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 "simgrid/s4u/Exec.hpp"
   7 #include "src/kernel/activity/ExecImpl.hpp"
   8 #include "simgrid/Exception.hpp"
   9 #include "simgrid/modelchecker.h"
  10 #include "src/mc/mc_replay.hpp"
  11 #include "src/surf/HostImpl.hpp"
  12 #include "src/surf/cpu_interface.hpp"
  13 #include "src/surf/surf_interface.hpp"
  14
  15 #include "simgrid/s4u/Host.hpp"
  16
  17 #include <boost/range/algorithm.hpp>
  18
  19 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(simix_process);
  20
  21 void simcall_HANDLER_execution_waitany_for(smx_simcall_t simcall, simgrid::kernel::activity::ExecImpl* execs[],
  22                                            size_t count, double timeout)
  23 {
  24   if (timeout < 0.0) {
  25     simcall->timeout_cb_ = nullptr;
  26   } else {
  27     simcall->timeout_cb_ = simgrid::simix::Timer::set(SIMIX_get_clock() + timeout, [simcall, execs, count]() {
  28       simcall->timeout_cb_ = nullptr;
  29       for (size_t i = 0; i < count; i++) {
  30         // Remove the first occurrence of simcall:
  31         auto* exec = execs[i];
  32         auto j     = boost::range::find(exec->simcalls_, simcall);
  33         if (j != exec->simcalls_.end())
  34           exec->simcalls_.erase(j);
  35       }
  36       simcall_execution_waitany_for__set__result(simcall, -1);
  37       simcall->issuer_->simcall_answer();
  38     });
  39   }
  40
  41   for (size_t i = 0; i < count; i++) {
  42     /* associate this simcall to the the synchro */
  43     auto* exec = execs[i];
  44     exec->simcalls_.push_back(simcall);
  45
  46     /* see if the synchro is already finished */
  47     if (exec->state_ != simgrid::kernel::activity::State::WAITING &&
  48         exec->state_ != simgrid::kernel::activity::State::RUNNING) {
  49       exec->finish();
  50       break;
  51     }
  52   }
  53 }
  54
  55 namespace simgrid {
  56 namespace kernel {
  57 namespace activity {
  58
  59 ExecImpl::ExecImpl()
  60 {
  61   piface_ = new s4u::Exec(this);
  62   actor::ActorImpl* self = actor::ActorImpl::self();
  63   if (self) {
  64     actor_ = self;
  65     self->activities_.emplace_back(this);
  66   }
  67 }
  68
  69 ExecImpl& ExecImpl::set_host(s4u::Host* host)
  70 {
  71   hosts_.assign(1, host);
  72   return *this;
  73 }
  74
  75 ExecImpl& ExecImpl::set_hosts(const std::vector<s4u::Host*>& hosts)
  76 {
  77   hosts_ = hosts;
  78   return *this;
  79 }
  80
  81 ExecImpl& ExecImpl::set_timeout(double timeout)
  82 {
  83   if (timeout >= 0 && not MC_is_active() && not MC_record_replay_is_active()) {
  84     timeout_detector_.reset(hosts_.front()->pimpl_cpu->sleep(timeout));
  85     timeout_detector_->set_activity(this);
  86   }
  87   return *this;
  88 }
  89
  90 ExecImpl& ExecImpl::set_flops_amount(double flops_amount)
  91 {
  92   flops_amounts_.assign(1, flops_amount);
  93   return *this;
  94 }
  95
  96 ExecImpl& ExecImpl::set_flops_amounts(const std::vector<double>& flops_amounts)
  97 {
  98   flops_amounts_ = flops_amounts;
  99   return *this;
 100 }
 101
 102 ExecImpl& ExecImpl::set_bytes_amounts(const std::vector<double>& bytes_amounts)
 103 {
 104   bytes_amounts_ = bytes_amounts;
 105
 106   return *this;
 107 }
 108
 109 ExecImpl* ExecImpl::start()
 110 {
 111   state_ = State::RUNNING;
 112   if (not MC_is_active() && not MC_record_replay_is_active()) {
 113     if (hosts_.size() == 1) {
 114       surf_action_ = hosts_.front()->pimpl_cpu->execution_start(flops_amounts_.front());
 115       surf_action_->set_sharing_penalty(sharing_penalty_);
 116       surf_action_->set_category(get_tracing_category());
 117
 118       if (bound_ > 0)
 119         surf_action_->set_bound(bound_);
 120     } else {
 121       surf_action_ = surf_host_model->execute_parallel(hosts_, flops_amounts_.data(), bytes_amounts_.data(), -1);
 122     }
 123     surf_action_->set_activity(this);
 124   }
 125
 126   XBT_DEBUG("Create execute synchro %p: %s", this, get_cname());
 127   return this;
 128 }
 129
 130 double ExecImpl::get_seq_remaining_ratio()
 131 {
 132   return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains() / surf_action_->get_cost();
 133 }
 134
 135 double ExecImpl::get_par_remaining_ratio()
 136 {
 137   // parallel task: their remain is already between 0 and 1
 138   return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains();
 139 }
 140
 141 ExecImpl& ExecImpl::set_bound(double bound)
 142 {
 143   bound_ = bound;
 144   return *this;
 145 }
 146
 147 ExecImpl& ExecImpl::set_sharing_penalty(double sharing_penalty)
 148 {
 149   sharing_penalty_ = sharing_penalty;
 150   return *this;
 151 }
 152
 153 void ExecImpl::post()
 154 {
 155   xbt_assert(surf_action_ != nullptr);
 156   if (hosts_.size() == 1 && not hosts_.front()->is_on()) { /* FIXME: handle resource failure for parallel tasks too */
 157     /* If the host running the synchro failed, notice it. This way, the asking
 158      * process can be killed if it runs on that host itself */
 159     state_ = State::FAILED;
 160   } else if (surf_action_->get_state() == resource::Action::State::FAILED) {
 161     /* If the host running the synchro didn't fail, then the synchro was canceled */
