1 /* Copyright (c) 2007-2020. 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/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"
14 #include "simgrid/s4u/Host.hpp"
16 #include <boost/range/algorithm.hpp>
18 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(simix_process);
20 void simcall_HANDLER_execution_wait(smx_simcall_t simcall, simgrid::kernel::activity::ExecImpl* synchro, double timeout)
22 XBT_DEBUG("Wait for execution of synchro %p, state %d", synchro, (int)synchro->state_);
23 xbt_assert(std::isfinite(timeout), "timeout is not finite!");
25 /* Associate this simcall to the synchro */
26 synchro->register_simcall(simcall);
28 /* set surf's synchro */
29 if (MC_is_active() || MC_record_replay_is_active()) {
30 int idx = SIMCALL_GET_MC_VALUE(*simcall);
32 synchro->state_ = simgrid::kernel::activity::State::DONE;
34 /* If we reached this point, the wait simcall must have a timeout */
35 /* Otherwise it shouldn't be enabled and executed by the MC */
38 synchro->state_ = simgrid::kernel::activity::State::TIMEOUT;
44 /* If the synchro is already finished then perform the error handling */
45 if (synchro->state_ != simgrid::kernel::activity::State::RUNNING) {
47 } else { /* we need a sleep action (even when there is no timeout) to be notified of host failures */
48 synchro->set_timeout(timeout);
52 void simcall_HANDLER_execution_waitany_for(smx_simcall_t simcall, simgrid::kernel::activity::ExecImpl* execs[],
53 size_t count, double timeout)
56 simcall->timeout_cb_ = nullptr;
58 simcall->timeout_cb_ = simgrid::simix::Timer::set(SIMIX_get_clock() + timeout, [simcall, execs, count]() {
59 for (size_t i = 0; i < count; i++) {
60 // Remove the first occurrence of simcall:
61 auto* exec = execs[i];
62 auto j = boost::range::find(exec->simcalls_, simcall);
63 if (j != exec->simcalls_.end())
64 exec->simcalls_.erase(j);
66 simcall_execution_waitany_for__set__result(simcall, -1);
67 simcall->issuer_->simcall_answer();
71 for (size_t i = 0; i < count; i++) {
72 /* associate this simcall to the the synchro */
73 auto* exec = execs[i];
74 exec->simcalls_.push_back(simcall);
76 /* see if the synchro is already finished */
77 if (exec->state_ != simgrid::kernel::activity::State::WAITING &&
78 exec->state_ != simgrid::kernel::activity::State::RUNNING) {
89 ExecImpl& ExecImpl::set_host(s4u::Host* host)
91 hosts_.assign(1, host);
95 ExecImpl& ExecImpl::set_hosts(const std::vector<s4u::Host*>& hosts)
101 ExecImpl& ExecImpl::set_timeout(double timeout)
103 if (timeout > 0 && not MC_is_active() && not MC_record_replay_is_active()) {
104 timeout_detector_.reset(hosts_.front()->pimpl_cpu->sleep(timeout));
105 timeout_detector_->set_activity(this);
110 ExecImpl& ExecImpl::set_flops_amount(double flops_amount)
112 flops_amounts_.assign(1, flops_amount);
116 ExecImpl& ExecImpl::set_flops_amounts(const std::vector<double>& flops_amounts)
118 flops_amounts_ = flops_amounts;
122 ExecImpl& ExecImpl::set_bytes_amounts(const std::vector<double>& bytes_amounts)
124 bytes_amounts_ = bytes_amounts;
129 ExecImpl* ExecImpl::start()
131 state_ = State::RUNNING;
132 if (not MC_is_active() && not MC_record_replay_is_active()) {
133 if (hosts_.size() == 1) {
134 surf_action_ = hosts_.front()->pimpl_cpu->execution_start(flops_amounts_.front());
135 surf_action_->set_sharing_penalty(sharing_penalty_);
136 surf_action_->set_category(get_tracing_category());
139 surf_action_->set_bound(bound_);
