 
      16           69 : OpMonManager::OpMonManager(std::string session, std::string name, facility_ptr_t f_ptr)
      17           69 :   : MonitorableObject(name, session)
      18              : {
      19              : 
      20           69 :   m_facility.store(f_ptr);
      21           69 : }
      22              : 
      23              : void
      24           14 : OpMonManager::start_monitoring()
      25              : {
      26              : 
      27           14 :   if (!m_cfg)
      28            1 :     throw MissingConfiguration(ERS_HERE);
      29              : 
      30           26 :   TLOG() << "Starting a new monitoring thread with interval " << m_cfg.load()->get_interval().count()
      31           26 :          << " seconds, at level " << get_opmon_level();
      32              : 
      33           13 :   auto running_function = std::bind(&OpMonManager::run, this, std::placeholders::_1);
      34           13 :   m_thread = std::jthread(running_function);
      35           13 :   auto handle = m_thread.native_handle();
      36           13 :   auto thread_name = "opmon";
      37           13 :   auto rc = pthread_setname_np(handle, thread_name);
      38           13 :   if (rc != 0) {
      39            0 :     ers::warning(ThreadNameTooLong(ERS_HERE, thread_name));
      40              :   }
      41           13 : }
      42              : 
      43              : void
      44            7 : OpMonManager::stop_monitoring()
      45              : {
      46              : 
      47           14 :   TLOG() << "Gracefully requesting the monitoring thread to stop";
      48              : 
      49            7 :   m_thread.request_stop();
      50            7 :   m_thread.join();
      51            7 : }
      52              : 
      53              : void
      54           13 : OpMonManager::run(std::stop_token stoken)
      55              : {
      56              : 
      57           26 :   TLOG() << "Monitoring thread started";
      58              : 
      59           13 :   auto sleep_interval = std::chrono::milliseconds(100);
      60           13 :   auto reporting_interval = m_cfg.load()->get_interval();
      61           13 :   auto last_collection_time = std::chrono::steady_clock::now();
      62              : 
      63          231 :   while (!stoken.stop_requested()) {
      64              : 
      65          205 :     std::this_thread::sleep_for(sleep_interval);
      66          205 :     auto time_span =
      67          205 :       std::chrono::duration_cast<std::chrono::milliseconds>(std::chrono::steady_clock::now() - last_collection_time);
      68              : 
      69          205 :     if (time_span < reporting_interval)
      70          205 :       continue;
      71              : 
      72            0 :     if (stoken.stop_requested())
      73              :       break;
      74              : 
      75            0 :     last_collection_time = std::chrono::steady_clock::now();
      76            0 :     publish(collect());
      77              :     // there is no catch here because collect is supposed to catch all possible exceptions
      78              :     // In this way we should garantee the collection of metrics on the system
      79              :   }
      80              : 
      81           26 :   TLOG() << "Exiting the monitoring thread";
      82           13 : }