src/kernel/activity/ExecImpl.cpp

   1 /* Copyright (c) 2007-2019. The SimGrid Team. All rights reserved.          */
   2
   3 /* This program is free software; you can redistribute it and/or modify it
   4  * under the terms of the license (GNU LGPL) which comes with this package. */
   5
   6 #include "src/kernel/activity/ExecImpl.hpp"
   7 #include "simgrid/Exception.hpp"
   8 #include "simgrid/modelchecker.h"
   9 #include "src/mc/mc_replay.hpp"
  10 #include "src/simix/smx_host_private.hpp"
  11 #include "src/surf/cpu_interface.hpp"
  12 #include "src/surf/surf_interface.hpp"
  13
  14 #include "simgrid/s4u/Host.hpp"
  15
  16 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(simix_process);
  17
  18 void simcall_HANDLER_execution_wait(smx_simcall_t simcall, smx_activity_t synchro)
  19 {
  20   XBT_DEBUG("Wait for execution of synchro %p, state %d", synchro.get(), (int)synchro->state_);
  21
  22   /* Associate this simcall to the synchro */
  23   synchro->simcalls_.push_back(simcall);
  24   simcall->issuer->waiting_synchro = synchro;
  25
  26   /* set surf's synchro */
  27   if (MC_is_active() || MC_record_replay_is_active()) {
  28     synchro->state_ = SIMIX_DONE;
  29     boost::static_pointer_cast<simgrid::kernel::activity::ExecImpl>(synchro)->finish();
  30     return;
  31   }
  32
  33   /* If the synchro is already finished then perform the error handling */
  34   if (synchro->state_ != SIMIX_RUNNING)
  35     boost::static_pointer_cast<simgrid::kernel::activity::ExecImpl>(synchro)->finish();
  36 }
  37
  38 void simcall_HANDLER_execution_test(smx_simcall_t simcall, smx_activity_t synchro)
  39 {
  40   int res = (synchro->state_ != SIMIX_WAITING && synchro->state_ != SIMIX_RUNNING);
  41   if (res) {
  42     synchro->simcalls_.push_back(simcall);
  43     boost::static_pointer_cast<simgrid::kernel::activity::ExecImpl>(synchro)->finish();
  44   } else {
  45     SIMIX_simcall_answer(simcall);
  46   }
  47   simcall_execution_test__set__result(simcall, res);
  48 }
  49
  50 namespace simgrid {
  51 namespace kernel {
  52 namespace activity {
  53
  54 ExecImpl::ExecImpl(std::string name, std::string tracing_category, resource::Action* timeout_detector, s4u::Host* host)
  55     : ActivityImpl(std::move(name)), host_(host), timeout_detector_(timeout_detector)
  56 {
  57   this->state_ = SIMIX_RUNNING;
  58   this->set_category(std::move(tracing_category));
  59
  60   if (timeout_detector != nullptr)
  61     timeout_detector_->set_data(this);
  62
  63   XBT_DEBUG("Create exec %p", this);
  64 }
  65
  66 ExecImpl::~ExecImpl()
  67 {
  68   if (surf_action_)
  69     surf_action_->unref();
  70   if (timeout_detector_)
  71     timeout_detector_->unref();
  72   XBT_DEBUG("Destroy exec %p", this);
  73 }
  74
  75 ExecImpl* ExecImpl::start(double flops_amount, double priority, double bound)
  76 {
  77   if (not MC_is_active() && not MC_record_replay_is_active()) {
  78     surf_action_ = host_->pimpl_cpu->execution_start(flops_amount);
  79     surf_action_->set_data(this);
  80     surf_action_->set_priority(priority);
  81     if (bound > 0)
  82       surf_action_->set_bound(bound);
  83   }
  84
  85   XBT_DEBUG("Create execute synchro %p: %s", this, get_cname());
  86   ExecImpl::on_creation(this);
  87   return this;
  88 }
  89
  90 void ExecImpl::cancel()
  91 {
  92   XBT_VERB("This exec %p is canceled", this);
  93   if (surf_action_ != nullptr)
  94     surf_action_->cancel();
  95 }
  96
  97 double ExecImpl::get_remaining()
  98 {
  99   xbt_assert(host_ != nullptr, "Calling remains() on a parallel execution is not allowed. "
 100                                "We would need to return a vector instead of a scalar. "
 101                                "Did you mean remainingRatio() instead?");
 102   return surf_action_ ? surf_action_->get_remains() : 0;
 103 }
 104
 105 double ExecImpl::get_remaining_ratio()
 106 {
 107   if (host_ ==
 108       nullptr) // parallel task: their remain is already between 0 and 1 (see comment in ExecImpl::get_remaining())
 109     return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains();
 110   else // Actually compute the ratio for sequential tasks
 111     return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains() / surf_action_->get_cost();
 112 }
 113
 114 void ExecImpl::set_bound(double bound)
