src/kernel/activity/ExecImpl.cpp

   1 /* Copyright (c) 2007-2020. 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/modelchecker.h"
   9 #include "src/mc/mc_replay.hpp"
  10 #include "src/surf/HostImpl.hpp"
  11 #include "src/surf/cpu_interface.hpp"
  12 #include "src/surf/surf_interface.hpp"
  13
  14 #include "simgrid/s4u/Host.hpp"
  15
  16 #include <boost/range/algorithm.hpp>
  17
  18 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(simix_process);
  19
  20 void simcall_HANDLER_execution_waitany_for(smx_simcall_t simcall, simgrid::kernel::activity::ExecImpl* execs[],
  21                                            size_t count, double timeout)
  22 {
  23   if (timeout < 0.0) {
  24     simcall->timeout_cb_ = nullptr;
  25   } else {
  26     simcall->timeout_cb_ = simgrid::simix::Timer::set(SIMIX_get_clock() + timeout, [simcall, execs, count]() {
  27       for (size_t i = 0; i < count; i++) {
  28         // Remove the first occurrence of simcall:
  29         auto* exec = execs[i];
  30         auto j     = boost::range::find(exec->simcalls_, simcall);
  31         if (j != exec->simcalls_.end())
  32           exec->simcalls_.erase(j);
  33       }
  34       simcall_execution_waitany_for__set__result(simcall, -1);
  35       simcall->issuer_->simcall_answer();
  36     });
  37   }
  38
  39   for (size_t i = 0; i < count; i++) {
  40     /* associate this simcall to the the synchro */
  41     auto* exec = execs[i];
  42     exec->simcalls_.push_back(simcall);
  43
  44     /* see if the synchro is already finished */
  45     if (exec->state_ != simgrid::kernel::activity::State::WAITING &&
  46         exec->state_ != simgrid::kernel::activity::State::RUNNING) {
  47       exec->finish();
  48       break;
  49     }
  50   }
  51 }
  52
  53 namespace simgrid {
  54 namespace kernel {
  55 namespace activity {
  56
  57 ExecImpl& ExecImpl::set_host(s4u::Host* host)
  58 {
  59   hosts_.assign(1, host);
  60   return *this;
  61 }
  62
  63 ExecImpl& ExecImpl::set_hosts(const std::vector<s4u::Host*>& hosts)
  64 {
  65   hosts_ = hosts;
  66   return *this;
  67 }
  68
  69 ExecImpl& ExecImpl::set_timeout(double timeout)
  70 {
  71   if (timeout > 0 && not MC_is_active() && not MC_record_replay_is_active()) {
  72     timeout_detector_.reset(hosts_.front()->pimpl_cpu->sleep(timeout));
  73     timeout_detector_->set_activity(this);
  74   }
  75   return *this;
  76 }
  77
  78 ExecImpl& ExecImpl::set_flops_amount(double flops_amount)
  79 {
  80   flops_amounts_.assign(1, flops_amount);
  81   return *this;
  82 }
  83
  84 ExecImpl& ExecImpl::set_flops_amounts(const std::vector<double>& flops_amounts)
  85 {
  86   flops_amounts_ = flops_amounts;
  87   return *this;
  88 }
  89
  90 ExecImpl& ExecImpl::set_bytes_amounts(const std::vector<double>& bytes_amounts)
  91 {
  92   bytes_amounts_ = bytes_amounts;
  93
  94   return *this;
  95 }
  96
  97 ExecImpl* ExecImpl::start()
  98 {
  99   state_ = State::RUNNING;
 100   if (not MC_is_active() && not MC_record_replay_is_active()) {
 101     if (hosts_.size() == 1) {
 102       surf_action_ = hosts_.front()->pimpl_cpu->execution_start(flops_amounts_.front());
 103       surf_action_->set_sharing_penalty(sharing_penalty_);
 104       surf_action_->set_category(get_tracing_category());
 105
 106       if (bound_ > 0)
 107         surf_action_->set_bound(bound_);
 108     } else {
 109       surf_action_ = surf_host_model->execute_parallel(hosts_, flops_amounts_.data(), bytes_amounts_.data(), -1);
 110     }
 111     surf_action_->set_activity(this);
 112   }
 113
 114   XBT_DEBUG("Create execute synchro %p: %s", this, get_cname());
 115   return this;
 116 }
 117
 118 double ExecImpl::get_seq_remaining_ratio()
 119 {
 120   return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains() / surf_action_->get_cost();
 121 }
 122
 123 double ExecImpl::get_par_remaining_ratio()
 124 {
 125   // parallel task: their remain is already between 0 and 1
 126   return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains();
 127 }
 128
 129 ExecImpl& ExecImpl::set_bound(double bound)
 130 {
 131   bound_ = bound;
 132   return *this;
 133 }
 134
 135 ExecImpl& ExecImpl::set_sharing_penalty(double sharing_penalty)
 136 {
 137   sharing_penalty_ = sharing_penalty;
 138   return *this;
 139 }
 140
 141 void ExecImpl::post()
 142 {
 143   if (hosts_.size() == 1 && not hosts_.front()->is_on()) { /* FIXME: handle resource failure for parallel tasks too */
 144     /* If the host running the synchro failed, notice it. This way, the asking
 145      * process can be killed if it runs on that host itself */
 146     state_ = State::FAILED;
 147   } else if (surf_action_ && surf_action_->get_state() == resource::Action::State::FAILED) {
 148     /* If the host running the synchro didn't fail, then the synchro was canceled */
