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