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
  74 ExecImpl* ExecImpl::start()
  75 {
  76   set_state(State::RUNNING);
  77   if (not MC_is_active() && not MC_record_replay_is_active()) {
  78     if (hosts_.size() == 1) {
  79       surf_action_ = hosts_.front()->get_cpu()->execution_start(flops_amounts_.front(), bound_);
  80       surf_action_->set_sharing_penalty(sharing_penalty_);
  81       surf_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 = hosts_.front()->get_netpoint()->get_englobing_zone()->get_host_model();
  85       surf_action_    = host_model->execute_parallel(hosts_, flops_amounts_.data(), bytes_amounts_.data(), -1);
  86     }
  87     surf_action_->set_activity(this);
  88     set_start_time(surf_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 (surf_action_ == nullptr) ? 0 : surf_action_->get_remains() / surf_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 (surf_action_ == nullptr) ? 0 : surf_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   surf_action_->set_sharing_penalty(sharing_penalty);
 133   return *this;
 134 }
 135
 136 void ExecImpl::post()
 137 {
 138   xbt_assert(surf_action_ != nullptr);
 139   if (std::any_of(hosts_.begin(), hosts_.end(), [](const s4u::Host* host) { return not host->is_on(); })) {
 140     /* If one of the hosts running the synchro failed, notice it. This way, the asking
 141      * process can be killed if it runs on that host itself */
 142     set_state(State::FAILED);
 143   } else if (surf_action_->get_state() == resource::Action::State::FAILED) {
 144     /* If all the hosts are running the synchro didn't fail, then the synchro was canceled */
 145     set_state(State::CANCELED);
 146   } else if (timeout_detector_ && timeout_detector_->get_state() == resource::Action::State::FINISHED) {
 147     if (surf_action_->get_remains() > 0.0) {
 148       surf_action_->set_state(resource::Action::State::FAILED);
 149       set_state(State::TIMEOUT);
 150     } else {
 151       set_state(State::DONE);
 152     }
 153   } else {
 154     set_state(State::DONE);
 155   }
 156
 157   clean_action();
 158   timeout_detector_.reset();
 159   if (get_actor() != nullptr) {
 160     get_actor()->activities_.remove(this);
 161   }
 162   if (get_state() != State::FAILED && cb_id_ >= 0)
 163     s4u::Host::on_state_change.disconnect(cb_id_);
 164   /* Answer all simcalls associated with the synchro */
 165   finish();
 166 }
 167
 168 void ExecImpl::set_exception(actor::ActorImpl* issuer)
 169 {
 170   switch (get_state()) {
 171     case State::FAILED:
 172       static_cast<s4u::Exec*>(get_iface())->complete(s4u::Activity::State::FAILED);
 173       if (issuer->get_host()->is_on())
 174         issuer->exception_ = std::make_exception_ptr(HostFailureException(XBT_THROW_POINT, "Host failed"));
 175       else /* else, the actor will be killed with no possibility to survive */
 176         issuer->context_->set_wannadie();
 177       break;
 178
 179     case State::CANCELED:
 180       issuer->exception_ = std::make_exception_ptr(CancelException(XBT_THROW_POINT, "Execution Canceled"));
 181       break;
 182
 183     case State::TIMEOUT:
 184       issuer->exception_ = std::make_exception_ptr(TimeoutException(XBT_THROW_POINT, "Timeouted"));
 185       break;
 186
 187     default:
 188       xbt_assert(get_state() == State::DONE, "Internal error in ExecImpl::finish(): unexpected synchro state %s",
 189                  get_state_str());
 190   }
 191 }
 192 void ExecImpl::finish()
 193 {
 194   XBT_DEBUG("ExecImpl::finish() in state %s", get_state_str());
 195   while (not simcalls_.empty()) {
 196     smx_simcall_t simcall = simcalls_.front();
 197     simcalls_.pop_front();
 198
 199     if (simcall->call_ == simix::Simcall::NONE) // FIXME: maybe a better way to handle this case
 200       continue;                                 // if process handling comm is killed
 201
 202     handle_activity_waitany(simcall);
 203
 204     set_exception(simcall->issuer_);
 205
 206     simcall->issuer_->waiting_synchro_ = nullptr;
 207     /* Fail the process if the host is down */
 208     if (simcall->issuer_->get_host()->is_on())
 209       simcall->issuer_->simcall_answer();
 210     else
 211       simcall->issuer_->context_->set_wannadie();
 212   }
 213 }
 214
 215 void ExecImpl::reset()
 216 {
 217   hosts_.clear();
 218   bytes_amounts_.clear();
 219   flops_amounts_.clear();
 220   set_start_time(-1.0);
 221 }
 222
 223 ActivityImpl* ExecImpl::migrate(s4u::Host* to)
 224 {
 225   if (not MC_is_active() && not MC_record_replay_is_active()) {
 226     resource::Action* old_action = this->surf_action_;
 227     resource::Action* new_action = to->get_cpu()->execution_start(old_action->get_cost(), old_action->get_user_bound());
 228     new_action->set_remains(old_action->get_remains());
 229     new_action->set_activity(this);
 230     new_action->set_sharing_penalty(old_action->get_sharing_penalty());
 231     new_action->set_user_bound(old_action->get_user_bound());
 232
 233     old_action->set_activity(nullptr);
 234     old_action->cancel();
 235     old_action->unref();
 236     this->surf_action_ = new_action;
 237   }
 238
 239   on_migration(*this, to);
 240   return this;
 241 }
 242
 243 /*************
 244  * Callbacks *
 245  *************/
 246 xbt::signal<void(ExecImpl const&, s4u::Host*)> ExecImpl::on_migration;
 247
 248 } // namespace activity
 249 } // namespace kernel
 250 } // namespace simgrid