src/kernel/activity/ExecImpl.cpp

   1 /* Copyright (c) 2007-2021. 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()
  59 {
  60   actor::ActorImpl* self = actor::ActorImpl::self();
  61   if (self) {
  62     actor_ = self;
  63     self->activities_.emplace_back(this);
  64   }
  65 }
  66
  67 ExecImpl& ExecImpl::set_host(s4u::Host* host)
  68 {
  69   hosts_.assign(1, host);
  70   return *this;
  71 }
  72
  73 ExecImpl& ExecImpl::set_hosts(const std::vector<s4u::Host*>& hosts)
  74 {
  75   hosts_ = hosts;
  76   return *this;
  77 }
  78
  79 ExecImpl& ExecImpl::set_timeout(double timeout)
  80 {
  81   if (timeout > 0 && not MC_is_active() && not MC_record_replay_is_active()) {
  82     timeout_detector_.reset(hosts_.front()->pimpl_cpu->sleep(timeout));
  83     timeout_detector_->set_activity(this);
  84   }
  85   return *this;
  86 }
  87
  88 ExecImpl& ExecImpl::set_flops_amount(double flops_amount)
  89 {
  90   flops_amounts_.assign(1, flops_amount);
  91   return *this;
  92 }
  93
  94 ExecImpl& ExecImpl::set_flops_amounts(const std::vector<double>& flops_amounts)
  95 {
  96   flops_amounts_ = flops_amounts;
  97   return *this;
  98 }
  99
 100 ExecImpl& ExecImpl::set_bytes_amounts(const std::vector<double>& bytes_amounts)
 101 {
 102   bytes_amounts_ = bytes_amounts;
 103
 104   return *this;
 105 }
 106
 107 ExecImpl* ExecImpl::start()
 108 {
 109   state_ = State::RUNNING;
 110   if (not MC_is_active() && not MC_record_replay_is_active()) {
 111     if (hosts_.size() == 1) {
 112       surf_action_ = hosts_.front()->pimpl_cpu->execution_start(flops_amounts_.front());
 113       surf_action_->set_sharing_penalty(sharing_penalty_);
 114       surf_action_->set_category(get_tracing_category());
 115
 116       if (bound_ > 0)
 117         surf_action_->set_bound(bound_);
 118     } else {
 119       surf_action_ = surf_host_model->execute_parallel(hosts_, flops_amounts_.data(), bytes_amounts_.data(), -1);
 120     }
 121     surf_action_->set_activity(this);
 122   }
 123
 124   XBT_DEBUG("Create execute synchro %p: %s", this, get_cname());
 125   return this;
 126 }
 127
 128 double ExecImpl::get_seq_remaining_ratio()
 129 {
 130   return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains() / surf_action_->get_cost();
 131 }
 132
 133 double ExecImpl::get_par_remaining_ratio()
 134 {
 135   // parallel task: their remain is already between 0 and 1
 136   return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains();
 137 }
 138
 139 ExecImpl& ExecImpl::set_bound(double bound)
 140 {
 141   bound_ = bound;
 142   return *this;
 143 }
 144
 145 ExecImpl& ExecImpl::set_sharing_penalty(double sharing_penalty)
 146 {
 147   sharing_penalty_ = sharing_penalty;
 148   return *this;
 149 }
 150
 151 void ExecImpl::post()
 152 {
 153   if (hosts_.size() == 1 && not hosts_.front()->is_on()) { /* FIXME: handle resource failure for parallel tasks too */
 154     /* If the host running the synchro failed, notice it. This way, the asking
 155      * process can be killed if it runs on that host itself */
 156     state_ = State::FAILED;
 157   } else if (surf_action_ && surf_action_->get_state() == resource::Action::State::FAILED) {
 158     /* If the host running the synchro didn't fail, then the synchro was canceled */
 159     state_ = State::CANCELED;
 160   } else if (timeout_detector_ && timeout_detector_->get_state() == resource::Action::State::FINISHED) {
 161     state_ = State::TIMEOUT;
 162   } else {
 163     state_ = State::DONE;
 164   }
 165
 166   clean_action();
 167   timeout_detector_.reset();
 168   if (actor_) {
 169     actor_->activities_.remove(this);
 170     actor_ = nullptr;
 171   }
 172   /* Answer all simcalls associated with the synchro */
 173   finish();
 174 }
 175
 176 void ExecImpl::finish()
 177 {
 178   while (not simcalls_.empty()) {
 179     smx_simcall_t simcall = simcalls_.front();
 180     simcalls_.pop_front();
 181
 182     /* If a waitany simcall is waiting for this synchro to finish, then remove it from the other synchros in the waitany
 183      * list. Afterwards, get the position of the actual synchro in the waitany list and return it as the result of the
