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