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 <opmonlib/OpMonManager.hpp>
      12              : #include "logging/Logging.hpp"
      13              : 
      14              : using namespace dunedaq::opmonlib;
      15              : 
      16           69 : OpMonManager::OpMonManager( std::string session,
      17              :                             std::string name,
      18           69 :                             facility_ptr_t f_ptr)
      19           69 :   : MonitorableObject( name, session) {
      20              : 
      21           69 :   m_facility.store(f_ptr);
      22              :   
      23           69 : }
      24              : 
      25           14 : void OpMonManager::start_monitoring() {
      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() << " seconds, at level " << get_opmon_level();
      31              : 
      32           13 :   auto running_function = std::bind( & OpMonManager::run, this, std::placeholders::_1);
      33           13 :   m_thread = std::jthread( running_function );
      34           13 :   auto handle = m_thread.native_handle();
      35           13 :   auto thread_name = "opmon";
      36           13 :   auto rc =  pthread_setname_np(handle, thread_name);    
      37           13 :    if (rc != 0) {
      38            0 :     ers::warning(ThreadNameTooLong(ERS_HERE, thread_name));
      39              :   }
      40           13 : }
      41              : 
      42              : 
      43            7 : void OpMonManager::stop_monitoring() {
      44              : 
      45           14 :   TLOG() << "Gracefully requesting the monitoring thread to stop";
      46              :   
      47            7 :   m_thread.request_stop();
      48            7 :   m_thread.join();
      49            7 : }
      50              : 
      51           13 : void OpMonManager::run(std::stop_token stoken ) {
      52              : 
      53           26 :   TLOG() << "Monitoring thread started";
      54              :   
      55           13 :   auto sleep_interval = std::chrono::milliseconds(100);
      56           13 :   auto reporting_interval = m_cfg.load()->get_interval();  
      57           13 :   auto last_collection_time = std::chrono::steady_clock::now();
      58              :   
      59          229 :   while( ! stoken.stop_requested() ) {
      60              :     
      61          203 :     std::this_thread::sleep_for(sleep_interval);
      62          203 :     auto time_span = std::chrono::duration_cast<std::chrono::milliseconds>( std::chrono::steady_clock::now() - last_collection_time);
      63              : 
      64          203 :     if ( time_span < reporting_interval ) continue;
      65              : 
      66            0 :     if ( stoken.stop_requested() ) break;
      67              :       
      68            0 :     last_collection_time = std::chrono::steady_clock::now();
      69            0 :     publish( collect() );
      70              :     // there is no catch here because collect is supposed to catch all possible exceptions
      71              :     // In this way we should garantee the collection of metrics on the system
      72              :     
      73              :   }
      74              : 
      75           26 :   TLOG() << "Exiting the monitoring thread";
      76           13 : }
      77              : 
      78              :