 149     state_ = State::CANCELED;
 150   } else if (timeout_detector_ && timeout_detector_->get_state() == resource::Action::State::FINISHED) {
 151     state_ = State::TIMEOUT;
 152   } else {
 153     state_ = State::DONE;
 154   }
 155
 156   clean_action();
 157   timeout_detector_.reset();
 158   /* Answer all simcalls associated with the synchro */
 159   finish();
 160 }
 161
 162 void ExecImpl::finish()
 163 {
 164   while (not simcalls_.empty()) {
 165     smx_simcall_t simcall = simcalls_.front();
 166     simcalls_.pop_front();
 167
 168     /* If a waitany simcall is waiting for this synchro to finish, then remove it from the other synchros in the waitany
 169      * list. Afterwards, get the position of the actual synchro in the waitany list and return it as the result of the
 170      * simcall */
 171
 172     if (simcall->call_ == SIMCALL_NONE) // FIXME: maybe a better way to handle this case
 173       continue;                        // if process handling comm is killed
 174     if (simcall->call_ == SIMCALL_EXECUTION_WAITANY_FOR) {
 175       simgrid::kernel::activity::ExecImpl** execs = simcall_execution_waitany_for__get__execs(simcall);
 176       size_t count                                = simcall_execution_waitany_for__get__count(simcall);
 177
 178       for (size_t i = 0; i < count; i++) {
 179         // Remove the first occurrence of simcall:
 180         auto* exec = execs[i];
 181         auto j     = boost::range::find(exec->simcalls_, simcall);
 182         if (j != exec->simcalls_.end())
 183           exec->simcalls_.erase(j);
 184
 185         if (simcall->timeout_cb_) {
 186           simcall->timeout_cb_->remove();
 187           simcall->timeout_cb_ = nullptr;
 188         }
 189       }
 190
 191       if (not MC_is_active() && not MC_record_replay_is_active()) {
 192         ExecImpl** element = std::find(execs, execs + count, this);
 193         int rank           = (element != execs + count) ? element - execs : -1;
 194         simcall_execution_waitany_for__set__result(simcall, rank);
 195       }
 196     }
 197
 198     switch (state_) {
 199       case State::DONE:
 200         /* do nothing, synchro done */
 201         XBT_DEBUG("ExecImpl::finish(): execution successful");
 202         break;
 203
 204       case State::FAILED:
 205         XBT_DEBUG("ExecImpl::finish(): host '%s' failed", simcall->issuer_->get_host()->get_cname());
 206         simcall->issuer_->context_->set_wannadie();
 207         if (simcall->issuer_->get_host()->is_on())
 208           simcall->issuer_->exception_ =
 209               std::make_exception_ptr(simgrid::HostFailureException(XBT_THROW_POINT, "Host failed"));
 210         /* else, the actor will be killed with no possibility to survive */
 211         break;
 212
 213       case State::CANCELED:
 214         XBT_DEBUG("ExecImpl::finish(): execution canceled");
 215         simcall->issuer_->exception_ =
 216             std::make_exception_ptr(simgrid::CancelException(XBT_THROW_POINT, "Execution Canceled"));
 217         break;
 218
 219       case State::TIMEOUT:
 220         XBT_DEBUG("ExecImpl::finish(): execution timeouted");
 221         simcall->issuer_->exception_ = std::make_exception_ptr(simgrid::TimeoutException(XBT_THROW_POINT, "Timeouted"));
 222         break;
 223
 224       default:
 225         xbt_die("Internal error in ExecImpl::finish(): unexpected synchro state %d", static_cast<int>(state_));
 226     }
 227
 228     simcall->issuer_->waiting_synchro = nullptr;
 229     /* Fail the process if the host is down */
 230     if (simcall->issuer_->get_host()->is_on())
 231       simcall->issuer_->simcall_answer();
 232     else
 233       simcall->issuer_->context_->set_wannadie();
 234   }
 235 }
 236
 237 ActivityImpl* ExecImpl::migrate(s4u::Host* to)
 238 {
 239   if (not MC_is_active() && not MC_record_replay_is_active()) {
 240     resource::Action* old_action = this->surf_action_;
 241     resource::Action* new_action = to->pimpl_cpu->execution_start(old_action->get_cost());
 242     new_action->set_remains(old_action->get_remains());
 243     new_action->set_activity(this);
 244     new_action->set_sharing_penalty(old_action->get_sharing_penalty());
 245
 246     // FIXME: the user-defined bound seem to not be kept by LMM, that seem to overwrite it for the multi-core modeling.
 247     // I hope that the user did not provide any.
 248
 249     old_action->set_activity(nullptr);
 250     old_action->cancel();
 251     old_action->unref();
 252     this->surf_action_ = new_action;
 253   }
 254
 255   on_migration(*this, to);
 256   return this;
 257 }
 258
 259 /*************
 260  * Callbacks *
 261  *************/
 262 xbt::signal<void(ExecImpl const&, s4u::Host*)> ExecImpl::on_migration;
 263
 264 } // namespace activity
 265 } // namespace kernel
 266 } // namespace simgrid