src/kernel/activity/ExecImpl.cpp

   1 /* Copyright (c) 2007-2023. 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/kernel/resource/HostImpl.hpp"
  16 #include "src/mc/mc_replay.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_.insert(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_flops_amount(double flops_amount)
  45 {
  46   flops_amounts_.assign(1, flops_amount);
  47   return *this;
  48 }
  49
  50 ExecImpl& ExecImpl::set_flops_amounts(const std::vector<double>& flops_amounts)
  51 {
  52   flops_amounts_ = flops_amounts;
  53   return *this;
  54 }
  55
  56 ExecImpl& ExecImpl::set_bytes_amounts(const std::vector<double>& bytes_amounts)
  57 {
  58   bytes_amounts_ = bytes_amounts;
  59
  60   return *this;
  61 }
  62 ExecImpl& ExecImpl::set_thread_count(int thread_count)
  63 {
  64   thread_count_ = thread_count;
  65
  66   return *this;
  67 }
  68
  69 ExecImpl* ExecImpl::start()
  70 {
  71   set_state(State::RUNNING);
  72   if (not MC_is_active() && not MC_record_replay_is_active()) {
  73     if (get_hosts().size() == 1) {
  74       xbt_assert(not flops_amounts_.empty(), "Cannot start Exec: no flops_amount defined.");
  75       if (thread_count_ == 1) {
  76         model_action_ = get_host()->get_cpu()->execution_start(flops_amounts_.front(), bound_);
  77         model_action_->set_sharing_penalty(sharing_penalty_);
  78       } else {
  79         auto host_model = get_host()->get_netpoint()->get_englobing_zone()->get_host_model();
  80         model_action_   = host_model->execute_thread(get_host(), flops_amounts_.front(), thread_count_);
  81       }
  82       model_action_->set_category(get_tracing_category());
  83     } else {
  84       // get the model from first host since we have only 1 by now
  85       auto host_model = get_host()->get_netpoint()->get_englobing_zone()->get_host_model();
  86       model_action_   = host_model->execute_parallel(get_hosts(), flops_amounts_.data(), bytes_amounts_.data(), -1);
  87     }
  88     model_action_->set_activity(this);
  89     set_start_time(model_action_->get_start_time());
  90   }
  91
  92   XBT_DEBUG("Create execute synchro %p: %s", this, get_cname());
  93   return this;
  94 }
  95
  96 double ExecImpl::get_remaining() const
  97 {
  98   if (get_state() == State::WAITING || get_state() == State::FAILED)
  99     return flops_amounts_.front();
 100   return ActivityImpl::get_remaining();
 101 }
 102
 103 double ExecImpl::get_seq_remaining_ratio()
 104 {
 105   if (get_state() == State::WAITING)
 106     return 1;
 107   return (model_action_ == nullptr) ? 0 : model_action_->get_remains() / model_action_->get_cost();
 108 }
 109
 110 double ExecImpl::get_par_remaining_ratio()
 111 {
 112   // parallel task: their remain is already between 0 and 1
 113   if (get_state() == State::WAITING)
 114     return 1;
 115   return (model_action_ == nullptr) ? 0 : model_action_->get_remains();
 116 }
 117
 118 ExecImpl& ExecImpl::set_bound(double bound)
 119 {
 120   bound_ = bound;
 121   return *this;
 122 }
 123
 124 ExecImpl& ExecImpl::set_sharing_penalty(double sharing_penalty)
 125 {
 126   sharing_penalty_ = sharing_penalty;
 127   return *this;
 128 }
 129
 130 ExecImpl& ExecImpl::update_sharing_penalty(double sharing_penalty)
 131 {
 132   sharing_penalty_ = sharing_penalty;
 133   model_action_->set_sharing_penalty(sharing_penalty);
 134   return *this;
 135 }
 136
 137 void ExecImpl::set_exception(actor::ActorImpl* issuer)
 138 {
 139   switch (get_state()) {
 140     case State::FAILED:
 141       static_cast<s4u::Exec*>(get_iface())->complete(s4u::Activity::State::FAILED);
 142       if (issuer->get_host()->is_on())
 143         issuer->exception_ = std::make_exception_ptr(HostFailureException(XBT_THROW_POINT, "Host failed"));
 144       else /* else, the actor will be killed with no possibility to survive */
 145         issuer->set_wannadie();
 146       break;
 147
 148     case State::CANCELED:
 149       issuer->exception_ = std::make_exception_ptr(CancelException(XBT_THROW_POINT, "Execution Canceled"));
 150       break;
 151
 152     case State::TIMEOUT:
 153       issuer->exception_ = std::make_exception_ptr(TimeoutException(XBT_THROW_POINT, "Timeouted"));
 154       break;
 155
 156     default:
 157       xbt_assert(get_state() == State::DONE, "Internal error in ExecImpl::finish(): unexpected synchro state %s",
 158                  get_state_str());
 159   }
 160 }
 161 void ExecImpl::finish()
 162 {
 163   XBT_DEBUG("ExecImpl::finish() in state %s", get_state_str());
 164   if (model_action_ != nullptr) {
 165     if (auto const& hosts = get_hosts();
 166         std::any_of(hosts.begin(), hosts.end(), [](const s4u::Host* host) { return not host->is_on(); })) {
 167       /* If one of the hosts running the synchro failed, notice it. This way, the asking
 168        * process can be killed if it runs on that host itself */
 169       set_state(State::FAILED);
 170     } else if (model_action_->get_state() == resource::Action::State::FAILED) {
 171       /* If all the hosts are running the synchro didn't fail, then the synchro was canceled */
 172       set_state(State::CANCELED);
 173     } else {
 174       set_state(State::DONE);
 175     }
 176
 177     clean_action();
 178   }
 179
 180   if (get_actor() != nullptr)
 181     get_actor()->activities_.erase(this);
 182
 183   while (not simcalls_.empty()) {
 184     actor::Simcall* simcall = simcalls_.front();
 185     simcalls_.pop_front();
 186
 187     if (simcall->call_ == actor::Simcall::Type::NONE) // FIXME: maybe a better way to handle this case
 188       continue;                                       // if process handling comm is killed
 189
 190     handle_activity_waitany(simcall);
 191
 192     set_exception(simcall->issuer_);
 193
 194     simcall->issuer_->waiting_synchro_ = nullptr;
 195     /* Fail the process if the host is down */
 196     if (simcall->issuer_->get_host()->is_on())
 197       simcall->issuer_->simcall_answer();
 198     else
 199       simcall->issuer_->set_wannadie();
 200   }
 201 }
 202
 203 void ExecImpl::reset()
 204 {
 205   clear_hosts();
 206   set_start_time(-1.0);
 207 }
 208
 209 ActivityImpl* ExecImpl::migrate(s4u::Host* to)
 210 {
 211   if (not MC_is_active() && not MC_record_replay_is_active()) {
 212     resource::Action* old_action = this->model_action_;
 213     resource::Action* new_action = to->get_cpu()->execution_start(old_action->get_cost(), old_action->get_user_bound());
 214     new_action->set_remains(old_action->get_remains());
 215     new_action->set_activity(this);
 216     new_action->set_sharing_penalty(old_action->get_sharing_penalty());
 217     new_action->set_user_bound(old_action->get_user_bound());
 218
 219     old_action->set_activity(nullptr);
 220     old_action->cancel();
 221     old_action->unref();
 222     this->model_action_ = new_action;
 223   }
 224
 225   on_migration(*this, to);
 226   return this;
 227 }
 228
 229 /*************
 230  * Callbacks *
 231  *************/
 232 xbt::signal<void(ExecImpl const&, s4u::Host*)> ExecImpl::on_migration;
 233
 234 } // namespace simgrid::kernel::activity