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/Engine.hpp"
  11 #include "simgrid/s4u/Exec.hpp"
  12 #include "src/kernel/actor/SimcallObserver.hpp"
  13 #include "src/mc/mc_replay.hpp"
  14 #include "src/surf/HostImpl.hpp"
  15 #include "src/surf/cpu_interface.hpp"
  16 #include "src/surf/surf_interface.hpp"
  17
  18 #include "simgrid/s4u/Host.hpp"
  19
  20 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(simix_process);
  21
  22 namespace simgrid {
  23 namespace kernel {
  24 namespace activity {
  25
  26 ExecImpl::ExecImpl()
  27 {
  28   piface_                = new s4u::Exec(this);
  29   actor::ActorImpl* self = actor::ActorImpl::self();
  30   if (self) {
  31     actor_ = self;
  32     self->activities_.emplace_back(this);
  33   }
  34 }
  35
  36 ExecImpl& ExecImpl::set_host(s4u::Host* host)
  37 {
  38   hosts_.assign(1, host);
  39   return *this;
  40 }
  41
  42 ExecImpl& ExecImpl::set_hosts(const std::vector<s4u::Host*>& hosts)
  43 {
  44   hosts_ = hosts;
  45   return *this;
  46 }
  47
  48 ExecImpl& ExecImpl::set_timeout(double timeout)
  49 {
  50   if (timeout >= 0 && not MC_is_active() && not MC_record_replay_is_active()) {
  51     timeout_detector_.reset(hosts_.front()->get_cpu()->sleep(timeout));
  52     timeout_detector_->set_activity(this);
  53   }
  54   return *this;
  55 }
  56
  57 ExecImpl& ExecImpl::set_flops_amount(double flops_amount)
  58 {
  59   flops_amounts_.assign(1, flops_amount);
  60   return *this;
  61 }
  62
  63 ExecImpl& ExecImpl::set_flops_amounts(const std::vector<double>& flops_amounts)
  64 {
  65   flops_amounts_ = flops_amounts;
  66   return *this;
  67 }
  68
  69 ExecImpl& ExecImpl::set_bytes_amounts(const std::vector<double>& bytes_amounts)
  70 {
  71   bytes_amounts_ = bytes_amounts;
  72
  73   return *this;
  74 }
  75
  76 ExecImpl* ExecImpl::start()
  77 {
  78   state_ = State::RUNNING;
  79   if (not MC_is_active() && not MC_record_replay_is_active()) {
  80     if (hosts_.size() == 1) {
  81       surf_action_ = hosts_.front()->get_cpu()->execution_start(flops_amounts_.front());
  82       surf_action_->set_sharing_penalty(sharing_penalty_);
  83       surf_action_->set_category(get_tracing_category());
  84
  85       if (bound_ > 0) {
  86         surf_action_->set_bound(bound_);
  87         surf_action_->set_user_bound(bound_);
  88       }
  89     } else {
  90       // FIXME[donassolo]: verify if all hosts belongs to the same netZone?
  91       auto host_model = hosts_.front()->get_netpoint()->get_englobing_zone()->get_host_model();
  92       surf_action_    = host_model->execute_parallel(hosts_, flops_amounts_.data(), bytes_amounts_.data(), -1);
  93     }
  94     surf_action_->set_activity(this);
  95     start_time_ = surf_action_->get_start_time();
  96   }
  97
  98   XBT_DEBUG("Create execute synchro %p: %s", this, get_cname());
  99   return this;
 100 }
 101
 102 double ExecImpl::get_seq_remaining_ratio()
 103 {
 104   return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains() / surf_action_->get_cost();
 105 }
 106
 107 double ExecImpl::get_par_remaining_ratio()
 108 {
 109   // parallel task: their remain is already between 0 and 1
 110   return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains();
 111 }
 112
 113 ExecImpl& ExecImpl::set_bound(double bound)
 114 {
 115   bound_ = bound;
 116   return *this;
 117 }
 118
 119 ExecImpl& ExecImpl::set_sharing_penalty(double sharing_penalty)
 120 {
 121   sharing_penalty_ = sharing_penalty;
 122   return *this;
 123 }
 124
 125 void ExecImpl::post()
 126 {
 127   xbt_assert(surf_action_ != nullptr);
 128   if (std::any_of(hosts_.begin(), hosts_.end(), [](const s4u::Host* host) { return not host->is_on(); })) {
 129     /* If one of the hosts running the synchro failed, notice it. This way, the asking
 130      * process can be killed if it runs on that host itself */
 131     state_ = State::FAILED;
 132   } else if (surf_action_->get_state() == resource::Action::State::FAILED) {
 133     /* If all the hosts are running the synchro didn't fail, then the synchro was canceled */
 134     state_ = State::CANCELED;
 135   } else if (timeout_detector_ && timeout_detector_->get_state() == resource::Action::State::FINISHED) {
 136     if (surf_action_->get_remains() > 0.0) {
 137       surf_action_->set_state(resource::Action::State::FAILED);
 138       state_ = State::TIMEOUT;
 139     } else {
 140       state_ = State::DONE;
 141     }
 142   } else {
 143     state_ = State::DONE;
 144   }
 145
 146   finish_time_ = surf_action_->get_finish_time();
 147
 148   clean_action();
 149   timeout_detector_.reset();
 150   if (actor_) {
 151     actor_->activities_.remove(this);
 152     actor_ = nullptr;
 153   }
 154   /* Answer all simcalls associated with the synchro */
 155   finish();
 156 }
 157
 158 void ExecImpl::finish()
 159 {
 160   XBT_DEBUG("ExecImpl::finish() in state %s", to_c_str(state_));
 161   while (not simcalls_.empty()) {
 162     smx_simcall_t simcall = simcalls_.front();
 163     simcalls_.pop_front();
 164
 165     /* If a waitany simcall is waiting for this synchro to finish, then remove it from the other synchros in the waitany
 166      * list. Afterwards, get the position of the actual synchro in the waitany list and return it as the result of the
 167      * simcall */
 168
 169     if (simcall->call_ == simix::Simcall::NONE) // FIXME: maybe a better way to handle this case
 170       continue;                                 // if process handling comm is killed
 171     if (auto* observer =
 172             dynamic_cast<kernel::actor::ExecutionWaitanySimcall*>(simcall->observer_)) { // simcall is a wait_any?
