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 <simgrid/Exception.hpp>
   7 #include <simgrid/kernel/routing/NetPoint.hpp>
   8 #include <simgrid/modelchecker.h>
   9 #include <simgrid/s4u/Engine.hpp>
  10
  11 #include "src/kernel/activity/ExecImpl.hpp"
  12 #include "src/kernel/actor/ActorImpl.hpp"
  13 #include "src/kernel/actor/SimcallObserver.hpp"
  14 #include "src/mc/mc_replay.hpp"
  15 #include "src/surf/HostImpl.hpp"
  16 #include "src/surf/cpu_interface.hpp"
  17
  18 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(simix_process);
  19
  20 namespace simgrid {
  21 namespace kernel {
  22 namespace activity {
  23
  24 ExecImpl::ExecImpl()
  25 {
  26   piface_                = new s4u::Exec(this);
  27   actor::ActorImpl* self = actor::ActorImpl::self();
  28   if (self) {
  29     actor_ = self;
  30     self->activities_.emplace_back(this);
  31   }
  32 }
  33
  34 ExecImpl& ExecImpl::set_host(s4u::Host* host)
  35 {
  36   hosts_.assign(1, host);
  37   return *this;
  38 }
  39
  40 ExecImpl& ExecImpl::set_hosts(const std::vector<s4u::Host*>& hosts)
  41 {
  42   hosts_ = hosts;
  43   return *this;
  44 }
  45
  46 ExecImpl& ExecImpl::set_timeout(double timeout)
  47 {
  48   if (timeout >= 0 && not MC_is_active() && not MC_record_replay_is_active()) {
  49     timeout_detector_.reset(hosts_.front()->get_cpu()->sleep(timeout));
  50     timeout_detector_->set_activity(this);
  51   }
  52   return *this;
  53 }
  54
  55 ExecImpl& ExecImpl::set_flops_amount(double flops_amount)
  56 {
  57   flops_amounts_.assign(1, flops_amount);
  58   return *this;
  59 }
  60
  61 ExecImpl& ExecImpl::set_flops_amounts(const std::vector<double>& flops_amounts)
  62 {
  63   flops_amounts_ = flops_amounts;
  64   return *this;
  65 }
  66
  67 ExecImpl& ExecImpl::set_bytes_amounts(const std::vector<double>& bytes_amounts)
  68 {
  69   bytes_amounts_ = bytes_amounts;
  70
  71   return *this;
  72 }
  73
  74 ExecImpl* ExecImpl::start()
  75 {
  76   state_ = State::RUNNING;
  77   if (not MC_is_active() && not MC_record_replay_is_active()) {
  78     if (hosts_.size() == 1) {
  79       surf_action_ = hosts_.front()->get_cpu()->execution_start(flops_amounts_.front(), bound_);
  80       surf_action_->set_sharing_penalty(sharing_penalty_);
  81       surf_action_->set_category(get_tracing_category());
  82     } else {
  83       // get the model from first host since we have only 1 by now
  84       auto host_model = hosts_.front()->get_netpoint()->get_englobing_zone()->get_host_model();
  85       surf_action_    = host_model->execute_parallel(hosts_, flops_amounts_.data(), bytes_amounts_.data(), -1);
  86     }
  87     surf_action_->set_activity(this);
  88     start_time_ = surf_action_->get_start_time();
  89   }
  90
  91   XBT_DEBUG("Create execute synchro %p: %s", this, get_cname());
  92   return this;
  93 }
  94
  95 double ExecImpl::get_seq_remaining_ratio()
  96 {
  97   return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains() / surf_action_->get_cost();
  98 }
  99
 100 double ExecImpl::get_par_remaining_ratio()
 101 {
 102   // parallel task: their remain is already between 0 and 1
 103   return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains();
 104 }
 105
 106 ExecImpl& ExecImpl::set_bound(double bound)
 107 {
 108   bound_ = bound;
 109   return *this;
 110 }
 111
 112 ExecImpl& ExecImpl::set_sharing_penalty(double sharing_penalty)
 113 {
 114   sharing_penalty_ = sharing_penalty;
 115   return *this;
 116 }
 117
 118 void ExecImpl::post()
 119 {
 120   xbt_assert(surf_action_ != nullptr);
 121   if (std::any_of(hosts_.begin(), hosts_.end(), [](const s4u::Host* host) { return not host->is_on(); })) {
 122     /* If one of the hosts running the synchro failed, notice it. This way, the asking
 123      * process can be killed if it runs on that host itself */
 124     state_ = State::FAILED;
 125   } else if (surf_action_->get_state() == resource::Action::State::FAILED) {
 126     /* If all the hosts are running the synchro didn't fail, then the synchro was canceled */
 127     state_ = State::CANCELED;
 128   } else if (timeout_detector_ && timeout_detector_->get_state() == resource::Action::State::FINISHED) {
 129     if (surf_action_->get_remains() > 0.0) {
 130       surf_action_->set_state(resource::Action::State::FAILED);
 131       state_ = State::TIMEOUT;
 132     } else {
 133       state_ = State::DONE;
 134     }
 135   } else {
 136     state_ = State::DONE;
 137   }
 138
 139   finish_time_ = surf_action_->get_finish_time();
 140
 141   clean_action();
 142   timeout_detector_.reset();
 143   if (actor_) {
 144     actor_->activities_.remove(this);
 145     actor_ = nullptr;
 146   }
 147   if (state_ != State::FAILED && cb_id_ >= 0)
 148     s4u::Host::on_state_change.disconnect(cb_id_);
 149   /* Answer all simcalls associated with the synchro */
 150   finish();
 151 }
 152
 153 void ExecImpl::finish()
 154 {
 155   XBT_DEBUG("ExecImpl::finish() in state %s", to_c_str(state_));
 156   while (not simcalls_.empty()) {
 157     smx_simcall_t simcall = simcalls_.front();
 158     simcalls_.pop_front();
 159
 160     /* If a waitany simcall is waiting for this synchro to finish, then remove it from the other synchros in the waitany
 161      * list. Afterwards, get the position of the actual synchro in the waitany list and return it as the result of the
 162      * simcall */
 163
 164     if (simcall->call_ == simix::Simcall::NONE) // FIXME: maybe a better way to handle this case
 165       continue;                                 // if process handling comm is killed
 166     if (auto* observer =
 167             dynamic_cast<kernel::actor::ExecutionWaitanySimcall*>(simcall->observer_)) { // simcall is a wait_any?
