1 /* Copyright (c) 2007-2021. The SimGrid Team. All rights reserved. */
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. */
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/kernel/actor/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"
17 #include "simgrid/s4u/Host.hpp"
19 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(simix_process);
27 piface_ = new s4u::Exec(this);
28 actor::ActorImpl* self = actor::ActorImpl::self();
31 self->activities_.emplace_back(this);
35 ExecImpl& ExecImpl::set_host(s4u::Host* host)
37 hosts_.assign(1, host);
41 ExecImpl& ExecImpl::set_hosts(const std::vector<s4u::Host*>& hosts)
47 ExecImpl& ExecImpl::set_timeout(double timeout)
49 if (timeout >= 0 && not MC_is_active() && not MC_record_replay_is_active()) {
50 timeout_detector_.reset(hosts_.front()->pimpl_cpu->sleep(timeout));
51 timeout_detector_->set_activity(this);
56 ExecImpl& ExecImpl::set_flops_amount(double flops_amount)
58 flops_amounts_.assign(1, flops_amount);
62 ExecImpl& ExecImpl::set_flops_amounts(const std::vector<double>& flops_amounts)
64 flops_amounts_ = flops_amounts;
68 ExecImpl& ExecImpl::set_bytes_amounts(const std::vector<double>& bytes_amounts)
70 bytes_amounts_ = bytes_amounts;
75 ExecImpl* ExecImpl::start()
77 state_ = State::RUNNING;
78 if (not MC_is_active() && not MC_record_replay_is_active()) {
79 if (hosts_.size() == 1) {
80 surf_action_ = hosts_.front()->pimpl_cpu->execution_start(flops_amounts_.front());
81 surf_action_->set_sharing_penalty(sharing_penalty_);
82 surf_action_->set_category(get_tracing_category());
85 surf_action_->set_bound(bound_);
86 surf_action_->set_user_bound(bound_);
89 // FIXME[donassolo]: verify if all hosts belongs to the same netZone?
90 auto host_model = hosts_.front()->get_netpoint()->get_englobing_zone()->get_host_model();
91 surf_action_ = host_model->execute_parallel(hosts_, flops_amounts_.data(), bytes_amounts_.data(), -1);
93 surf_action_->set_activity(this);
96 XBT_DEBUG("Create execute synchro %p: %s", this, get_cname());
100 double ExecImpl::get_seq_remaining_ratio()
102 return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains() / surf_action_->get_cost();
105 double ExecImpl::get_par_remaining_ratio()
107 // parallel task: their remain is already between 0 and 1
108 return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains();
111 ExecImpl& ExecImpl::set_bound(double bound)
117 ExecImpl& ExecImpl::set_sharing_penalty(double sharing_penalty)
119 sharing_penalty_ = sharing_penalty;
123 void ExecImpl::post()
125 xbt_assert(surf_action_ != nullptr);
126 if (std::any_of(hosts_.begin(), hosts_.end(), [](const s4u::Host* host) { return not host->is_on(); })) {
127 /* If one of the hosts running the synchro failed, notice it. This way, the asking
128 * process can be killed if it runs on that host itself */
129 state_ = State::FAILED;
130 } else if (surf_action_->get_state() == resource::Action::State::FAILED) {
131 /* If all the hosts are running the synchro didn't fail, then the synchro was canceled */
132 state_ = State::CANCELED;
133 } else if (timeout_detector_ && timeout_detector_->get_state() == resource::Action::State::FINISHED) {
134 if (surf_action_->get_remains() > 0.0) {
135 surf_action_->set_state(resource::Action::State::FAILED);
136 state_ = State::TIMEOUT;
138 state_ = State::DONE;
141 state_ = State::DONE;
144 get_iface()->set_finish_time(surf_action_->get_finish_time());
147 timeout_detector_.reset();
149 actor_->activities_.remove(this);
152 /* Answer all simcalls associated with the synchro */
156 void ExecImpl::finish()
158 while (not simcalls_.empty()) {
159 smx_simcall_t simcall = simcalls_.front();
160 simcalls_.pop_front();
162 /* If a waitany simcall is waiting for this synchro to finish, then remove it from the other synchros in the waitany
163 * list. Afterwards, get the position of the actual synchro in the waitany list and return it as the result of the
166 if (simcall->call_ == simix::Simcall::NONE) // FIXME: maybe a better way to handle this case
167 continue; // if process handling comm is killed
169 dynamic_cast<kernel::actor::ExecutionWaitanySimcall*>(simcall->observer_)) { // simcall is a wait_any?
