src/kernel/activity/ExecImpl.cpp

   1 /* Copyright (c) 2007-2022. 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/kernel/resource/CpuImpl.hpp"
  15 #include "src/mc/mc_replay.hpp"
  16 #include "src/surf/HostImpl.hpp"
  17
  18 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(ker_cpu, kernel, "Kernel cpu-related synchronization");
  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     set_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 ExecImpl& ExecImpl::set_thread_count(int thread_count)
  74 {
  75   thread_count_ = thread_count;
  76
  77   return *this;
  78 }
  79
  80 ExecImpl* ExecImpl::start()
  81 {
  82   set_state(State::RUNNING);
  83   if (not MC_is_active() && not MC_record_replay_is_active()) {
  84     if (hosts_.size() == 1) {
  85       if (thread_count_ == 1) {
  86         surf_action_ = hosts_.front()->get_cpu()->execution_start(flops_amounts_.front(), bound_);
  87         surf_action_->set_sharing_penalty(sharing_penalty_);
  88       } else {
  89         auto host_model = hosts_.front()->get_netpoint()->get_englobing_zone()->get_host_model();
  90         surf_action_    = host_model->execute_thread(hosts_.front(), flops_amounts_.front(), thread_count_);
  91       }
  92       surf_action_->set_category(get_tracing_category());
  93     } else {
  94       // get the model from first host since we have only 1 by now
  95       auto host_model = hosts_.front()->get_netpoint()->get_englobing_zone()->get_host_model();
  96       surf_action_    = host_model->execute_parallel(hosts_, flops_amounts_.data(), bytes_amounts_.data(), -1);
  97     }
  98     surf_action_->set_activity(this);
  99     set_start_time(surf_action_->get_start_time());
 100   }
 101
 102   XBT_DEBUG("Create execute synchro %p: %s", this, get_cname());
 103   return this;
 104 }
 105
 106 double ExecImpl::get_remaining() const
 107 {
 108   if (get_state() == State::WAITING || get_state() == State::FAILED)
 109     return flops_amounts_.front();
 110   return ActivityImpl::get_remaining();
 111 }
 112
 113 double ExecImpl::get_seq_remaining_ratio()
 114 {
 115   if (get_state() == State::WAITING)
 116     return 1;
 117   return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains() / surf_action_->get_cost();
 118 }
 119
 120 double ExecImpl::get_par_remaining_ratio()
 121 {
 122   // parallel task: their remain is already between 0 and 1
 123   if (get_state() == State::WAITING)
 124     return 1;
 125   return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains();
 126 }
 127
 128 ExecImpl& ExecImpl::set_bound(double bound)
 129 {
 130   bound_ = bound;
 131   return *this;
 132 }
 133
 134 ExecImpl& ExecImpl::set_sharing_penalty(double sharing_penalty)
 135 {
 136   sharing_penalty_ = sharing_penalty;
 137   return *this;
 138 }
 139
 140 ExecImpl& ExecImpl::update_sharing_penalty(double sharing_penalty)
 141 {
 142   sharing_penalty_ = sharing_penalty;
 143   surf_action_->set_sharing_penalty(sharing_penalty);
 144   return *this;
 145 }
 146
 147 void ExecImpl::post()
 148 {
 149   xbt_assert(surf_action_ != nullptr);
 150   if (std::any_of(hosts_.begin(), hosts_.end(), [](const s4u::Host* host) { return not host->is_on(); })) {
 151     /* If one of the hosts running the synchro failed, notice it. This way, the asking
 152      * process can be killed if it runs on that host itself */
 153     set_state(State::FAILED);
 154   } else if (surf_action_->get_state() == resource::Action::State::FAILED) {
 155     /* If all the hosts are running the synchro didn't fail, then the synchro was canceled */
 156     set_state(State::CANCELED);
 157   } else if (timeout_detector_ && timeout_detector_->get_state() == resource::Action::State::FINISHED) {
