src/kernel/activity/ExecImpl.cpp

   1 /* Copyright (c) 2007-2020. 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/surf/HostImpl.hpp"
  11 #include "src/surf/cpu_interface.hpp"
  12 #include "src/surf/surf_interface.hpp"
  13
  14 #include "simgrid/s4u/Host.hpp"
  15
  16 #include <boost/range/algorithm.hpp>
  17
  18 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(simix_process);
  19
  20 void simcall_HANDLER_execution_wait(smx_simcall_t simcall, simgrid::kernel::activity::ExecImpl* synchro, double timeout)
  21 {
  22   synchro->wait_for(simcall->issuer_, timeout);
  23 }
  24
  25 void simcall_HANDLER_execution_waitany_for(smx_simcall_t simcall, simgrid::kernel::activity::ExecImpl* execs[],
  26                                            size_t count, double timeout)
  27 {
  28   if (timeout < 0.0) {
  29     simcall->timeout_cb_ = nullptr;
  30   } else {
  31     simcall->timeout_cb_ = simgrid::simix::Timer::set(SIMIX_get_clock() + timeout, [simcall, execs, count]() {
  32       for (size_t i = 0; i < count; i++) {
  33         // Remove the first occurrence of simcall:
  34         auto* exec = execs[i];
  35         auto j     = boost::range::find(exec->simcalls_, simcall);
  36         if (j != exec->simcalls_.end())
  37           exec->simcalls_.erase(j);
  38       }
  39       simcall_execution_waitany_for__set__result(simcall, -1);
  40       simcall->issuer_->simcall_answer();
  41     });
  42   }
  43
  44   for (size_t i = 0; i < count; i++) {
  45     /* associate this simcall to the the synchro */
  46     auto* exec = execs[i];
  47     exec->simcalls_.push_back(simcall);
  48
  49     /* see if the synchro is already finished */
  50     if (exec->state_ != simgrid::kernel::activity::State::WAITING &&
  51         exec->state_ != simgrid::kernel::activity::State::RUNNING) {
  52       exec->finish();
  53       break;
  54     }
  55   }
  56 }
  57
  58 namespace simgrid {
  59 namespace kernel {
  60 namespace activity {
  61
  62 ExecImpl& ExecImpl::set_host(s4u::Host* host)
  63 {
  64   hosts_.assign(1, host);
  65   return *this;
  66 }
  67
  68 ExecImpl& ExecImpl::set_hosts(const std::vector<s4u::Host*>& hosts)
  69 {
  70   hosts_ = hosts;
  71   return *this;
  72 }
  73
  74 ExecImpl& ExecImpl::set_timeout(double timeout)
  75 {
  76   if (timeout > 0 && not MC_is_active() && not MC_record_replay_is_active()) {
  77     timeout_detector_.reset(hosts_.front()->pimpl_cpu->sleep(timeout));
  78     timeout_detector_->set_activity(this);
  79   }
  80   return *this;
  81 }
  82
  83 ExecImpl& ExecImpl::set_flops_amount(double flops_amount)
  84 {
  85   flops_amounts_.assign(1, flops_amount);
  86   return *this;
  87 }
  88
  89 ExecImpl& ExecImpl::set_flops_amounts(const std::vector<double>& flops_amounts)
  90 {
  91   flops_amounts_ = flops_amounts;
  92   return *this;
  93 }
  94
  95 ExecImpl& ExecImpl::set_bytes_amounts(const std::vector<double>& bytes_amounts)
  96 {
  97   bytes_amounts_ = bytes_amounts;
  98
  99   return *this;
 100 }
 101
 102 ExecImpl* ExecImpl::start()
 103 {
 104   state_ = State::RUNNING;
 105   if (not MC_is_active() && not MC_record_replay_is_active()) {
 106     if (hosts_.size() == 1) {
 107       surf_action_ = hosts_.front()->pimpl_cpu->execution_start(flops_amounts_.front());
 108       surf_action_->set_sharing_penalty(sharing_penalty_);
 109       surf_action_->set_category(get_tracing_category());
 110
 111       if (bound_ > 0)
 112         surf_action_->set_bound(bound_);
 113     } else {
 114       surf_action_ = surf_host_model->execute_parallel(hosts_, flops_amounts_.data(), bytes_amounts_.data(), -1);
 115     }
 116     surf_action_->set_activity(this);
 117   }
 118
 119   XBT_DEBUG("Create execute synchro %p: %s", this, get_cname());
 120   return this;
 121 }
 122
 123 double ExecImpl::get_seq_remaining_ratio()
 124 {
 125   return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains() / surf_action_->get_cost();
 126 }
 127
 128 double ExecImpl::get_par_remaining_ratio()
 129 {
 130   // parallel task: their remain is already between 0 and 1
 131   return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains();
 132 }
 133
 134 ExecImpl& ExecImpl::set_bound(double bound)
 135 {
 136   bound_ = bound;
 137   return *this;
 138 }
 139
 140 ExecImpl& ExecImpl::set_sharing_penalty(double sharing_penalty)
 141 {
 142   sharing_penalty_ = sharing_penalty;
 143   return *this;
 144 }
 145
 146 void ExecImpl::post()
 147 {
 148   if (hosts_.size() == 1 && not hosts_.front()->is_on()) { /* FIXME: handle resource failure for parallel tasks too */
 149     /* If the host running the synchro failed, notice it. This way, the asking
 150      * process can be killed if it runs on that host itself */
 151     state_ = State::FAILED;
 152   } else if (surf_action_ && surf_action_->get_state() == resource::Action::State::FAILED) {
