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/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)
  21 {
  22   XBT_DEBUG("Wait for execution of synchro %p, state %d", synchro, (int)synchro->state_);
  23
  24   /* Associate this simcall to the synchro */
  25   synchro->register_simcall(simcall);
  26
  27   /* set surf's synchro */
  28   if (MC_is_active() || MC_record_replay_is_active()) {
  29     synchro->state_ = simgrid::kernel::activity::State::DONE;
  30     synchro->finish();
  31     return;
  32   }
  33
  34   /* If the synchro is already finished then perform the error handling */
  35   if (synchro->state_ != simgrid::kernel::activity::State::RUNNING)
  36     synchro->finish();
  37 }
  38
  39 void simcall_HANDLER_execution_test(smx_simcall_t simcall, simgrid::kernel::activity::ExecImpl* synchro)
  40 {
  41   bool res = (synchro->state_ != simgrid::kernel::activity::State::WAITING &&
  42               synchro->state_ != simgrid::kernel::activity::State::RUNNING);
  43   if (res) {
  44     synchro->simcalls_.push_back(simcall);
  45     synchro->finish();
  46   } else {
  47     simcall->issuer_->simcall_answer();
  48   }
  49   simcall_execution_test__set__result(simcall, res);
  50 }
  51
  52 void simcall_HANDLER_execution_waitany_for(smx_simcall_t simcall, simgrid::kernel::activity::ExecImpl* execs[],
  53                                            size_t count, double timeout)
  54 {
  55   if (timeout < 0.0) {
  56     simcall->timeout_cb_ = nullptr;
  57   } else {
  58     simcall->timeout_cb_ = simgrid::simix::Timer::set(SIMIX_get_clock() + timeout, [simcall, execs, count]() {
  59       for (size_t i = 0; i < count; i++) {
  60         // Remove the first occurrence of simcall:
  61         auto* exec = execs[i];
  62         auto j     = boost::range::find(exec->simcalls_, simcall);
  63         if (j != exec->simcalls_.end())
  64           exec->simcalls_.erase(j);
  65       }
  66       simcall_execution_waitany_for__set__result(simcall, -1);
  67       simcall->issuer_->simcall_answer();
  68     });
  69   }
  70
  71   for (size_t i = 0; i < count; i++) {
  72     /* associate this simcall to the the synchro */
  73     auto* exec = execs[i];
  74     exec->simcalls_.push_back(simcall);
  75
  76     /* see if the synchro is already finished */
  77     if (exec->state_ != simgrid::kernel::activity::State::WAITING &&
  78         exec->state_ != simgrid::kernel::activity::State::RUNNING) {
  79       exec->finish();
  80       break;
  81     }
  82   }
  83 }
  84
  85 namespace simgrid {
  86 namespace kernel {
  87 namespace activity {
  88
  89 ExecImpl::~ExecImpl()
  90 {
  91   if (timeout_detector_)
  92     timeout_detector_->unref();
  93   XBT_DEBUG("Destroy exec %p", this);
  94 }
  95
  96 ExecImpl& ExecImpl::set_host(s4u::Host* host)
  97 {
  98   if (not hosts_.empty())
  99     hosts_.clear();
 100   hosts_.push_back(host);
 101   return *this;
 102 }
 103
 104 ExecImpl& ExecImpl::set_hosts(const std::vector<s4u::Host*>& hosts)
 105 {
 106   hosts_ = hosts;
 107   return *this;
 108 }
 109
 110 ExecImpl& ExecImpl::set_timeout(double timeout)
 111 {
 112   if (timeout > 0 && not MC_is_active() && not MC_record_replay_is_active()) {
 113     timeout_detector_ = hosts_.front()->pimpl_cpu->sleep(timeout);
 114     timeout_detector_->set_activity(this);
 115   }
 116   return *this;
 117 }
 118
 119 ExecImpl& ExecImpl::set_flops_amount(double flops_amount)
 120 {
 121   if (not flops_amounts_.empty())
 122     flops_amounts_.clear();
 123   flops_amounts_.push_back(flops_amount);
 124   return *this;
 125 }
