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 "simgrid/s4u/Exec.hpp"
7 #include "src/kernel/activity/ExecImpl.hpp"
8 #include "simgrid/Exception.hpp"
9 #include "simgrid/modelchecker.h"
10 #include "src/mc/mc_replay.hpp"
11 #include "src/surf/HostImpl.hpp"
12 #include "src/surf/cpu_interface.hpp"
13 #include "src/surf/surf_interface.hpp"
15 #include "simgrid/s4u/Host.hpp"
17 #include <boost/range/algorithm.hpp>
19 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(simix_process);
21 void simcall_HANDLER_execution_waitany_for(smx_simcall_t simcall, simgrid::kernel::activity::ExecImpl* execs[],
22 size_t count, double timeout)
25 simcall->timeout_cb_ = nullptr;
27 simcall->timeout_cb_ = simgrid::simix::Timer::set(SIMIX_get_clock() + timeout, [simcall, execs, count]() {
28 simcall->timeout_cb_ = nullptr;
29 for (size_t i = 0; i < count; i++) {
30 // Remove the first occurrence of simcall:
31 auto* exec = execs[i];
32 auto j = boost::range::find(exec->simcalls_, simcall);
33 if (j != exec->simcalls_.end())
34 exec->simcalls_.erase(j);
36 simcall_execution_waitany_for__set__result(simcall, -1);
37 simcall->issuer_->simcall_answer();
41 for (size_t i = 0; i < count; i++) {
42 /* associate this simcall to the the synchro */
43 auto* exec = execs[i];
44 exec->simcalls_.push_back(simcall);
46 /* see if the synchro is already finished */
47 if (exec->state_ != simgrid::kernel::activity::State::WAITING &&
48 exec->state_ != simgrid::kernel::activity::State::RUNNING) {
61 piface_ = new s4u::Exec(this);
62 actor::ActorImpl* self = actor::ActorImpl::self();
65 self->activities_.emplace_back(this);
69 ExecImpl& ExecImpl::set_host(s4u::Host* host)
71 hosts_.assign(1, host);
75 ExecImpl& ExecImpl::set_hosts(const std::vector<s4u::Host*>& hosts)
81 ExecImpl& ExecImpl::set_timeout(double timeout)
83 if (timeout >= 0 && not MC_is_active() && not MC_record_replay_is_active()) {
84 timeout_detector_.reset(hosts_.front()->pimpl_cpu->sleep(timeout));
85 timeout_detector_->set_activity(this);
90 ExecImpl& ExecImpl::set_flops_amount(double flops_amount)
92 flops_amounts_.assign(1, flops_amount);
96 ExecImpl& ExecImpl::set_flops_amounts(const std::vector<double>& flops_amounts)
98 flops_amounts_ = flops_amounts;
102 ExecImpl& ExecImpl::set_bytes_amounts(const std::vector<double>& bytes_amounts)
104 bytes_amounts_ = bytes_amounts;
109 ExecImpl* ExecImpl::start()
111 state_ = State::RUNNING;
112 if (not MC_is_active() && not MC_record_replay_is_active()) {
113 if (hosts_.size() == 1) {
114 surf_action_ = hosts_.front()->pimpl_cpu->execution_start(flops_amounts_.front());
115 surf_action_->set_sharing_penalty(sharing_penalty_);
116 surf_action_->set_category(get_tracing_category());
119 surf_action_->set_bound(bound_);
121 surf_action_ = surf_host_model->execute_parallel(hosts_, flops_amounts_.data(), bytes_amounts_.data(), -1);
123 surf_action_->set_activity(this);
126 XBT_DEBUG("Create execute synchro %p: %s", this, get_cname());
130 double ExecImpl::get_seq_remaining_ratio()
132 return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains() / surf_action_->get_cost();
135 double ExecImpl::get_par_remaining_ratio()
137 // parallel task: their remain is already between 0 and 1
138 return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains();
141 ExecImpl& ExecImpl::set_bound(double bound)
147 ExecImpl& ExecImpl::set_sharing_penalty(double sharing_penalty)
149 sharing_penalty_ = sharing_penalty;
153 void ExecImpl::post()
155 xbt_assert(surf_action_ != nullptr);
156 if (hosts_.size() == 1 && not hosts_.front()->is_on()) { /* FIXME: handle resource failure for parallel tasks too */
157 /* If the host running the synchro failed, notice it. This way, the asking
158 * process can be killed if it runs on that host itself */
159 state_ = State::FAILED;
160 } else if (surf_action_->get_state() == resource::Action::State::FAILED) {