 158     if (surf_action_->get_remains() > 0.0) {
 159       surf_action_->set_state(resource::Action::State::FAILED);
 160       set_state(State::TIMEOUT);
 161     } else {
 162       set_state(State::DONE);
 163     }
 164   } else {
 165     set_state(State::DONE);
 166   }
 167
 168   clean_action();
 169   timeout_detector_.reset();
 170   if (get_actor() != nullptr) {
 171     get_actor()->activities_.remove(this);
 172   }
 173   if (get_state() != State::FAILED && cb_id_ >= 0)
 174     s4u::Host::on_state_change.disconnect(cb_id_);
 175   /* Answer all simcalls associated with the synchro */
 176   finish();
 177 }
 178
 179 void ExecImpl::set_exception(actor::ActorImpl* issuer)
 180 {
 181   switch (get_state()) {
 182     case State::FAILED:
 183       static_cast<s4u::Exec*>(get_iface())->complete(s4u::Activity::State::FAILED);
 184       if (issuer->get_host()->is_on())
 185         issuer->exception_ = std::make_exception_ptr(HostFailureException(XBT_THROW_POINT, "Host failed"));
 186       else /* else, the actor will be killed with no possibility to survive */
 187         issuer->set_wannadie();
 188       break;
 189
 190     case State::CANCELED:
 191       issuer->exception_ = std::make_exception_ptr(CancelException(XBT_THROW_POINT, "Execution Canceled"));
 192       break;
 193
 194     case State::TIMEOUT:
 195       issuer->exception_ = std::make_exception_ptr(TimeoutException(XBT_THROW_POINT, "Timeouted"));
 196       break;
 197
 198     default:
 199       xbt_assert(get_state() == State::DONE, "Internal error in ExecImpl::finish(): unexpected synchro state %s",
 200                  get_state_str());
 201   }
 202 }
 203 void ExecImpl::finish()
 204 {
 205   XBT_DEBUG("ExecImpl::finish() in state %s", get_state_str());
 206   while (not simcalls_.empty()) {
 207     actor::Simcall* simcall = simcalls_.front();
 208     simcalls_.pop_front();
 209
 210     if (simcall->call_ == actor::Simcall::Type::NONE) // FIXME: maybe a better way to handle this case
 211       continue;                                       // if process handling comm is killed
 212
 213     handle_activity_waitany(simcall);
 214
 215     set_exception(simcall->issuer_);
 216
 217     simcall->issuer_->waiting_synchro_ = nullptr;
 218     /* Fail the process if the host is down */
 219     if (simcall->issuer_->get_host()->is_on())
 220       simcall->issuer_->simcall_answer();
 221     else
 222       simcall->issuer_->set_wannadie();
 223   }
 224 }
 225
 226 void ExecImpl::reset()
 227 {
 228   hosts_.clear();
 229   bytes_amounts_.clear();
 230   flops_amounts_.clear();
 231   set_start_time(-1.0);
 232 }
 233
 234 ActivityImpl* ExecImpl::migrate(s4u::Host* to)
 235 {
 236   if (not MC_is_active() && not MC_record_replay_is_active()) {
 237     resource::Action* old_action = this->surf_action_;
 238     resource::Action* new_action = to->get_cpu()->execution_start(old_action->get_cost(), old_action->get_user_bound());
 239     new_action->set_remains(old_action->get_remains());
 240     new_action->set_activity(this);
 241     new_action->set_sharing_penalty(old_action->get_sharing_penalty());
 242     new_action->set_user_bound(old_action->get_user_bound());
 243
 244     old_action->set_activity(nullptr);
 245     old_action->cancel();
 246     old_action->unref();
 247     this->surf_action_ = new_action;
 248   }
 249
 250   on_migration(*this, to);
 251   return this;
 252 }
 253
 254 /*************
 255  * Callbacks *
 256  *************/
 257 xbt::signal<void(ExecImpl const&, s4u::Host*)> ExecImpl::on_migration;
 258
 259 } // namespace activity
 260 } // namespace kernel
 261 } // namespace simgrid