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::kernel::activity {
  21
  22 ExecImpl::ExecImpl()
  23 {
  24   piface_                = new s4u::Exec(this);
  25   actor::ActorImpl* self = actor::ActorImpl::self();
  26   if (self) {
  27     set_actor(self);
  28     self->activities_.emplace_back(this);
  29   }
  30 }
  31
  32 ExecImpl& ExecImpl::set_host(s4u::Host* host)
  33 {
  34   hosts_.assign(1, host);
  35   return *this;
  36 }
  37
  38 ExecImpl& ExecImpl::set_hosts(const std::vector<s4u::Host*>& hosts)
  39 {
  40   hosts_ = hosts;
  41   return *this;
  42 }
  43
  44 ExecImpl& ExecImpl::set_timeout(double timeout)
  45 {
  46   if (timeout >= 0 && not MC_is_active() && not MC_record_replay_is_active()) {
  47     timeout_detector_.reset(hosts_.front()->get_cpu()->sleep(timeout));
  48     timeout_detector_->set_activity(this);
  49   }
  50   return *this;
  51 }
  52
  53 ExecImpl& ExecImpl::set_flops_amount(double flops_amount)
  54 {
  55   flops_amounts_.assign(1, flops_amount);
  56   return *this;
  57 }
  58
  59 ExecImpl& ExecImpl::set_flops_amounts(const std::vector<double>& flops_amounts)
  60 {
  61   flops_amounts_ = flops_amounts;
  62   return *this;
  63 }
  64
  65 ExecImpl& ExecImpl::set_bytes_amounts(const std::vector<double>& bytes_amounts)
  66 {
  67   bytes_amounts_ = bytes_amounts;
  68
  69   return *this;
  70 }
  71 ExecImpl& ExecImpl::set_thread_count(int thread_count)
  72 {
  73   thread_count_ = thread_count;
  74
  75   return *this;
  76 }
  77
  78 ExecImpl* ExecImpl::start()
  79 {
  80   set_state(State::RUNNING);
  81   if (not MC_is_active() && not MC_record_replay_is_active()) {
  82     if (hosts_.size() == 1) {
  83       if (thread_count_ == 1) {
  84         surf_action_ = hosts_.front()->get_cpu()->execution_start(flops_amounts_.front(), bound_);
  85         surf_action_->set_sharing_penalty(sharing_penalty_);
  86       } else {
  87         auto host_model = hosts_.front()->get_netpoint()->get_englobing_zone()->get_host_model();
  88         surf_action_    = host_model->execute_thread(hosts_.front(), flops_amounts_.front(), thread_count_);
  89       }
  90       surf_action_->set_category(get_tracing_category());
  91     } else {
  92       // get the model from first host since we have only 1 by now
  93       auto host_model = hosts_.front()->get_netpoint()->get_englobing_zone()->get_host_model();
  94       surf_action_    = host_model->execute_parallel(hosts_, flops_amounts_.data(), bytes_amounts_.data(), -1);
  95     }
  96     surf_action_->set_activity(this);
  97     set_start_time(surf_action_->get_start_time());
  98   }
  99
 100   XBT_DEBUG("Create execute synchro %p: %s", this, get_cname());
 101   return this;
 102 }
 103
 104 double ExecImpl::get_remaining() const
 105 {
 106   if (get_state() == State::WAITING || get_state() == State::FAILED)
 107     return flops_amounts_.front();
 108   return ActivityImpl::get_remaining();
 109 }
 110
 111 double ExecImpl::get_seq_remaining_ratio()
 112 {
 113   if (get_state() == State::WAITING)
 114     return 1;
 115   return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains() / surf_action_->get_cost();
 116 }
 117
 118 double ExecImpl::get_par_remaining_ratio()
 119 {
 120   // parallel task: their remain is already between 0 and 1
 121   if (get_state() == State::WAITING)
 122     return 1;
 123   return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains();
 124 }
 125
 126 ExecImpl& ExecImpl::set_bound(double bound)
 127 {
 128   bound_ = bound;
 129   return *this;
 130 }
 131
 132 ExecImpl& ExecImpl::set_sharing_penalty(double sharing_penalty)
 133 {
 134   sharing_penalty_ = sharing_penalty;
 135   return *this;
 136 }
 137
 138 ExecImpl& ExecImpl::update_sharing_penalty(double sharing_penalty)
 139 {
 140   sharing_penalty_ = sharing_penalty;
 141   surf_action_->set_sharing_penalty(sharing_penalty);
 142   return *this;
 143 }
 144
 145 void ExecImpl::post()
 146 {
 147   xbt_assert(surf_action_ != nullptr);
 148   if (std::any_of(hosts_.begin(), hosts_.end(), [](const s4u::Host* host) { return not host->is_on(); })) {