 115 {
 116   if (surf_action_)
 117     surf_action_->set_bound(bound);
 118 }
 119 void ExecImpl::set_priority(double priority)
 120 {
 121   if (surf_action_)
 122     surf_action_->set_priority(priority);
 123 }
 124
 125 void ExecImpl::post()
 126 {
 127   if (host_ && not host_->is_on()) { /* FIXME: handle resource failure for parallel tasks too */
 128     /* If the host running the synchro failed, notice it. This way, the asking
 129      * process can be killed if it runs on that host itself */
 130     state_ = SIMIX_FAILED;
 131   } else if (surf_action_ && surf_action_->get_state() == resource::Action::State::FAILED) {
 132     /* If the host running the synchro didn't fail, then the synchro was canceled */
 133     state_ = SIMIX_CANCELED;
 134   } else if (timeout_detector_ && timeout_detector_->get_state() == resource::Action::State::FINISHED) {
 135     state_ = SIMIX_TIMEOUT;
 136   } else {
 137     state_ = SIMIX_DONE;
 138   }
 139
 140   on_completion(this);
 141
 142   if (surf_action_) {
 143     surf_action_->unref();
 144     surf_action_ = nullptr;
 145   }
 146   if (timeout_detector_) {
 147     timeout_detector_->unref();
 148     timeout_detector_ = nullptr;
 149   }
 150
 151   /* If there are simcalls associated with the synchro, then answer them */
 152   if (not simcalls_.empty())
 153     finish();
 154 }
 155
 156 void ExecImpl::finish()
 157 {
 158   while (not simcalls_.empty()) {
 159     smx_simcall_t simcall = simcalls_.front();
 160     simcalls_.pop_front();
 161     switch (state_) {
 162
 163       case SIMIX_DONE:
 164         /* do nothing, synchro done */
 165         XBT_DEBUG("SIMIX_execution_finished: execution successful");
 166         break;
 167
 168       case SIMIX_FAILED:
 169         XBT_DEBUG("SIMIX_execution_finished: host '%s' failed", simcall->issuer->get_host()->get_cname());
 170         simcall->issuer->context_->iwannadie = true;
 171         simcall->issuer->exception_ =
 172             std::make_exception_ptr(simgrid::HostFailureException(XBT_THROW_POINT, "Host failed"));
 173         break;
 174
 175       case SIMIX_CANCELED:
 176         XBT_DEBUG("SIMIX_execution_finished: execution canceled");
 177         simcall->issuer->exception_ =
 178             std::make_exception_ptr(simgrid::CancelException(XBT_THROW_POINT, "Execution Canceled"));
 179         break;
 180
 181       case SIMIX_TIMEOUT:
 182         XBT_DEBUG("SIMIX_execution_finished: execution timeouted");
 183         simcall->issuer->exception_ = std::make_exception_ptr(simgrid::TimeoutError(XBT_THROW_POINT, "Timeouted"));
 184         break;
 185
 186       default:
 187         xbt_die("Internal error in SIMIX_execution_finish: unexpected synchro state %d", static_cast<int>(state_));
 188     }
 189
 190     simcall->issuer->waiting_synchro = nullptr;
 191     simcall_execution_wait__set__result(simcall, state_);
 192
 193     /* Fail the process if the host is down */
 194     if (simcall->issuer->get_host()->is_on())
 195       SIMIX_simcall_answer(simcall);
 196     else
 197       simcall->issuer->context_->iwannadie = true;
 198   }
 199 }
 200
 201 ActivityImpl* ExecImpl::migrate(simgrid::s4u::Host* to)
 202 {
 203   if (not MC_is_active() && not MC_record_replay_is_active()) {
 204     resource::Action* old_action = this->surf_action_;
 205     resource::Action* new_action = to->pimpl_cpu->execution_start(old_action->get_cost());
 206     new_action->set_remains(old_action->get_remains());
 207     new_action->set_data(this);
 208     new_action->set_priority(old_action->get_priority());
 209
 210     // FIXME: the user-defined bound seem to not be kept by LMM, that seem to overwrite it for the multi-core modeling.
 211     // I hope that the user did not provide any.
 212
 213     old_action->set_data(nullptr);
 214     old_action->cancel();
 215     old_action->unref();
 216     this->surf_action_ = new_action;
 217   }
 218
 219   on_migration(this, to);
 220   return this;
 221 }
 222
 223 /*************
 224  * Callbacks *
 225  *************/
 226 xbt::signal<void(ExecImplPtr)> ExecImpl::on_creation;
 227 xbt::signal<void(ExecImplPtr)> ExecImpl::on_completion;
 228 xbt::signal<void(ExecImplPtr, s4u::Host*)> ExecImpl::on_migration;
 229
 230 } // namespace activity
 231 } // namespace kernel
 232 } // namespace simgrid