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(), bound_);
  82       surf_action_->set_sharing_penalty(sharing_penalty_);
  83       surf_action_->set_category(get_tracing_category());
  84     } else {
  85       // get the model from first host since we have only 1 by now
  86       auto host_model = hosts_.front()->get_netpoint()->get_englobing_zone()->get_host_model();
  87       surf_action_    = host_model->execute_parallel(hosts_, flops_amounts_.data(), bytes_amounts_.data(), -1);
  88     }
  89     surf_action_->set_activity(this);
  90     start_time_ = surf_action_->get_start_time();
  91   }
  92
  93   XBT_DEBUG("Create execute synchro %p: %s", this, get_cname());
  94   return this;
  95 }
  96
  97 double ExecImpl::get_seq_remaining_ratio()
  98 {
  99   return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains() / surf_action_->get_cost();
 100 }
 101
 102 double ExecImpl::get_par_remaining_ratio()
 103 {
 104   // parallel task: their remain is already between 0 and 1
 105   return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains();
 106 }
 107
 108 ExecImpl& ExecImpl::set_bound(double bound)
 109 {
 110   bound_ = bound;
 111   return *this;
 112 }
 113
 114 ExecImpl& ExecImpl::set_sharing_penalty(double sharing_penalty)
 115 {
 116   sharing_penalty_ = sharing_penalty;
 117   return *this;
 118 }
 119
 120 void ExecImpl::post()
 121 {
 122   xbt_assert(surf_action_ != nullptr);
 123   if (std::any_of(hosts_.begin(), hosts_.end(), [](const s4u::Host* host) { return not host->is_on(); })) {
 124     /* If one of the hosts running the synchro failed, notice it. This way, the asking
 125      * process can be killed if it runs on that host itself */
 126     state_ = State::FAILED;
 127   } else if (surf_action_->get_state() == resource::Action::State::FAILED) {
 128     /* If all the hosts are running the synchro didn't fail, then the synchro was canceled */
 129     state_ = State::CANCELED;
 130   } else if (timeout_detector_ && timeout_detector_->get_state() == resource::Action::State::FINISHED) {
 131     if (surf_action_->get_remains() > 0.0) {
 132       surf_action_->set_state(resource::Action::State::FAILED);
 133       state_ = State::TIMEOUT;
 134     } else {
 135       state_ = State::DONE;
 136     }
 137   } else {
 138     state_ = State::DONE;
 139   }
 140
 141   finish_time_ = surf_action_->get_finish_time();
 142
 143   clean_action();
 144   timeout_detector_.reset();
 145   if (actor_) {
 146     actor_->activities_.remove(this);
 147     actor_ = nullptr;
 148   }
 149   if (state_ != State::FAILED && cb_id_ >= 0)
 150     s4u::Host::on_state_change.disconnect(cb_id_);
 151   /* Answer all simcalls associated with the synchro */
 152   finish();
 153 }
 154
 155 void ExecImpl::finish()
 156 {
 157   XBT_DEBUG("ExecImpl::finish() in state %s", to_c_str(state_));
 158   while (not simcalls_.empty()) {
 159     smx_simcall_t simcall = simcalls_.front();
 160     simcalls_.pop_front();
 161
 162     /* If a waitany simcall is waiting for this synchro to finish, then remove it from the other synchros in the waitany
 163      * list. Afterwards, get the position of the actual synchro in the waitany list and return it as the result of the
 164      * simcall */
 165
 166     if (simcall->call_ == simix::Simcall::NONE) // FIXME: maybe a better way to handle this case
 167       continue;                                 // if process handling comm is killed
 168     if (auto* observer =
 169             dynamic_cast<kernel::actor::ExecutionWaitanySimcall*>(simcall->observer_)) { // simcall is a wait_any?
