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