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/kernel/routing/NetPoint.hpp"
   9 #include "simgrid/modelchecker.h"
  10 #include "simgrid/s4u/Exec.hpp"
  11 #include "src/mc/mc_replay.hpp"
  12 #include "src/surf/HostImpl.hpp"
  13 #include "src/surf/cpu_interface.hpp"
  14 #include "src/surf/surf_interface.hpp"
  15
  16 #include "simgrid/s4u/Host.hpp"
  17
  18 #include <boost/range/algorithm.hpp>
  19
  20 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(simix_process);
  21
  22 void simcall_HANDLER_execution_waitany_for(smx_simcall_t simcall, simgrid::kernel::activity::ExecImpl* execs[],
  23                                            size_t count, double timeout)
  24 {
  25   if (timeout < 0.0) {
  26     simcall->timeout_cb_ = nullptr;
  27   } else {
  28     simcall->timeout_cb_ = simgrid::simix::Timer::set(SIMIX_get_clock() + timeout, [simcall, execs, count]() {
  29       simcall->timeout_cb_ = nullptr;
  30       for (size_t i = 0; i < count; i++) {
  31         // Remove the first occurrence of simcall:
  32         auto* exec = execs[i];
  33         auto j     = boost::range::find(exec->simcalls_, simcall);
  34         if (j != exec->simcalls_.end())
  35           exec->simcalls_.erase(j);
  36       }
  37       simcall_execution_waitany_for__set__result(simcall, -1);
  38       simcall->issuer_->simcall_answer();
  39     });
  40   }
  41
  42   for (size_t i = 0; i < count; i++) {
  43     /* associate this simcall to the the synchro */
  44     auto* exec = execs[i];
  45     exec->simcalls_.push_back(simcall);
  46
  47     /* see if the synchro is already finished */
  48     if (exec->state_ != simgrid::kernel::activity::State::WAITING &&
  49         exec->state_ != simgrid::kernel::activity::State::RUNNING) {
  50       exec->finish();
  51       break;
  52     }
  53   }
  54 }
  55
  56 namespace simgrid {
  57 namespace kernel {
  58 namespace activity {
  59
  60 ExecImpl::ExecImpl()
  61 {
  62   piface_                = new s4u::Exec(this);
  63   actor::ActorImpl* self = actor::ActorImpl::self();
  64   if (self) {
  65     actor_ = self;
  66     self->activities_.emplace_back(this);
  67   }
  68 }
  69
  70 ExecImpl& ExecImpl::set_host(s4u::Host* host)
  71 {
  72   hosts_.assign(1, host);
  73   return *this;
  74 }
  75
  76 ExecImpl& ExecImpl::set_hosts(const std::vector<s4u::Host*>& hosts)
  77 {
  78   hosts_ = hosts;
  79   return *this;
  80 }
  81
  82 ExecImpl& ExecImpl::set_timeout(double timeout)
  83 {
  84   if (timeout >= 0 && not MC_is_active() && not MC_record_replay_is_active()) {
  85     timeout_detector_.reset(hosts_.front()->pimpl_cpu->sleep(timeout));
  86     timeout_detector_->set_activity(this);
  87   }
  88   return *this;
  89 }
  90
  91 ExecImpl& ExecImpl::set_flops_amount(double flops_amount)
  92 {
  93   flops_amounts_.assign(1, flops_amount);
  94   return *this;
  95 }
  96
  97 ExecImpl& ExecImpl::set_flops_amounts(const std::vector<double>& flops_amounts)
  98 {
  99   flops_amounts_ = flops_amounts;
 100   return *this;
 101 }
 102
 103 ExecImpl& ExecImpl::set_bytes_amounts(const std::vector<double>& bytes_amounts)
 104 {
 105   bytes_amounts_ = bytes_amounts;
 106
 107   return *this;
 108 }
 109
 110 ExecImpl* ExecImpl::start()
 111 {
 112   state_ = State::RUNNING;
 113   if (not MC_is_active() && not MC_record_replay_is_active()) {
 114     if (hosts_.size() == 1) {
 115       surf_action_ = hosts_.front()->pimpl_cpu->execution_start(flops_amounts_.front());
 116       surf_action_->set_sharing_penalty(sharing_penalty_);
 117       surf_action_->set_category(get_tracing_category());
 118
 119       if (bound_ > 0) {
 120         surf_action_->set_bound(bound_);
 121         surf_action_->set_user_bound(bound_);
 122       }
 123     } else {
 124       // FIXME[donassolo]: verify if all hosts belongs to the same netZone?
