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     start_time_ = surf_action_->get_start_time();
  95   }
  96
  97   XBT_DEBUG("Create execute synchro %p: %s", this, get_cname());
  98   return this;
  99 }
 100
 101 double ExecImpl::get_seq_remaining_ratio()
 102 {
 103   return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains() / surf_action_->get_cost();
 104 }
 105
 106 double ExecImpl::get_par_remaining_ratio()
 107 {
 108   // parallel task: their remain is already between 0 and 1
 109   return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains();
 110 }
 111
 112 ExecImpl& ExecImpl::set_bound(double bound)
 113 {
 114   bound_ = bound;
 115   return *this;
 116 }
 117
 118 ExecImpl& ExecImpl::set_sharing_penalty(double sharing_penalty)
 119 {
 120   sharing_penalty_ = sharing_penalty;
 121   return *this;
 122 }
 123
 124 void ExecImpl::post()
 125 {
 126   xbt_assert(surf_action_ != nullptr);
 127   if (std::any_of(hosts_.begin(), hosts_.end(), [](const s4u::Host* host) { return not host->is_on(); })) {
 128     /* If one of the hosts running the synchro failed, notice it. This way, the asking
 129      * process can be killed if it runs on that host itself */
 130     state_ = State::FAILED;
 131   } else if (surf_action_->get_state() == resource::Action::State::FAILED) {
 132     /* If all the hosts are running the synchro didn't fail, then the synchro was canceled */
 133     state_ = State::CANCELED;
 134   } else if (timeout_detector_ && timeout_detector_->get_state() == resource::Action::State::FINISHED) {
 135     if (surf_action_->get_remains() > 0.0) {
 136       surf_action_->set_state(resource::Action::State::FAILED);
 137       state_ = State::TIMEOUT;
 138     } else {
 139       state_ = State::DONE;
 140     }
 141   } else {
 142     state_ = State::DONE;
 143   }
 144
 145   finish_time_ = surf_action_->get_finish_time();
 146
 147   clean_action();
 148   timeout_detector_.reset();
 149   if (actor_) {
 150     actor_->activities_.remove(this);
 151     actor_ = nullptr;
 152   }
 153   /* Answer all simcalls associated with the synchro */
 154   finish();
 155 }
 156
 157 void ExecImpl::finish()
 158 {
 159   XBT_DEBUG("ExecImpl::finish() in state %s", to_c_str(state_));
 160   while (not simcalls_.empty()) {
 161     smx_simcall_t simcall = simcalls_.front();
 162     simcalls_.pop_front();
 163
 164     /* If a waitany simcall is waiting for this synchro to finish, then remove it from the other synchros in the waitany
 165      * list. Afterwards, get the position of the actual synchro in the waitany list and return it as the result of the
 166      * simcall */
 167
 168     if (simcall->call_ == simix::Simcall::NONE) // FIXME: maybe a better way to handle this case
 169       continue;                                 // if process handling comm is killed
 170     if (auto* observer =
 171             dynamic_cast<kernel::actor::ExecutionWaitanySimcall*>(simcall->observer_)) { // simcall is a wait_any?
 172       const auto& execs = observer->get_execs();
 173
 174       for (auto* exec : execs) {
 175         exec->unregister_simcall(simcall);
 176
 177         if (simcall->timeout_cb_) {
 178           simcall->timeout_cb_->remove();
 179           simcall->timeout_cb_ = nullptr;
 180         }
 181       }
 182
 183       if (not MC_is_active() && not MC_record_replay_is_active()) {
 184         auto element = std::find(execs.begin(), execs.end(), this);
 185         int rank     = element != execs.end() ? static_cast<int>(std::distance(execs.begin(), element)) : -1;
 186         observer->set_result(rank);
 187       }
 188     }
 189     switch (state_) {
 190       case State::FAILED:
 191         simcall->issuer_->context_->set_wannadie();
 192         if (simcall->issuer_->get_host()->is_on())
 193           simcall->issuer_->exception_ = std::make_exception_ptr(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_ = std::make_exception_ptr(CancelException(XBT_THROW_POINT, "Execution Canceled"));
 199         break;
 200
 201       case State::TIMEOUT:
 202         simcall->issuer_->exception_ = std::make_exception_ptr(TimeoutException(XBT_THROW_POINT, "Timeouted"));
 203         break;
 204
 205       default:
 206         xbt_assert(state_ == State::DONE, "Internal error in ExecImpl::finish(): unexpected synchro state %s",
 207                    to_c_str(state_));
 208     }
 209
 210     simcall->issuer_->waiting_synchro_ = nullptr;
 211     /* Fail the process if the host is down */
 212     if (simcall->issuer_->get_host()->is_on())
 213       simcall->issuer_->simcall_answer();
 214     else
 215       simcall->issuer_->context_->set_wannadie();
 216   }
 217 }
 218
 219 ActivityImpl* ExecImpl::migrate(s4u::Host* to)
 220 {
 221   if (not MC_is_active() && not MC_record_replay_is_active()) {
 222     resource::Action* old_action = this->surf_action_;
 223     resource::Action* new_action = to->pimpl_cpu->execution_start(old_action->get_cost());
 224     new_action->set_remains(old_action->get_remains());
 225     new_action->set_activity(this);
 226     new_action->set_sharing_penalty(old_action->get_sharing_penalty());
 227     new_action->set_user_bound(old_action->get_user_bound());
 228
 229     old_action->set_activity(nullptr);
 230     old_action->cancel();
 231     old_action->unref();
 232     this->surf_action_ = new_action;
 233   }
 234
 235   on_migration(*this, to);
 236   return this;
 237 }
 238
 239 void ExecImpl::wait_any_for(actor::ActorImpl* issuer, const std::vector<ExecImpl*>& execs, double timeout)
 240 {
 241   if (timeout < 0.0) {
 242     issuer->simcall_.timeout_cb_ = nullptr;
 243   } else {
 244     issuer->simcall_.timeout_cb_ = timer::Timer::set(SIMIX_get_clock() + timeout, [issuer, &execs]() {
 245       issuer->simcall_.timeout_cb_ = nullptr;
 246       for (auto* exec : execs)
 247         exec->unregister_simcall(&issuer->simcall_);
 248       // default result (-1) is set in actor::ExecutionWaitanySimcall
 249       issuer->simcall_answer();
 250     });
 251   }
 252
 253   for (auto* exec : execs) {
 254     /* associate this simcall to the the synchro */
 255     exec->simcalls_.push_back(&issuer->simcall_);
 256
 257     /* see if the synchro is already finished */
 258     if (exec->state_ != State::WAITING && exec->state_ != State::RUNNING) {
 259       exec->finish();
 260       break;
 261     }
 262   }
 263 }
 264
 265 /*************
 266  * Callbacks *
 267  *************/
 268 xbt::signal<void(ExecImpl const&, s4u::Host*)> ExecImpl::on_migration;
 269
 270 } // namespace activity
 271 } // namespace kernel
 272 } // namespace simgrid