161 /* If the host running the synchro didn't fail, then the synchro was canceled */
162 state_ = State::CANCELED;
163 } else if (timeout_detector_ && timeout_detector_->get_state() == resource::Action::State::FINISHED) {
164 state_ = State::TIMEOUT;
166 state_ = State::DONE;
169 get_iface()->set_finish_time(surf_action_->get_finish_time());
172 timeout_detector_.reset();
174 actor_->activities_.remove(this);
177 /* Answer all simcalls associated with the synchro */
181 void ExecImpl::finish()
183 while (not simcalls_.empty()) {
184 smx_simcall_t simcall = simcalls_.front();
185 simcalls_.pop_front();
187 /* If a waitany simcall is waiting for this synchro to finish, then remove it from the other synchros in the waitany
188 * list. Afterwards, get the position of the actual synchro in the waitany list and return it as the result of the
191 if (simcall->call_ == simix::Simcall::NONE) // FIXME: maybe a better way to handle this case
192 continue; // if process handling comm is killed
193 if (simcall->call_ == simix::Simcall::EXECUTION_WAITANY_FOR) {
194 simgrid::kernel::activity::ExecImpl** execs = simcall_execution_waitany_for__get__execs(simcall);
195 size_t count = simcall_execution_waitany_for__get__count(simcall);
197 for (size_t i = 0; i < count; i++) {
198 // Remove the first occurrence of simcall:
199 auto* exec = execs[i];
200 auto j = boost::range::find(exec->simcalls_, simcall);
201 if (j != exec->simcalls_.end())
202 exec->simcalls_.erase(j);
204 if (simcall->timeout_cb_) {
205 simcall->timeout_cb_->remove();
206 simcall->timeout_cb_ = nullptr;
210 if (not MC_is_active() && not MC_record_replay_is_active()) {
211 ExecImpl** element = std::find(execs, execs + count, this);
212 int rank = (element != execs + count) ? element - execs : -1;
213 simcall_execution_waitany_for__set__result(simcall, rank);
218 /* do nothing, synchro done */
219 XBT_DEBUG("ExecImpl::finish(): execution successful");
223 XBT_DEBUG("ExecImpl::finish(): host '%s' failed", simcall->issuer_->get_host()->get_cname());
224 simcall->issuer_->context_->set_wannadie();
225 if (simcall->issuer_->get_host()->is_on())
226 simcall->issuer_->exception_ =
227 std::make_exception_ptr(simgrid::HostFailureException(XBT_THROW_POINT, "Host failed"));
228 /* else, the actor will be killed with no possibility to survive */
231 case State::CANCELED:
232 XBT_DEBUG("ExecImpl::finish(): execution canceled");
233 simcall->issuer_->exception_ =
234 std::make_exception_ptr(simgrid::CancelException(XBT_THROW_POINT, "Execution Canceled"));
238 XBT_DEBUG("ExecImpl::finish(): execution timeouted");
239 simcall->issuer_->exception_ = std::make_exception_ptr(simgrid::TimeoutException(XBT_THROW_POINT, "Timeouted"));
243 xbt_die("Internal error in ExecImpl::finish(): unexpected synchro state %d", static_cast<int>(state_));
246 simcall->issuer_->waiting_synchro_ = nullptr;
247 /* Fail the process if the host is down */
248 if (simcall->issuer_->get_host()->is_on())
249 simcall->issuer_->simcall_answer();
251 simcall->issuer_->context_->set_wannadie();
255 ActivityImpl* ExecImpl::migrate(s4u::Host* to)
257 if (not MC_is_active() && not MC_record_replay_is_active()) {
258 resource::Action* old_action = this->surf_action_;
259 resource::Action* new_action = to->pimpl_cpu->execution_start(old_action->get_cost());
260 new_action->set_remains(old_action->get_remains());
261 new_action->set_activity(this);
262 new_action->set_sharing_penalty(old_action->get_sharing_penalty());
264 // FIXME: the user-defined bound seem to not be kept by LMM, that seem to overwrite it for the multi-core modeling.
265 // I hope that the user did not provide any.
267 old_action->set_activity(nullptr);
268 old_action->cancel();
270 this->surf_action_ = new_action;
273 on_migration(*this, to);
280 xbt::signal<void(ExecImpl const&, s4u::Host*)> ExecImpl::on_migration;
282 } // namespace activity
283 } // namespace kernel
284 } // namespace simgrid