 173       const auto& execs = observer->get_execs();
 174
 175       for (auto* exec : execs) {
 176         exec->unregister_simcall(simcall);
 177
 178         if (simcall->timeout_cb_) {
 179           simcall->timeout_cb_->remove();
 180           simcall->timeout_cb_ = nullptr;
 181         }
 182       }
 183
 184       if (not MC_is_active() && not MC_record_replay_is_active()) {
 185         auto element = std::find(execs.begin(), execs.end(), this);
 186         int rank     = element != execs.end() ? static_cast<int>(std::distance(execs.begin(), element)) : -1;
 187         observer->set_result(rank);
 188       }
 189     }
 190     switch (state_) {
 191       case State::FAILED:
 192         simcall->issuer_->context_->set_wannadie();
 193         if (simcall->issuer_->get_host()->is_on())
 194           simcall->issuer_->exception_ = std::make_exception_ptr(HostFailureException(XBT_THROW_POINT, "Host failed"));
 195         /* else, the actor will be killed with no possibility to survive */
 196         break;
 197
 198       case State::CANCELED:
 199         simcall->issuer_->exception_ = std::make_exception_ptr(CancelException(XBT_THROW_POINT, "Execution Canceled"));
 200         break;
 201
 202       case State::TIMEOUT:
 203         simcall->issuer_->exception_ = std::make_exception_ptr(TimeoutException(XBT_THROW_POINT, "Timeouted"));
 204         break;
 205
 206       default:
 207         xbt_assert(state_ == State::DONE, "Internal error in ExecImpl::finish(): unexpected synchro state %s",
 208                    to_c_str(state_));
 209     }
 210
 211     simcall->issuer_->waiting_synchro_ = nullptr;
 212     /* Fail the process if the host is down */
 213     if (simcall->issuer_->get_host()->is_on())
 214       simcall->issuer_->simcall_answer();
 215     else
 216       simcall->issuer_->context_->set_wannadie();
 217   }
 218 }
 219
 220 ActivityImpl* ExecImpl::migrate(s4u::Host* to)
 221 {
 222   if (not MC_is_active() && not MC_record_replay_is_active()) {
 223     resource::Action* old_action = this->surf_action_;
 224     resource::Action* new_action = to->get_cpu()->execution_start(old_action->get_cost());
 225     new_action->set_remains(old_action->get_remains());
 226     new_action->set_activity(this);
 227     new_action->set_sharing_penalty(old_action->get_sharing_penalty());
 228     new_action->set_user_bound(old_action->get_user_bound());
 229
 230     old_action->set_activity(nullptr);
 231     old_action->cancel();
 232     old_action->unref();
 233     this->surf_action_ = new_action;
 234   }
 235
 236   on_migration(*this, to);
 237   return this;
 238 }
 239
 240 void ExecImpl::wait_any_for(actor::ActorImpl* issuer, const std::vector<ExecImpl*>& execs, double timeout)
 241 {
 242   if (timeout < 0.0) {
 243     issuer->simcall_.timeout_cb_ = nullptr;
 244   } else {
 245     issuer->simcall_.timeout_cb_ = timer::Timer::set(s4u::Engine::get_clock() + timeout, [issuer, &execs]() {
 246       issuer->simcall_.timeout_cb_ = nullptr;
 247       for (auto* exec : execs)
 248         exec->unregister_simcall(&issuer->simcall_);
 249       // default result (-1) is set in actor::ExecutionWaitanySimcall
 250       issuer->simcall_answer();
 251     });
 252   }
 253
 254   for (auto* exec : execs) {
 255     /* associate this simcall to the the synchro */
 256     exec->simcalls_.push_back(&issuer->simcall_);
 257
 258     /* see if the synchro is already finished */
 259     if (exec->state_ != State::WAITING && exec->state_ != State::RUNNING) {
 260       exec->finish();
 261       break;
 262     }
 263   }
 264 }
 265
 266 /*************
 267  * Callbacks *
 268  *************/
 269 xbt::signal<void(ExecImpl const&, s4u::Host*)> ExecImpl::on_migration;
 270
 271 } // namespace activity
 272 } // namespace kernel
 273 } // namespace simgrid