 126
 127 ExecImpl& ExecImpl::set_flops_amounts(const std::vector<double>& flops_amounts)
 128 {
 129   flops_amounts_ = flops_amounts;
 130   return *this;
 131 }
 132
 133 ExecImpl& ExecImpl::set_bytes_amounts(const std::vector<double>& bytes_amounts)
 134 {
 135   bytes_amounts_ = bytes_amounts;
 136
 137   return *this;
 138 }
 139
 140 ExecImpl* ExecImpl::start()
 141 {
 142   state_ = State::RUNNING;
 143   if (not MC_is_active() && not MC_record_replay_is_active()) {
 144     if (hosts_.size() == 1) {
 145       surf_action_ = hosts_.front()->pimpl_cpu->execution_start(flops_amounts_.front());
 146       surf_action_->set_sharing_penalty(sharing_penalty_);
 147       surf_action_->set_category(get_tracing_category());
 148
 149       if (bound_ > 0)
 150         surf_action_->set_bound(bound_);
 151     } else {
 152       surf_action_ = surf_host_model->execute_parallel(hosts_, flops_amounts_.data(), bytes_amounts_.data(), -1);
 153     }
 154     surf_action_->set_activity(this);
 155   }
 156
 157   XBT_DEBUG("Create execute synchro %p: %s", this, get_cname());
 158   return this;
 159 }
 160
 161 double ExecImpl::get_seq_remaining_ratio()
 162 {
 163   return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains() / surf_action_->get_cost();
 164 }
 165
 166 double ExecImpl::get_par_remaining_ratio()
 167 {
 168   // parallel task: their remain is already between 0 and 1
 169   return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains();
 170 }
 171
 172 ExecImpl& ExecImpl::set_bound(double bound)
 173 {
 174   bound_ = bound;
 175   return *this;
 176 }
 177
 178 ExecImpl& ExecImpl::set_sharing_penalty(double sharing_penalty)
 179 {
 180   sharing_penalty_ = sharing_penalty;
 181   return *this;
 182 }
 183
 184 void ExecImpl::post()
 185 {
 186   if (hosts_.size() == 1 && not hosts_.front()->is_on()) { /* FIXME: handle resource failure for parallel tasks too */
 187     /* If the host running the synchro failed, notice it. This way, the asking
 188      * process can be killed if it runs on that host itself */
 189     state_ = State::FAILED;
 190   } else if (surf_action_ && surf_action_->get_state() == resource::Action::State::FAILED) {
 191     /* If the host running the synchro didn't fail, then the synchro was canceled */
 192     state_ = State::CANCELED;
 193   } else if (timeout_detector_ && timeout_detector_->get_state() == resource::Action::State::FINISHED) {
 194     state_ = State::TIMEOUT;
 195   } else {
 196     state_ = State::DONE;
 197   }
 198
 199   clean_action();
 200
 201   if (timeout_detector_) {
 202     timeout_detector_->unref();
 203     timeout_detector_ = nullptr;
 204   }
 205
 206   /* Answer all simcalls associated with the synchro */
 207   finish();
 208 }
 209
 210 void ExecImpl::finish()
 211 {
 212   while (not simcalls_.empty()) {
 213     smx_simcall_t simcall = simcalls_.front();
 214     simcalls_.pop_front();
 215
 216     /* If a waitany simcall is waiting for this synchro to finish, then remove it from the other synchros in the waitany
 217      * list. Afterwards, get the position of the actual synchro in the waitany list and return it as the result of the
 218      * simcall */
 219
 220     if (simcall->call_ == SIMCALL_NONE) // FIXME: maybe a better way to handle this case
 221       continue;                        // if process handling comm is killed
 222     if (simcall->call_ == SIMCALL_EXECUTION_WAITANY_FOR) {
 223       simgrid::kernel::activity::ExecImpl** execs = simcall_execution_waitany_for__get__execs(simcall);
 224       size_t count                                = simcall_execution_waitany_for__get__count(simcall);