 184      * simcall */
 185
 186     if (simcall->call_ == simix::Simcall::NONE) // FIXME: maybe a better way to handle this case
 187       continue;                        // if process handling comm is killed
 188     if (simcall->call_ == simix::Simcall::EXECUTION_WAITANY_FOR) {
 189       simgrid::kernel::activity::ExecImpl** execs = simcall_execution_waitany_for__get__execs(simcall);
 190       size_t count                                = simcall_execution_waitany_for__get__count(simcall);
 191
 192       for (size_t i = 0; i < count; i++) {
 193         // Remove the first occurrence of simcall:
 194         auto* exec = execs[i];
 195         auto j     = boost::range::find(exec->simcalls_, simcall);
 196         if (j != exec->simcalls_.end())
 197           exec->simcalls_.erase(j);
 198
 199         if (simcall->timeout_cb_) {
 200           simcall->timeout_cb_->remove();
 201           simcall->timeout_cb_ = nullptr;
 202         }
 203       }
 204
 205       if (not MC_is_active() && not MC_record_replay_is_active()) {
 206         ExecImpl** element = std::find(execs, execs + count, this);
 207         int rank           = (element != execs + count) ? element - execs : -1;
 208         simcall_execution_waitany_for__set__result(simcall, rank);
 209       }
 210     }
 211
 212     switch (state_) {
 213       case State::DONE:
 214         /* do nothing, synchro done */
 215         XBT_DEBUG("ExecImpl::finish(): execution successful");
 216         break;
 217
 218       case State::FAILED:
 219         XBT_DEBUG("ExecImpl::finish(): host '%s' failed", simcall->issuer_->get_host()->get_cname());
 220         simcall->issuer_->context_->set_wannadie();
 221         if (simcall->issuer_->get_host()->is_on())
 222           simcall->issuer_->exception_ =
 223               std::make_exception_ptr(simgrid::HostFailureException(XBT_THROW_POINT, "Host failed"));
 224         /* else, the actor will be killed with no possibility to survive */
 225         break;
 226
 227       case State::CANCELED:
 228         XBT_DEBUG("ExecImpl::finish(): execution canceled");
 229         simcall->issuer_->exception_ =
 230             std::make_exception_ptr(simgrid::CancelException(XBT_THROW_POINT, "Execution Canceled"));
 231         break;
 232
 233       case State::TIMEOUT:
 234         XBT_DEBUG("ExecImpl::finish(): execution timeouted");
 235         simcall->issuer_->exception_ = std::make_exception_ptr(simgrid::TimeoutException(XBT_THROW_POINT, "Timeouted"));
 236         break;
 237
 238       default:
 239         xbt_die("Internal error in ExecImpl::finish(): unexpected synchro state %d", static_cast<int>(state_));
 240     }
 241
 242     simcall->issuer_->waiting_synchro_ = nullptr;
 243     /* Fail the process if the host is down */
 244     if (simcall->issuer_->get_host()->is_on())
 245       simcall->issuer_->simcall_answer();
 246     else
 247       simcall->issuer_->context_->set_wannadie();
 248   }
 249 }
 250
 251 ActivityImpl* ExecImpl::migrate(s4u::Host* to)
 252 {
 253   if (not MC_is_active() && not MC_record_replay_is_active()) {
 254     resource::Action* old_action = this->surf_action_;
 255     resource::Action* new_action = to->pimpl_cpu->execution_start(old_action->get_cost());
 256     new_action->set_remains(old_action->get_remains());
 257     new_action->set_activity(this);
 258     new_action->set_sharing_penalty(old_action->get_sharing_penalty());
 259
 260     // FIXME: the user-defined bound seem to not be kept by LMM, that seem to overwrite it for the multi-core modeling.
 261     // I hope that the user did not provide any.
 262
 263     old_action->set_activity(nullptr);
 264     old_action->cancel();
 265     old_action->unref();
 266     this->surf_action_ = new_action;
 267   }
 268
 269   on_migration(*this, to);
 270   return this;
 271 }
 272
 273 /*************
 274  * Callbacks *
 275  *************/
 276 xbt::signal<void(ExecImpl const&, s4u::Host*)> ExecImpl::on_migration;
 277
 278 } // namespace activity
 279 } // namespace kernel
 280 } // namespace simgrid