src/kernel/activity/ExecImpl.cpp

   1 /* Copyright (c) 2007-2021. 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 "src/kernel/activity/ExecImpl.hpp"
   7 #include "simgrid/Exception.hpp"
   8 #include "simgrid/kernel/routing/NetPoint.hpp"
   9 #include "simgrid/modelchecker.h"
  10 #include "simgrid/s4u/Exec.hpp"
  11 #include "src/mc/checker/SimcallObserver.hpp"
  12 #include "src/mc/mc_replay.hpp"
  13 #include "src/surf/HostImpl.hpp"
  14 #include "src/surf/cpu_interface.hpp"
  15 #include "src/surf/surf_interface.hpp"
  16
  17 #include "simgrid/s4u/Host.hpp"
  18
  19 #include <boost/range/algorithm.hpp>
  20
  21 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(simix_process);
  22
  23 namespace simgrid {
  24 namespace kernel {
  25 namespace activity {
  26
  27 ExecImpl::ExecImpl()
  28 {
  29   piface_                = new s4u::Exec(this);
  30   actor::ActorImpl* self = actor::ActorImpl::self();
  31   if (self) {
  32     actor_ = self;
  33     self->activities_.emplace_back(this);
  34   }
  35 }
  36
  37 ExecImpl& ExecImpl::set_host(s4u::Host* host)
  38 {
  39   hosts_.assign(1, host);
  40   return *this;
  41 }
  42
  43 ExecImpl& ExecImpl::set_hosts(const std::vector<s4u::Host*>& hosts)
  44 {
  45   hosts_ = hosts;
  46   return *this;
  47 }
  48
  49 ExecImpl& ExecImpl::set_timeout(double timeout)
  50 {
  51   if (timeout >= 0 && not MC_is_active() && not MC_record_replay_is_active()) {
  52     timeout_detector_.reset(hosts_.front()->pimpl_cpu->sleep(timeout));
  53     timeout_detector_->set_activity(this);
  54   }
  55   return *this;
  56 }
  57
  58 ExecImpl& ExecImpl::set_flops_amount(double flops_amount)
  59 {
  60   flops_amounts_.assign(1, flops_amount);
  61   return *this;
  62 }
  63
  64 ExecImpl& ExecImpl::set_flops_amounts(const std::vector<double>& flops_amounts)
  65 {
  66   flops_amounts_ = flops_amounts;
  67   return *this;
  68 }
  69
  70 ExecImpl& ExecImpl::set_bytes_amounts(const std::vector<double>& bytes_amounts)
  71 {
  72   bytes_amounts_ = bytes_amounts;
  73
  74   return *this;
  75 }
  76
  77 ExecImpl* ExecImpl::start()
  78 {
  79   state_ = State::RUNNING;
  80   if (not MC_is_active() && not MC_record_replay_is_active()) {
  81     if (hosts_.size() == 1) {
  82       surf_action_ = hosts_.front()->pimpl_cpu->execution_start(flops_amounts_.front());
  83       surf_action_->set_sharing_penalty(sharing_penalty_);
  84       surf_action_->set_category(get_tracing_category());
  85
  86       if (bound_ > 0) {
  87         surf_action_->set_bound(bound_);
  88         surf_action_->set_user_bound(bound_);
  89       }
  90     } else {
  91       // FIXME[donassolo]: verify if all hosts belongs to the same netZone?