 168       const auto& execs = observer->get_execs();
 169
 170       for (auto* exec : execs) {
 171         exec->unregister_simcall(simcall);
 172
 173         if (simcall->timeout_cb_) {
 174           simcall->timeout_cb_->remove();
 175           simcall->timeout_cb_ = nullptr;
 176         }
 177       }
 178
 179       if (not MC_is_active() && not MC_record_replay_is_active()) {
 180         auto element = std::find(execs.begin(), execs.end(), this);
 181         int rank     = element != execs.end() ? static_cast<int>(std::distance(execs.begin(), element)) : -1;
 182         observer->set_result(rank);
 183       }
 184     }
 185     switch (state_) {
 186       case State::FAILED:
 187         piface_->complete(s4u::Activity::State::FAILED);
 188         if (simcall->issuer_->get_host()->is_on())
 189           simcall->issuer_->exception_ = std::make_exception_ptr(HostFailureException(XBT_THROW_POINT, "Host failed"));
 190         else /* else, the actor will be killed with no possibility to survive */
 191           simcall->issuer_->context_->set_wannadie();
 192         break;
 193
 194       case State::CANCELED:
 195         simcall->issuer_->exception_ = std::make_exception_ptr(CancelException(XBT_THROW_POINT, "Execution Canceled"));
 196         break;
 197
 198       case State::TIMEOUT:
 199         simcall->issuer_->exception_ = std::make_exception_ptr(TimeoutException(XBT_THROW_POINT, "Timeouted"));
 200         break;
 201
 202       default:
 203         xbt_assert(state_ == State::DONE, "Internal error in ExecImpl::finish(): unexpected synchro state %s",
 204                    to_c_str(state_));
 205     }
 206
 207     simcall->issuer_->waiting_synchro_ = nullptr;
 208     /* Fail the process if the host is down */
 209     if (simcall->issuer_->get_host()->is_on())
 210       simcall->issuer_->simcall_answer();
 211     else
 212       simcall->issuer_->context_->set_wannadie();
 213   }
 214 }
 215
 216 ActivityImpl* ExecImpl::migrate(s4u::Host* to)
 217 {
 218   if (not MC_is_active() && not MC_record_replay_is_active()) {
 219     resource::Action* old_action = this->surf_action_;
 220     resource::Action* new_action = to->get_cpu()->execution_start(old_action->get_cost(), old_action->get_user_bound());
 221     new_action->set_remains(old_action->get_remains());
 222     new_action->set_activity(this);
 223     new_action->set_sharing_penalty(old_action->get_sharing_penalty());
 224     new_action->set_user_bound(old_action->get_user_bound());
 225
 226     old_action->set_activity(nullptr);
 227     old_action->cancel();
 228     old_action->unref();
 229     this->surf_action_ = new_action;
 230   }
 231
 232   on_migration(*this, to);
 233   return this;
 234 }
 235
 236 void ExecImpl::wait_any_for(actor::ActorImpl* issuer, const std::vector<ExecImpl*>& execs, double timeout)
 237 {
 238   if (timeout < 0.0) {
 239     issuer->simcall_.timeout_cb_ = nullptr;
 240   } else {
 241     issuer->simcall_.timeout_cb_ = timer::Timer::set(s4u::Engine::get_clock() + timeout, [issuer, &execs]() {
 242       issuer->simcall_.timeout_cb_ = nullptr;
 243       for (auto* exec : execs)
 244         exec->unregister_simcall(&issuer->simcall_);
 245       // default result (-1) is set in actor::ExecutionWaitanySimcall
 246       issuer->simcall_answer();
 247     });
 248   }
 249
 250   for (auto* exec : execs) {
 251     /* associate this simcall to the the synchro */
 252     exec->simcalls_.push_back(&issuer->simcall_);
 253     /* see if the synchro is already finished */
 254     if (exec->state_ != State::WAITING && exec->state_ != State::RUNNING) {
 255       exec->finish();
 256       break;
 257     }
 258   }
 259 }
 260
 261 /*************
 262  * Callbacks *
 263  *************/
 264 xbt::signal<void(ExecImpl const&, s4u::Host*)> ExecImpl::on_migration;
 265
 266 } // namespace activity
 267 } // namespace kernel
 268 } // namespace simgrid