 170       const auto& execs = observer->get_execs();
 171
 172       for (auto* exec : execs) {
 173         exec->unregister_simcall(simcall);
 174
 175         if (simcall->timeout_cb_) {
 176           simcall->timeout_cb_->remove();
 177           simcall->timeout_cb_ = nullptr;
 178         }
 179       }
 180
 181       if (not MC_is_active() && not MC_record_replay_is_active()) {
 182         auto element = std::find(execs.begin(), execs.end(), this);
 183         int rank     = element != execs.end() ? static_cast<int>(std::distance(execs.begin(), element)) : -1;
 184         observer->set_result(rank);
 185       }
 186     }
 187     switch (state_) {
 188       case State::FAILED:
 189         piface_->complete(s4u::Activity::State::FAILED);
 190         if (simcall->issuer_->get_host()->is_on())
 191           simcall->issuer_->exception_ = std::make_exception_ptr(HostFailureException(XBT_THROW_POINT, "Host failed"));
 192         else /* else, the actor will be killed with no possibility to survive */
 193           simcall->issuer_->context_->set_wannadie();
 194         break;
 195
 196       case State::CANCELED:
 197         simcall->issuer_->exception_ = std::make_exception_ptr(CancelException(XBT_THROW_POINT, "Execution Canceled"));
 198         break;
 199
 200       case State::TIMEOUT:
 201         simcall->issuer_->exception_ = std::make_exception_ptr(TimeoutException(XBT_THROW_POINT, "Timeouted"));
 202         break;
 203
 204       default:
 205         xbt_assert(state_ == State::DONE, "Internal error in ExecImpl::finish(): unexpected synchro state %s",
 206                    to_c_str(state_));
 207     }
 208
 209     simcall->issuer_->waiting_synchro_ = nullptr;
 210     /* Fail the process if the host is down */
 211     if (simcall->issuer_->get_host()->is_on())
 212       simcall->issuer_->simcall_answer();
 213     else
 214       simcall->issuer_->context_->set_wannadie();
 215   }
 216 }
 217
 218 ActivityImpl* ExecImpl::migrate(s4u::Host* to)
 219 {
 220   if (not MC_is_active() && not MC_record_replay_is_active()) {
 221     resource::Action* old_action = this->surf_action_;
 222     resource::Action* new_action = to->get_cpu()->execution_start(old_action->get_cost(), old_action->get_user_bound());
 223     new_action->set_remains(old_action->get_remains());
 224     new_action->set_activity(this);
 225     new_action->set_sharing_penalty(old_action->get_sharing_penalty());
 226     new_action->set_user_bound(old_action->get_user_bound());
 227
 228     old_action->set_activity(nullptr);
 229     old_action->cancel();
 230     old_action->unref();
 231     this->surf_action_ = new_action;
 232   }
 233
 234   on_migration(*this, to);
 235   return this;
 236 }
 237
 238 void ExecImpl::wait_any_for(actor::ActorImpl* issuer, const std::vector<ExecImpl*>& execs, double timeout)
 239 {
 240   if (timeout < 0.0) {
 241     issuer->simcall_.timeout_cb_ = nullptr;
 242   } else {
 243     issuer->simcall_.timeout_cb_ = timer::Timer::set(s4u::Engine::get_clock() + timeout, [issuer, &execs]() {
 244       issuer->simcall_.timeout_cb_ = nullptr;
 245       for (auto* exec : execs)
 246         exec->unregister_simcall(&issuer->simcall_);
 247       // default result (-1) is set in actor::ExecutionWaitanySimcall
 248       issuer->simcall_answer();
 249     });
 250   }
 251
 252   for (auto* exec : execs) {
 253     /* associate this simcall to the the synchro */
 254     exec->simcalls_.push_back(&issuer->simcall_);
 255     /* see if the synchro is already finished */
 256     if (exec->state_ != State::WAITING && exec->state_ != State::RUNNING) {
 257       exec->finish();
 258       break;
 259     }
 260   }
 261 }
 262
 263 /*************
 264  * Callbacks *
 265  *************/
 266 xbt::signal<void(ExecImpl const&, s4u::Host*)> ExecImpl::on_migration;
 267
 268 } // namespace activity
 269 } // namespace kernel
 270 } // namespace simgrid