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   ActivityImpl::set_hosts({host});
  35   return *this;
  36 }
  37
  38 ExecImpl& ExecImpl::set_hosts(const std::vector<s4u::Host*>& hosts)
  39 {
  40   ActivityImpl::set_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(get_host()->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 (get_hosts().size() == 1) {
  83       if (thread_count_ == 1) {
  84         surf_action_ = get_host()->get_cpu()->execution_start(flops_amounts_.front(), bound_);
  85         surf_action_->set_sharing_penalty(sharing_penalty_);
  86       } else {
  87         auto host_model = get_host()->get_netpoint()->get_englobing_zone()->get_host_model();
  88         surf_action_    = host_model->execute_thread(get_host(), 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 = get_host()->get_netpoint()->get_englobing_zone()->get_host_model();
  94       surf_action_    = host_model->execute_parallel(get_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   auto const& hosts = get_hosts();
 149   if (std::any_of(hosts.begin(), hosts.end(), [](const s4u::Host* host) { return not host->is_on(); })) {
 150     /* If one of the hosts running the synchro failed, notice it. This way, the asking
 151      * process can be killed if it runs on that host itself */
 152     set_state(State::FAILED);
 153   } else if (surf_action_->get_state() == resource::Action::State::FAILED) {
 154     /* If all the hosts are running the synchro didn't fail, then the synchro was canceled */
 155     set_state(State::CANCELED);
 156   } else if (timeout_detector_ && timeout_detector_->get_state() == resource::Action::State::FINISHED) {
 157     if (surf_action_->get_remains() > 0.0) {
 158       surf_action_->set_state(resource::Action::State::FAILED);
 159       set_state(State::TIMEOUT);
 160     } else {
 161       set_state(State::DONE);
 162     }
 163   } else {
 164     set_state(State::DONE);
 165   }
 166
 167   clean_action();
 168   timeout_detector_.reset();
 169   if (get_actor() != nullptr) {
 170     get_actor()->activities_.remove(this);
 171   }
 172   if (get_state() != State::FAILED && cb_id_ >= 0)
 173     s4u::Host::on_state_change.disconnect(cb_id_);
 174   /* Answer all simcalls associated with the synchro */
 175   finish();
 176 }
 177
 178 void ExecImpl::set_exception(actor::ActorImpl* issuer)
 179 {
 180   switch (get_state()) {
 181     case State::FAILED:
 182       static_cast<s4u::Exec*>(get_iface())->complete(s4u::Activity::State::FAILED);
 183       if (issuer->get_host()->is_on())
 184         issuer->exception_ = std::make_exception_ptr(HostFailureException(XBT_THROW_POINT, "Host failed"));
 185       else /* else, the actor will be killed with no possibility to survive */
 186         issuer->set_wannadie();
 187       break;
 188
 189     case State::CANCELED:
 190       issuer->exception_ = std::make_exception_ptr(CancelException(XBT_THROW_POINT, "Execution Canceled"));
 191       break;
 192
 193     case State::TIMEOUT:
 194       issuer->exception_ = std::make_exception_ptr(TimeoutException(XBT_THROW_POINT, "Timeouted"));
 195       break;
 196
 197     default:
 198       xbt_assert(get_state() == State::DONE, "Internal error in ExecImpl::finish(): unexpected synchro state %s",
 199                  get_state_str());
 200   }
 201 }
 202 void ExecImpl::finish()
 203 {
 204   XBT_DEBUG("ExecImpl::finish() in state %s", get_state_str());
 205   while (not simcalls_.empty()) {
 206     actor::Simcall* simcall = simcalls_.front();
 207     simcalls_.pop_front();
 208
 209     if (simcall->call_ == actor::Simcall::Type::NONE) // FIXME: maybe a better way to handle this case
 210       continue;                                       // if process handling comm is killed
 211
 212     handle_activity_waitany(simcall);
 213
 214     set_exception(simcall->issuer_);
 215
 216     simcall->issuer_->waiting_synchro_ = nullptr;
 217     /* Fail the process if the host is down */
 218     if (simcall->issuer_->get_host()->is_on())
 219       simcall->issuer_->simcall_answer();
 220     else
 221       simcall->issuer_->set_wannadie();
 222   }
 223 }
 224
 225 void ExecImpl::reset()
 226 {
 227   clear_hosts();
 228   bytes_amounts_.clear();
 229   flops_amounts_.clear();
 230   set_start_time(-1.0);
 231 }
 232
 233 ActivityImpl* ExecImpl::migrate(s4u::Host* to)
 234 {
 235   if (not MC_is_active() && not MC_record_replay_is_active()) {
 236     resource::Action* old_action = this->surf_action_;
 237     resource::Action* new_action = to->get_cpu()->execution_start(old_action->get_cost(), old_action->get_user_bound());
 238     new_action->set_remains(old_action->get_remains());
 239     new_action->set_activity(this);
 240     new_action->set_sharing_penalty(old_action->get_sharing_penalty());
 241     new_action->set_user_bound(old_action->get_user_bound());
 242
 243     old_action->set_activity(nullptr);
 244     old_action->cancel();
 245     old_action->unref();
 246     this->surf_action_ = new_action;
 247   }
 248
 249   on_migration(*this, to);
 250   return this;
 251 }
 252
 253 /*************
 254  * Callbacks *
 255  *************/
 256 xbt::signal<void(ExecImpl const&, s4u::Host*)> ExecImpl::on_migration;
 257
 258 } // namespace simgrid::kernel::activity