 162     state_ = State::CANCELED;
 163   } else if (timeout_detector_ && timeout_detector_->get_state() == resource::Action::State::FINISHED) {
 164     if (surf_action_->get_remains() > 0.0) {
 165       surf_action_->set_state(resource::Action::State::FAILED);
 166       state_ = State::TIMEOUT;
 167     } else {
 168       state_ = State::DONE;
 169     }
 170   } else {
 171     state_ = State::DONE;
 172   }
 173
 174   get_iface()->set_finish_time(surf_action_->get_finish_time());
 175
 176   clean_action();
 177   timeout_detector_.reset();
 178   if (actor_) {
 179     actor_->activities_.remove(this);
 180     actor_ = nullptr;
 181   }
 182   /* Answer all simcalls associated with the synchro */
 183   finish();
 184 }
 185
 186 void ExecImpl::finish()
 187 {
 188   while (not simcalls_.empty()) {
 189     smx_simcall_t simcall = simcalls_.front();
 190     simcalls_.pop_front();
 191
 192     /* If a waitany simcall is waiting for this synchro to finish, then remove it from the other synchros in the waitany
 193      * list. Afterwards, get the position of the actual synchro in the waitany list and return it as the result of the
 194      * simcall */
 195
 196     if (simcall->call_ == simix::Simcall::NONE) // FIXME: maybe a better way to handle this case
 197       continue;                        // if process handling comm is killed
 198     if (simcall->call_ == simix::Simcall::EXECUTION_WAITANY_FOR) {
 199       simgrid::kernel::activity::ExecImpl** execs = simcall_execution_waitany_for__get__execs(simcall);
 200       size_t count                                = simcall_execution_waitany_for__get__count(simcall);
 201
 202       for (size_t i = 0; i < count; i++) {
 203         // Remove the first occurrence of simcall:
 204         auto* exec = execs[i];
 205         auto j     = boost::range::find(exec->simcalls_, simcall);
 206         if (j != exec->simcalls_.end())
 207           exec->simcalls_.erase(j);
 208
 209         if (simcall->timeout_cb_) {
 210           simcall->timeout_cb_->remove();
 211           simcall->timeout_cb_ = nullptr;
 212         }
 213       }
 214
 215       if (not MC_is_active() && not MC_record_replay_is_active()) {
 216         ExecImpl** element = std::find(execs, execs + count, this);
 217         int rank           = (element != execs + count) ? element - execs : -1;
 218         simcall_execution_waitany_for__set__result(simcall, rank);
 219       }
 220     }
 221     switch (state_) {
 222       case State::DONE:
 223         /* do nothing, synchro done */
 224         XBT_DEBUG("ExecImpl::finish(): execution successful");
 225         break;
 226
 227       case State::FAILED:
 228         XBT_DEBUG("ExecImpl::finish(): host '%s' failed", simcall->issuer_->get_host()->get_cname());
 229         simcall->issuer_->context_->set_wannadie();
 230         if (simcall->issuer_->get_host()->is_on())
 231           simcall->issuer_->exception_ =
 232               std::make_exception_ptr(simgrid::HostFailureException(XBT_THROW_POINT, "Host failed"));
 233         /* else, the actor will be killed with no possibility to survive */
 234         break;
 235
 236       case State::CANCELED:
 237         XBT_DEBUG("ExecImpl::finish(): execution canceled");
 238         simcall->issuer_->exception_ =
 239             std::make_exception_ptr(simgrid::CancelException(XBT_THROW_POINT, "Execution Canceled"));
 240         break;
 241
 242       case State::TIMEOUT:
 243         XBT_DEBUG("ExecImpl::finish(): execution timeouted");
 244         simcall->issuer_->exception_ = std::make_exception_ptr(simgrid::TimeoutException(XBT_THROW_POINT, "Timeouted"));
 245         break;
 246
 247       default:
 248         xbt_die("Internal error in ExecImpl::finish(): unexpected synchro state %d", static_cast<int>(state_));
 249     }
 250
 251     simcall->issuer_->waiting_synchro_ = nullptr;
 252     /* Fail the process if the host is down */
 253     if (simcall->issuer_->get_host()->is_on())
 254       simcall->issuer_->simcall_answer();
 255     else
 256       simcall->issuer_->context_->set_wannadie();
 257   }
 258 }
 259
 260 ActivityImpl* ExecImpl::migrate(s4u::Host* to)
 261 {
 262   if (not MC_is_active() && not MC_record_replay_is_active()) {
 263     resource::Action* old_action = this->surf_action_;
 264     resource::Action* new_action = to->pimpl_cpu->execution_start(old_action->get_cost());
 265     new_action->set_remains(old_action->get_remains());
 266     new_action->set_activity(this);
 267     new_action->set_sharing_penalty(old_action->get_sharing_penalty());
 268
 269     // FIXME: the user-defined bound seem to not be kept by LMM, that seem to overwrite it for the multi-core modeling.
 270     // I hope that the user did not provide any.
 271
 272     old_action->set_activity(nullptr);
 273     old_action->cancel();
 274     old_action->unref();
 275     this->surf_action_ = new_action;
 276   }
 277
 278   on_migration(*this, to);
 279   return this;
 280 }
 281
 282 /*************
 283  * Callbacks *
 284  *************/
 285 xbt::signal<void(ExecImpl const&, s4u::Host*)> ExecImpl::on_migration;
 286
 287 } // namespace activity
 288 } // namespace kernel
 289 } // namespace simgrid