 225
 226       for (size_t i = 0; i < count; i++) {
 227         // Remove the first occurrence of simcall:
 228         auto* exec = execs[i];
 229         auto j     = boost::range::find(exec->simcalls_, simcall);
 230         if (j != exec->simcalls_.end())
 231           exec->simcalls_.erase(j);
 232
 233         if (simcall->timeout_cb_) {
 234           simcall->timeout_cb_->remove();
 235           simcall->timeout_cb_ = nullptr;
 236         }
 237       }
 238
 239       if (not MC_is_active() && not MC_record_replay_is_active()) {
 240         ExecImpl** element = std::find(execs, execs + count, this);
 241         int rank           = (element != execs + count) ? element - execs : -1;
 242         simcall_execution_waitany_for__set__result(simcall, rank);
 243       }
 244     }
 245
 246     switch (state_) {
 247
 248       case State::DONE:
 249         /* do nothing, synchro done */
 250         XBT_DEBUG("ExecImpl::finish(): execution successful");
 251         break;
 252
 253       case State::FAILED:
 254         XBT_DEBUG("ExecImpl::finish(): host '%s' failed", simcall->issuer_->get_host()->get_cname());
 255         simcall->issuer_->context_->iwannadie = true;
 256         if (simcall->issuer_->get_host()->is_on())
 257           simcall->issuer_->exception_ =
 258               std::make_exception_ptr(simgrid::HostFailureException(XBT_THROW_POINT, "Host failed"));
 259         /* else, the actor will be killed with no possibility to survive */
 260         break;
 261
 262       case State::CANCELED:
 263         XBT_DEBUG("ExecImpl::finish(): execution canceled");
 264         simcall->issuer_->exception_ =
 265             std::make_exception_ptr(simgrid::CancelException(XBT_THROW_POINT, "Execution Canceled"));
 266         break;
 267
 268       case State::TIMEOUT:
 269         XBT_DEBUG("ExecImpl::finish(): execution timeouted");
 270         simcall->issuer_->exception_ = std::make_exception_ptr(simgrid::TimeoutException(XBT_THROW_POINT, "Timeouted"));
 271         break;
 272
 273       default:
 274         xbt_die("Internal error in ExecImpl::finish(): unexpected synchro state %d", static_cast<int>(state_));
 275     }
 276
 277     simcall->issuer_->waiting_synchro = nullptr;
 278     /* Fail the process if the host is down */
 279     if (simcall->issuer_->get_host()->is_on())
 280       simcall->issuer_->simcall_answer();
 281     else
 282       simcall->issuer_->context_->iwannadie = true;
 283   }
 284 }
 285
 286 ActivityImpl* ExecImpl::migrate(s4u::Host* to)
 287 {
 288   if (not MC_is_active() && not MC_record_replay_is_active()) {
 289     resource::Action* old_action = this->surf_action_;
 290     resource::Action* new_action = to->pimpl_cpu->execution_start(old_action->get_cost());
 291     new_action->set_remains(old_action->get_remains());
 292     new_action->set_activity(this);
 293     new_action->set_sharing_penalty(old_action->get_sharing_penalty());
 294
 295     // FIXME: the user-defined bound seem to not be kept by LMM, that seem to overwrite it for the multi-core modeling.
 296     // I hope that the user did not provide any.
 297
 298     old_action->set_activity(nullptr);
 299     old_action->cancel();
 300     old_action->unref();
 301     this->surf_action_ = new_action;
 302   }
 303
 304   on_migration(*this, to);
 305   return this;
 306 }
 307
 308 /*************
 309  * Callbacks *
 310  *************/
 311 xbt::signal<void(ExecImpl const&, s4u::Host*)> ExecImpl::on_migration;
 312
 313 } // namespace activity
 314 } // namespace kernel
 315 } // namespace simgrid