 125       auto host_model = hosts_.front()->get_netpoint()->get_englobing_zone()->get_host_model();
 126       surf_action_    = host_model->execute_parallel(hosts_, flops_amounts_.data(), bytes_amounts_.data(), -1);
 127     }
 128     surf_action_->set_activity(this);
 129   }
 130
 131   XBT_DEBUG("Create execute synchro %p: %s", this, get_cname());
 132   return this;
 133 }
 134
 135 double ExecImpl::get_seq_remaining_ratio()
 136 {
 137   return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains() / surf_action_->get_cost();
 138 }
 139
 140 double ExecImpl::get_par_remaining_ratio()
 141 {
 142   // parallel task: their remain is already between 0 and 1
 143   return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains();
 144 }
 145
 146 ExecImpl& ExecImpl::set_bound(double bound)
 147 {
 148   bound_ = bound;
 149   return *this;
 150 }
 151
 152 ExecImpl& ExecImpl::set_sharing_penalty(double sharing_penalty)
 153 {
 154   sharing_penalty_ = sharing_penalty;
 155   return *this;
 156 }
 157
 158 void ExecImpl::post()
 159 {
 160   xbt_assert(surf_action_ != nullptr);
 161   if (hosts_.size() == 1 && not hosts_.front()->is_on()) { /* FIXME: handle resource failure for parallel tasks too */
 162     /* If the host running the synchro failed, notice it. This way, the asking
 163      * process can be killed if it runs on that host itself */
 164     state_ = State::FAILED;
 165   } else if (surf_action_->get_state() == resource::Action::State::FAILED) {
 166     /* If the host running the synchro didn't fail, then the synchro was canceled */
 167     state_ = State::CANCELED;
 168   } else if (timeout_detector_ && timeout_detector_->get_state() == resource::Action::State::FINISHED) {
 169     if (surf_action_->get_remains() > 0.0) {
 170       surf_action_->set_state(resource::Action::State::FAILED);
 171       state_ = State::TIMEOUT;
 172     } else {
 173       state_ = State::DONE;
 174     }
 175   } else {
 176     state_ = State::DONE;
 177   }
 178
 179   get_iface()->set_finish_time(surf_action_->get_finish_time());
 180
 181   clean_action();
 182   timeout_detector_.reset();
 183   if (actor_) {
 184     actor_->activities_.remove(this);
 185     actor_ = nullptr;
 186   }
 187   /* Answer all simcalls associated with the synchro */
 188   finish();
 189 }
 190
 191 void ExecImpl::finish()
 192 {
 193   while (not simcalls_.empty()) {
 194     smx_simcall_t simcall = simcalls_.front();
 195     simcalls_.pop_front();
 196
 197     /* If a waitany simcall is waiting for this synchro to finish, then remove it from the other synchros in the waitany
 198      * list. Afterwards, get the position of the actual synchro in the waitany list and return it as the result of the
 199      * simcall */
 200
 201     if (simcall->call_ == simix::Simcall::NONE) // FIXME: maybe a better way to handle this case
 202       continue;                                 // if process handling comm is killed
 203     if (simcall->call_ == simix::Simcall::EXECUTION_WAITANY_FOR) {
 204       simgrid::kernel::activity::ExecImpl** execs = simcall_execution_waitany_for__get__execs(simcall);
 205       size_t count                                = simcall_execution_waitany_for__get__count(simcall);
 206
 207       for (size_t i = 0; i < count; i++) {
 208         // Remove the first occurrence of simcall:
 209         auto* exec = execs[i];