  92       auto host_model = hosts_.front()->get_netpoint()->get_englobing_zone()->get_host_model();
  93       surf_action_    = host_model->execute_parallel(hosts_, flops_amounts_.data(), bytes_amounts_.data(), -1);
  94     }
  95     surf_action_->set_activity(this);
  96   }
  97
  98   XBT_DEBUG("Create execute synchro %p: %s", this, get_cname());
  99   return this;
 100 }
 101
 102 double ExecImpl::get_seq_remaining_ratio()
 103 {
 104   return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains() / surf_action_->get_cost();
 105 }
 106
 107 double ExecImpl::get_par_remaining_ratio()
 108 {
 109   // parallel task: their remain is already between 0 and 1
 110   return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains();
 111 }
 112
 113 ExecImpl& ExecImpl::set_bound(double bound)
 114 {
 115   bound_ = bound;
 116   return *this;
 117 }
 118
 119 ExecImpl& ExecImpl::set_sharing_penalty(double sharing_penalty)
 120 {
 121   sharing_penalty_ = sharing_penalty;
 122   return *this;
 123 }
 124
 125 void ExecImpl::post()
 126 {
 127   xbt_assert(surf_action_ != nullptr);
 128   if (std::any_of(hosts_.begin(), hosts_.end(), [](const s4u::Host* host) { return not host->is_on(); })) {
 129     /* If one of the hosts running the synchro failed, notice it. This way, the asking
 130      * process can be killed if it runs on that host itself */
 131     state_ = State::FAILED;
 132   } else if (surf_action_->get_state() == resource::Action::State::FAILED) {
 133     /* If all the hosts are running the synchro didn't fail, then the synchro was canceled */
 134     state_ = State::CANCELED;
 135   } else if (timeout_detector_ && timeout_detector_->get_state() == resource::Action::State::FINISHED) {
 136     if (surf_action_->get_remains() > 0.0) {
 137       surf_action_->set_state(resource::Action::State::FAILED);
 138       state_ = State::TIMEOUT;
 139     } else {
 140       state_ = State::DONE;
 141     }
 142   } else {
 143     state_ = State::DONE;
 144   }
 145
 146   get_iface()->set_finish_time(surf_action_->get_finish_time());
 147
 148   clean_action();
 149   timeout_detector_.reset();
 150   if (actor_) {
 151     actor_->activities_.remove(this);
 152     actor_ = nullptr;
 153   }
 154   /* Answer all simcalls associated with the synchro */
 155   finish();
 156 }
 157
 158 void ExecImpl::finish()
 159 {
 160   while (not simcalls_.empty()) {
 161     smx_simcall_t simcall = simcalls_.front();
 162     simcalls_.pop_front();
 163
 164     /* If a waitany simcall is waiting for this synchro to finish, then remove it from the other synchros in the waitany
 165      * list. Afterwards, get the position of the actual synchro in the waitany list and return it as the result of the
 166      * simcall */
 167
 168     if (simcall->call_ == simix::Simcall::NONE) // FIXME: maybe a better way to handle this case
 169       continue;                                 // if process handling comm is killed
 170     if (const auto* observer = dynamic_cast<mc::ExecutionWaitanySimcall*>(simcall->observer_)) {
 171       const auto* execs = observer->get_execs();
 172
 173       for (auto* exec : *execs) {
 174         // Remove the first occurrence of simcall:
 175         auto j = boost::range::find(exec->simcalls_, simcall);
 176         if (j != exec->simcalls_.end())
 177           exec->simcalls_.erase(j);
 178
 179         if (simcall->timeout_cb_) {
 180           simcall->timeout_cb_->remove();
 181           simcall->timeout_cb_ = nullptr;
 182         }
 183       }
 184
 185       if (not MC_is_active() && not MC_record_replay_is_active()) {
 186         auto element = std::find(execs->begin(), execs->end(), this);
 187         int rank     = (element != execs->end()) ? std::distance(execs->begin(), element) : -1;
 188         simix::marshal<int>(simcall->result_, rank);
 189       }