170 const auto* execs = observer->get_execs();
172 for (auto* exec : *execs) {
173 exec->unregister_simcall(simcall);
175 if (simcall->timeout_cb_) {
176 simcall->timeout_cb_->remove();
177 simcall->timeout_cb_ = nullptr;
181 if (not MC_is_active() && not MC_record_replay_is_active()) {
182 auto element = std::find(execs->begin(), execs->end(), this);
183 int rank = element != execs->end() ? static_cast<int>(std::distance(execs->begin(), element)) : -1;
184 observer->set_result(rank);
189 /* do nothing, synchro done */
190 XBT_DEBUG("ExecImpl::finish(): execution successful");
194 XBT_DEBUG("ExecImpl::finish(): host '%s' failed", simcall->issuer_->get_host()->get_cname());
195 simcall->issuer_->context_->set_wannadie();
196 if (simcall->issuer_->get_host()->is_on())
197 simcall->issuer_->exception_ =
198 std::make_exception_ptr(simgrid::HostFailureException(XBT_THROW_POINT, "Host failed"));
199 /* else, the actor will be killed with no possibility to survive */
202 case State::CANCELED:
203 XBT_DEBUG("ExecImpl::finish(): execution canceled");
204 simcall->issuer_->exception_ =
205 std::make_exception_ptr(simgrid::CancelException(XBT_THROW_POINT, "Execution Canceled"));
209 XBT_DEBUG("ExecImpl::finish(): execution timeouted");
210 simcall->issuer_->exception_ = std::make_exception_ptr(simgrid::TimeoutException(XBT_THROW_POINT, "Timeouted"));
214 xbt_die("Internal error in ExecImpl::finish(): unexpected synchro state %d", static_cast<int>(state_));
217 simcall->issuer_->waiting_synchro_ = nullptr;
218 /* Fail the process if the host is down */
219 if (simcall->issuer_->get_host()->is_on())
220 simcall->issuer_->simcall_answer();
222 simcall->issuer_->context_->set_wannadie();
226 ActivityImpl* ExecImpl::migrate(s4u::Host* to)
228 if (not MC_is_active() && not MC_record_replay_is_active()) {
229 resource::Action* old_action = this->surf_action_;
230 resource::Action* new_action = to->pimpl_cpu->execution_start(old_action->get_cost());
231 new_action->set_remains(old_action->get_remains());
232 new_action->set_activity(this);
233 new_action->set_sharing_penalty(old_action->get_sharing_penalty());
234 new_action->set_user_bound(old_action->get_user_bound());
236 old_action->set_activity(nullptr);
237 old_action->cancel();
239 this->surf_action_ = new_action;
242 on_migration(*this, to);
246 void ExecImpl::wait_any_for(actor::ActorImpl* issuer, const std::vector<ExecImpl*>* execs, double timeout)
249 issuer->simcall_.timeout_cb_ = nullptr;
251 issuer->simcall_.timeout_cb_ = simgrid::simix::Timer::set(SIMIX_get_clock() + timeout, [issuer, execs]() {
252 issuer->simcall_.timeout_cb_ = nullptr;
253 for (auto* exec : *execs)
254 exec->unregister_simcall(&issuer->simcall_);
255 // default result (-1) is set in mc::ExecutionWaitanySimcall
256 issuer->simcall_answer();
260 for (auto* exec : *execs) {
261 /* associate this simcall to the the synchro */
262 exec->simcalls_.push_back(&issuer->simcall_);
264 /* see if the synchro is already finished */
265 if (exec->state_ != simgrid::kernel::activity::State::WAITING &&
266 exec->state_ != simgrid::kernel::activity::State::RUNNING) {
276 xbt::signal<void(ExecImpl const&, s4u::Host*)> ExecImpl::on_migration;
278 } // namespace activity
279 } // namespace kernel
280 } // namespace simgrid