 210         auto j     = boost::range::find(exec->simcalls_, simcall);
 211         if (j != exec->simcalls_.end())
 212           exec->simcalls_.erase(j);
 213
 214         if (simcall->timeout_cb_) {
 215           simcall->timeout_cb_->remove();
 216           simcall->timeout_cb_ = nullptr;
 217         }
 218       }
 219
 220       if (not MC_is_active() && not MC_record_replay_is_active()) {
 221         ExecImpl** element = std::find(execs, execs + count, this);
 222         int rank           = (element != execs + count) ? element - execs : -1;
 223         simcall_execution_waitany_for__set__result(simcall, rank);
 224       }
 225     }
 226     switch (state_) {
 227       case State::DONE:
 228         /* do nothing, synchro done */
 229         XBT_DEBUG("ExecImpl::finish(): execution successful");
 230         break;
 231
 232       case State::FAILED:
 233         XBT_DEBUG("ExecImpl::finish(): host '%s' failed", simcall->issuer_->get_host()->get_cname());
 234         simcall->issuer_->context_->set_wannadie();
 235         if (simcall->issuer_->get_host()->is_on())
 236           simcall->issuer_->exception_ =
 237               std::make_exception_ptr(simgrid::HostFailureException(XBT_THROW_POINT, "Host failed"));
 238         /* else, the actor will be killed with no possibility to survive */
 239         break;
 240
 241       case State::CANCELED:
 242         XBT_DEBUG("ExecImpl::finish(): execution canceled");
 243         simcall->issuer_->exception_ =
 244             std::make_exception_ptr(simgrid::CancelException(XBT_THROW_POINT, "Execution Canceled"));
 245         break;
 246
 247       case State::TIMEOUT:
 248         XBT_DEBUG("ExecImpl::finish(): execution timeouted");
 249         simcall->issuer_->exception_ = std::make_exception_ptr(simgrid::TimeoutException(XBT_THROW_POINT, "Timeouted"));
 250         break;
 251
 252       default:
 253         xbt_die("Internal error in ExecImpl::finish(): unexpected synchro state %d", static_cast<int>(state_));
 254     }
 255
 256     simcall->issuer_->waiting_synchro_ = nullptr;
 257     /* Fail the process if the host is down */
 258     if (simcall->issuer_->get_host()->is_on())
 259       simcall->issuer_->simcall_answer();
 260     else
 261       simcall->issuer_->context_->set_wannadie();
 262   }
 263 }
 264
 265 ActivityImpl* ExecImpl::migrate(s4u::Host* to)
 266 {
 267   if (not MC_is_active() && not MC_record_replay_is_active()) {
 268     resource::Action* old_action = this->surf_action_;
 269     resource::Action* new_action = to->pimpl_cpu->execution_start(old_action->get_cost());
 270     new_action->set_remains(old_action->get_remains());
 271     new_action->set_activity(this);
 272     new_action->set_sharing_penalty(old_action->get_sharing_penalty());
 273
 274     // FIXME: the user-defined bound seem to not be kept by LMM, that seem to overwrite it for the multi-core modeling.
 275     // I hope that the user did not provide any.
 276
 277     old_action->set_activity(nullptr);
 278     old_action->cancel();
 279     old_action->unref();
 280     this->surf_action_ = new_action;
 281   }
 282
 283   on_migration(*this, to);
 284   return this;
 285 }
 286
 287 /*************
 288  * Callbacks *
 289  *************/
 290 xbt::signal<void(ExecImpl const&, s4u::Host*)> ExecImpl::on_migration;
 291
 292 } // namespace activity
 293 } // namespace kernel
 294 } // namespace simgrid