 190     }
 191     switch (state_) {
 192       case State::DONE:
 193         /* do nothing, synchro done */
 194         XBT_DEBUG("ExecImpl::finish(): execution successful");
 195         break;
 196
 197       case State::FAILED:
 198         XBT_DEBUG("ExecImpl::finish(): host '%s' failed", simcall->issuer_->get_host()->get_cname());
 199         simcall->issuer_->context_->set_wannadie();
 200         if (simcall->issuer_->get_host()->is_on())
 201           simcall->issuer_->exception_ =
 202               std::make_exception_ptr(simgrid::HostFailureException(XBT_THROW_POINT, "Host failed"));
 203         /* else, the actor will be killed with no possibility to survive */
 204         break;
 205
 206       case State::CANCELED:
 207         XBT_DEBUG("ExecImpl::finish(): execution canceled");
 208         simcall->issuer_->exception_ =
 209             std::make_exception_ptr(simgrid::CancelException(XBT_THROW_POINT, "Execution Canceled"));
 210         break;
 211
 212       case State::TIMEOUT:
 213         XBT_DEBUG("ExecImpl::finish(): execution timeouted");
 214         simcall->issuer_->exception_ = std::make_exception_ptr(simgrid::TimeoutException(XBT_THROW_POINT, "Timeouted"));
 215         break;
 216
 217       default:
 218         xbt_die("Internal error in ExecImpl::finish(): unexpected synchro state %d", static_cast<int>(state_));
 219     }
 220
 221     simcall->issuer_->waiting_synchro_ = nullptr;
 222     /* Fail the process if the host is down */
 223     if (simcall->issuer_->get_host()->is_on())
 224       simcall->issuer_->simcall_answer();
 225     else
 226       simcall->issuer_->context_->set_wannadie();
 227   }
 228 }
 229
 230 ActivityImpl* ExecImpl::migrate(s4u::Host* to)
 231 {
 232   if (not MC_is_active() && not MC_record_replay_is_active()) {
 233     resource::Action* old_action = this->surf_action_;
 234     resource::Action* new_action = to->pimpl_cpu->execution_start(old_action->get_cost());
 235     new_action->set_remains(old_action->get_remains());
 236     new_action->set_activity(this);
 237     new_action->set_sharing_penalty(old_action->get_sharing_penalty());
 238     new_action->set_user_bound(old_action->get_user_bound());
 239
 240     old_action->set_activity(nullptr);
 241     old_action->cancel();
 242     old_action->unref();
 243     this->surf_action_ = new_action;
 244   }
 245
 246   on_migration(*this, to);
 247   return this;
 248 }
 249
 250 void ExecImpl::wait_any_for(actor::ActorImpl* issuer, const std::vector<ExecImpl*>* execs, double timeout)
 251 {
 252   if (timeout < 0.0) {
 253     issuer->simcall_.timeout_cb_ = nullptr;
 254   } else {
 255     issuer->simcall_.timeout_cb_ = simgrid::simix::Timer::set(SIMIX_get_clock() + timeout, [issuer, execs]() {
 256       issuer->simcall_.timeout_cb_ = nullptr;
 257       for (auto* exec : *execs) {
 258         // Remove the first occurrence of simcall:
 259         auto j = boost::range::find(exec->simcalls_, &issuer->simcall_);
 260         if (j != exec->simcalls_.end())
 261           exec->simcalls_.erase(j);
 262       }
 263       simix::marshal<int>(issuer->simcall_.result_, -1);
 264       issuer->simcall_answer();
 265     });
 266   }
 267
 268   for (auto* exec : *execs) {
 269     /* associate this simcall to the the synchro */
 270     exec->simcalls_.push_back(&issuer->simcall_);
 271
 272     /* see if the synchro is already finished */
 273     if (exec->state_ != simgrid::kernel::activity::State::WAITING &&
 274         exec->state_ != simgrid::kernel::activity::State::RUNNING) {
 275       exec->finish();
 276       break;
 277     }
 278   }
 279 }
 280
 281 /*************
 282  * Callbacks *
 283  *************/
 284 xbt::signal<void(ExecImpl const&, s4u::Host*)> ExecImpl::on_migration;
 285
 286 } // namespace activity
 287 } // namespace kernel
 288 } // namespace simgrid