1 /* Copyright (c) 2007-2022. 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 <simgrid/Exception.hpp>
7 #include <simgrid/kernel/routing/NetPoint.hpp>
8 #include <simgrid/modelchecker.h>
9 #include <simgrid/s4u/Engine.hpp>
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"
18 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(ker_actor);
26 piface_ = new s4u::Exec(this);
27 actor::ActorImpl* self = actor::ActorImpl::self();
30 self->activities_.emplace_back(this);
34 ExecImpl& ExecImpl::set_host(s4u::Host* host)
36 hosts_.assign(1, host);
40 ExecImpl& ExecImpl::set_hosts(const std::vector<s4u::Host*>& hosts)
46 ExecImpl& ExecImpl::set_timeout(double timeout)
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);
55 ExecImpl& ExecImpl::set_flops_amount(double flops_amount)
57 flops_amounts_.assign(1, flops_amount);
61 ExecImpl& ExecImpl::set_flops_amounts(const std::vector<double>& flops_amounts)
63 flops_amounts_ = flops_amounts;
67 ExecImpl& ExecImpl::set_bytes_amounts(const std::vector<double>& bytes_amounts)
69 bytes_amounts_ = bytes_amounts;
74 ExecImpl* ExecImpl::start()
76 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());
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);
87 surf_action_->set_activity(this);
88 start_time_ = surf_action_->get_start_time();
91 XBT_DEBUG("Create execute synchro %p: %s", this, get_cname());
95 double ExecImpl::get_remaining() const
97 if (state_ == State::WAITING || state_ == State::FAILED)
98 return flops_amounts_.front();
99 return ActivityImpl::get_remaining();
102 double ExecImpl::get_seq_remaining_ratio()
104 if (state_ == State::WAITING)
106 return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains() / surf_action_->get_cost();
109 double ExecImpl::get_par_remaining_ratio()
111 // parallel task: their remain is already between 0 and 1
112 if (state_ == State::WAITING)
114 return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains();
117 ExecImpl& ExecImpl::set_bound(double bound)
123 ExecImpl& ExecImpl::set_sharing_penalty(double sharing_penalty)
125 sharing_penalty_ = sharing_penalty;
129 ExecImpl& ExecImpl::update_sharing_penalty(double sharing_penalty)
131 sharing_penalty_ = sharing_penalty;
132 surf_action_->set_sharing_penalty(sharing_penalty);
136 void ExecImpl::post()
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 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 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 state_ = State::TIMEOUT;
151 state_ = State::DONE;
154 state_ = State::DONE;
158 timeout_detector_.reset();
159 if (get_actor() != nullptr) {
160 get_actor()->activities_.remove(this);
162 if (state_ != State::FAILED && cb_id_ >= 0)
163 s4u::Host::on_state_change.disconnect(cb_id_);
164 /* Answer all simcalls associated with the synchro */
168 void ExecImpl::set_exception(actor::ActorImpl* issuer)
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();
179 case State::CANCELED:
180 issuer->exception_ = std::make_exception_ptr(CancelException(XBT_THROW_POINT, "Execution Canceled"));
184 issuer->exception_ = std::make_exception_ptr(TimeoutException(XBT_THROW_POINT, "Timeouted"));
188 xbt_assert(state_ == State::DONE, "Internal error in ExecImpl::finish(): unexpected synchro state %s",
192 void ExecImpl::finish()
194 XBT_DEBUG("ExecImpl::finish() in state %s", to_c_str(state_));
195 while (not simcalls_.empty()) {
196 smx_simcall_t simcall = simcalls_.front();
197 simcalls_.pop_front();
199 if (simcall->call_ == simix::Simcall::NONE) // FIXME: maybe a better way to handle this case
200 continue; // if process handling comm is killed
202 /* If a waitany simcall is waiting for this synchro to finish, then remove it from the other synchros in the waitany
203 * list. Afterwards, get the position of the actual synchro in the waitany list and return it as the result of the
205 if (auto* observer = dynamic_cast<actor::ExecutionWaitanySimcall*>(simcall->observer_)) {
206 const auto& execs = observer->get_execs();
208 for (auto* exec : execs)
209 exec->unregister_simcall(simcall);
211 if (simcall->timeout_cb_) {
212 simcall->timeout_cb_->remove();
213 simcall->timeout_cb_ = nullptr;
216 if (not MC_is_active() && not MC_record_replay_is_active()) {
217 auto element = std::find(execs.begin(), execs.end(), this);
218 int rank = element != execs.end() ? static_cast<int>(std::distance(execs.begin(), element)) : -1;
219 observer->set_result(rank);
223 set_exception(simcall->issuer_);
225 simcall->issuer_->waiting_synchro_ = nullptr;
226 /* Fail the process if the host is down */
227 if (simcall->issuer_->get_host()->is_on())
228 simcall->issuer_->simcall_answer();
230 simcall->issuer_->context_->set_wannadie();
234 void ExecImpl::reset()
237 bytes_amounts_.clear();
238 flops_amounts_.clear();
242 ActivityImpl* ExecImpl::migrate(s4u::Host* to)
244 if (not MC_is_active() && not MC_record_replay_is_active()) {
245 resource::Action* old_action = this->surf_action_;
246 resource::Action* new_action = to->get_cpu()->execution_start(old_action->get_cost(), old_action->get_user_bound());
247 new_action->set_remains(old_action->get_remains());
248 new_action->set_activity(this);
249 new_action->set_sharing_penalty(old_action->get_sharing_penalty());
250 new_action->set_user_bound(old_action->get_user_bound());
252 old_action->set_activity(nullptr);
253 old_action->cancel();
255 this->surf_action_ = new_action;
258 on_migration(*this, to);
262 void ExecImpl::wait_any_for(actor::ActorImpl* issuer, const std::vector<ExecImpl*>& execs, double timeout)
265 issuer->simcall_.timeout_cb_ = nullptr;
267 issuer->simcall_.timeout_cb_ = timer::Timer::set(s4u::Engine::get_clock() + timeout, [issuer, &execs]() {
268 issuer->simcall_.timeout_cb_ = nullptr;
269 for (auto* exec : execs)
270 exec->unregister_simcall(&issuer->simcall_);
271 // default result (-1) is set in actor::ExecutionWaitanySimcall
272 issuer->simcall_answer();
276 for (auto* exec : execs) {
277 /* associate this simcall to the the synchro */
278 exec->simcalls_.push_back(&issuer->simcall_);
279 /* see if the synchro is already finished */
280 if (exec->state_ != State::WAITING && exec->state_ != State::RUNNING) {
290 xbt::signal<void(ExecImpl const&, s4u::Host*)> ExecImpl::on_migration;
292 } // namespace activity
293 } // namespace kernel
294 } // namespace simgrid