 153     /* If the host running the synchro didn't fail, then the synchro was canceled */
 154     state_ = State::CANCELED;
 155   } else if (timeout_detector_ && timeout_detector_->get_state() == resource::Action::State::FINISHED) {
 156     state_ = State::TIMEOUT;
 157   } else {
 158     state_ = State::DONE;
 159   }
 160
 161   clean_action();
 162   timeout_detector_.reset();
 163   /* Answer all simcalls associated with the synchro */
 164   finish();
 165 }
 166
 167 void ExecImpl::finish()
 168 {
 169   while (not simcalls_.empty()) {
 170     smx_simcall_t simcall = simcalls_.front();
 171     simcalls_.pop_front();
 172
 173     /* If a waitany simcall is waiting for this synchro to finish, then remove it from the other synchros in the waitany
 174      * list. Afterwards, get the position of the actual synchro in the waitany list and return it as the result of the
 175      * simcall */
 176
 177     if (simcall->call_ == SIMCALL_NONE) // FIXME: maybe a better way to handle this case
 178       continue;                        // if process handling comm is killed
 179     if (simcall->call_ == SIMCALL_EXECUTION_WAITANY_FOR) {
 180       simgrid::kernel::activity::ExecImpl** execs = simcall_execution_waitany_for__get__execs(simcall);
 181       size_t count                                = simcall_execution_waitany_for__get__count(simcall);
 182
 183       for (size_t i = 0; i < count; i++) {
 184         // Remove the first occurrence of simcall:
 185         auto* exec = execs[i];
 186         auto j     = boost::range::find(exec->simcalls_, simcall);
 187         if (j != exec->simcalls_.end())
 188           exec->simcalls_.erase(j);
 189
 190         if (simcall->timeout_cb_) {
 191           simcall->timeout_cb_->remove();
 192           simcall->timeout_cb_ = nullptr;
 193         }
 194       }
 195
 196       if (not MC_is_active() && not MC_record_replay_is_active()) {
 197         ExecImpl** element = std::find(execs, execs + count, this);
 198         int rank           = (element != execs + count) ? element - execs : -1;
 199         simcall_execution_waitany_for__set__result(simcall, rank);
 200       }
 201     }
 202
 203     switch (state_) {
 204       case State::DONE:
 205         /* do nothing, synchro done */
 206         XBT_DEBUG("ExecImpl::finish(): execution successful");
 207         break;
 208
 209       case State::FAILED:
 210         XBT_DEBUG("ExecImpl::finish(): host '%s' failed", simcall->issuer_->get_host()->get_cname());
 211         simcall->issuer_->context_->set_wannadie();
 212         if (simcall->issuer_->get_host()->is_on())
 213           simcall->issuer_->exception_ =
 214               std::make_exception_ptr(simgrid::HostFailureException(XBT_THROW_POINT, "Host failed"));
 215         /* else, the actor will be killed with no possibility to survive */
 216         break;
 217
 218       case State::CANCELED:
 219         XBT_DEBUG("ExecImpl::finish(): execution canceled");
 220         simcall->issuer_->exception_ =
 221             std::make_exception_ptr(simgrid::CancelException(XBT_THROW_POINT, "Execution Canceled"));
 222         break;
 223
 224       case State::TIMEOUT:
 225         XBT_DEBUG("ExecImpl::finish(): execution timeouted");
 226         simcall->issuer_->exception_ = std::make_exception_ptr(simgrid::TimeoutException(XBT_THROW_POINT, "Timeouted"));
 227         break;
 228
 229       default:
 230         xbt_die("Internal error in ExecImpl::finish(): unexpected synchro state %d", static_cast<int>(state_));
 231     }
 232
 233     simcall->issuer_->waiting_synchro = nullptr;
 234     /* Fail the process if the host is down */
 235     if (simcall->issuer_->get_host()->is_on())
 236       simcall->issuer_->simcall_answer();
 237     else
 238       simcall->issuer_->context_->set_wannadie();
 239   }
 240 }
 241
 242 ActivityImpl* ExecImpl::migrate(s4u::Host* to)
 243 {
 244   if (not MC_is_active() && not MC_record_replay_is_active()) {
 245     resource::Action* old_action = this->surf_action_;
 246     resource::Action* new_action = to->pimpl_cpu->execution_start(old_action->get_cost());
 247     new_action->set_remains(old_action->get_remains());
 248     new_action->set_activity(this);
 249     new_action->set_sharing_penalty(old_action->get_sharing_penalty());
 250
 251     // FIXME: the user-defined bound seem to not be kept by LMM, that seem to overwrite it for the multi-core modeling.
 252     // I hope that the user did not provide any.
 253
 254     old_action->set_activity(nullptr);
 255     old_action->cancel();
 256     old_action->unref();
 257     this->surf_action_ = new_action;
 258   }
 259
 260   on_migration(*this, to);
 261   return this;
 262 }
 263
 264 /*************
 265  * Callbacks *
 266  *************/
 267 xbt::signal<void(ExecImpl const&, s4u::Host*)> ExecImpl::on_migration;
 268
 269 } // namespace activity
 270 } // namespace kernel
 271 } // namespace simgrid