 149     /* If one of the hosts running the synchro failed, notice it. This way, the asking
 150      * process can be killed if it runs on that host itself */
 151     set_state(State::FAILED);
 152   } else if (surf_action_->get_state() == resource::Action::State::FAILED) {
 153     /* If all the hosts are running the synchro didn't fail, then the synchro was canceled */
 154     set_state(State::CANCELED);
 155   } else if (timeout_detector_ && timeout_detector_->get_state() == resource::Action::State::FINISHED) {
 156     if (surf_action_->get_remains() > 0.0) {
 157       surf_action_->set_state(resource::Action::State::FAILED);
 158       set_state(State::TIMEOUT);
 159     } else {
 160       set_state(State::DONE);
 161     }
 162   } else {
 163     set_state(State::DONE);
 164   }
 165
 166   clean_action();
 167   timeout_detector_.reset();
 168   if (get_actor() != nullptr) {
 169     get_actor()->activities_.remove(this);
 170   }
 171   if (get_state() != State::FAILED && cb_id_ >= 0)
 172     s4u::Host::on_state_change.disconnect(cb_id_);
 173   /* Answer all simcalls associated with the synchro */
 174   finish();
 175 }
 176
 177 void ExecImpl::set_exception(actor::ActorImpl* issuer)
 178 {
 179   switch (get_state()) {
 180     case State::FAILED:
 181       static_cast<s4u::Exec*>(get_iface())->complete(s4u::Activity::State::FAILED);
 182       if (issuer->get_host()->is_on())
 183         issuer->exception_ = std::make_exception_ptr(HostFailureException(XBT_THROW_POINT, "Host failed"));
 184       else /* else, the actor will be killed with no possibility to survive */
 185         issuer->set_wannadie();
 186       break;
 187
 188     case State::CANCELED:
 189       issuer->exception_ = std::make_exception_ptr(CancelException(XBT_THROW_POINT, "Execution Canceled"));
 190       break;
 191
 192     case State::TIMEOUT:
 193       issuer->exception_ = std::make_exception_ptr(TimeoutException(XBT_THROW_POINT, "Timeouted"));
 194       break;
 195
 196     default:
 197       xbt_assert(get_state() == State::DONE, "Internal error in ExecImpl::finish(): unexpected synchro state %s",
 198                  get_state_str());
 199   }
 200 }
 201 void ExecImpl::finish()
 202 {
 203   XBT_DEBUG("ExecImpl::finish() in state %s", get_state_str());
 204   while (not simcalls_.empty()) {
 205     actor::Simcall* simcall = simcalls_.front();
 206     simcalls_.pop_front();
 207
 208     if (simcall->call_ == actor::Simcall::Type::NONE) // FIXME: maybe a better way to handle this case
 209       continue;                                       // if process handling comm is killed
 210
 211     handle_activity_waitany(simcall);
 212
 213     set_exception(simcall->issuer_);
 214
 215     simcall->issuer_->waiting_synchro_ = nullptr;
 216     /* Fail the process if the host is down */
 217     if (simcall->issuer_->get_host()->is_on())
 218       simcall->issuer_->simcall_answer();
 219     else
 220       simcall->issuer_->set_wannadie();
 221   }
 222 }
 223
 224 void ExecImpl::reset()
 225 {
 226   hosts_.clear();
 227   bytes_amounts_.clear();
 228   flops_amounts_.clear();
 229   set_start_time(-1.0);
 230 }
 231
 232 ActivityImpl* ExecImpl::migrate(s4u::Host* to)
 233 {
 234   if (not MC_is_active() && not MC_record_replay_is_active()) {
 235     resource::Action* old_action = this->surf_action_;
 236     resource::Action* new_action = to->get_cpu()->execution_start(old_action->get_cost(), old_action->get_user_bound());
 237     new_action->set_remains(old_action->get_remains());
 238     new_action->set_activity(this);
 239     new_action->set_sharing_penalty(old_action->get_sharing_penalty());
 240     new_action->set_user_bound(old_action->get_user_bound());
 241
 242     old_action->set_activity(nullptr);
 243     old_action->cancel();
 244     old_action->unref();
 245     this->surf_action_ = new_action;
 246   }
 247
 248   on_migration(*this, to);
 249   return this;
 250 }
 251
 252 /*************
 253  * Callbacks *
 254  *************/
 255 xbt::signal<void(ExecImpl const&, s4u::Host*)> ExecImpl::on_migration;
 256
 257 } // namespace simgrid::kernel::activity