141 surf_action_ = surf_host_model->execute_parallel(hosts_, flops_amounts_.data(), bytes_amounts_.data(), -1);
143 surf_action_->set_activity(this);
146 XBT_DEBUG("Create execute synchro %p: %s", this, get_cname());
150 double ExecImpl::get_seq_remaining_ratio()
152 return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains() / surf_action_->get_cost();
155 double ExecImpl::get_par_remaining_ratio()
157 // parallel task: their remain is already between 0 and 1
158 return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains();
161 ExecImpl& ExecImpl::set_bound(double bound)
167 ExecImpl& ExecImpl::set_sharing_penalty(double sharing_penalty)
169 sharing_penalty_ = sharing_penalty;
173 void ExecImpl::post()
175 if (hosts_.size() == 1 && not hosts_.front()->is_on()) { /* FIXME: handle resource failure for parallel tasks too */
176 /* If the host running the synchro failed, notice it. This way, the asking
177 * process can be killed if it runs on that host itself */
178 state_ = State::FAILED;
179 } else if (surf_action_ && surf_action_->get_state() == resource::Action::State::FAILED) {
180 /* If the host running the synchro didn't fail, then the synchro was canceled */
181 state_ = State::CANCELED;
182 } else if (timeout_detector_ && timeout_detector_->get_state() == resource::Action::State::FINISHED) {
183 state_ = State::TIMEOUT;
185 state_ = State::DONE;
189 timeout_detector_.reset();
190 /* Answer all simcalls associated with the synchro */
194 void ExecImpl::finish()
196 while (not simcalls_.empty()) {
197 smx_simcall_t simcall = simcalls_.front();
198 simcalls_.pop_front();
200 /* If a waitany simcall is waiting for this synchro to finish, then remove it from the other synchros in the waitany
201 * list. Afterwards, get the position of the actual synchro in the waitany list and return it as the result of the
204 if (simcall->call_ == SIMCALL_NONE) // FIXME: maybe a better way to handle this case
205 continue; // if process handling comm is killed
206 if (simcall->call_ == SIMCALL_EXECUTION_WAITANY_FOR) {
207 simgrid::kernel::activity::ExecImpl** execs = simcall_execution_waitany_for__get__execs(simcall);
208 size_t count = simcall_execution_waitany_for__get__count(simcall);
210 for (size_t i = 0; i < count; i++) {
211 // Remove the first occurrence of simcall:
212 auto* exec = execs[i];
213 auto j = boost::range::find(exec->simcalls_, simcall);
214 if (j != exec->simcalls_.end())
215 exec->simcalls_.erase(j);
217 if (simcall->timeout_cb_) {
218 simcall->timeout_cb_->remove();
219 simcall->timeout_cb_ = nullptr;
223 if (not MC_is_active() && not MC_record_replay_is_active()) {
224 ExecImpl** element = std::find(execs, execs + count, this);
225 int rank = (element != execs + count) ? element - execs : -1;
226 simcall_execution_waitany_for__set__result(simcall, rank);
232 /* do nothing, synchro done */
233 XBT_DEBUG("ExecImpl::finish(): execution successful");
237 XBT_DEBUG("ExecImpl::finish(): host '%s' failed", simcall->issuer_->get_host()->get_cname());
238 simcall->issuer_->context_->set_wannadie();
239 if (simcall->issuer_->get_host()->is_on())
240 simcall->issuer_->exception_ =
241 std::make_exception_ptr(simgrid::HostFailureException(XBT_THROW_POINT, "Host failed"));
242 /* else, the actor will be killed with no possibility to survive */
245 case State::CANCELED:
246 XBT_DEBUG("ExecImpl::finish(): execution canceled");
247 simcall->issuer_->exception_ =
248 std::make_exception_ptr(simgrid::CancelException(XBT_THROW_POINT, "Execution Canceled"));
252 XBT_DEBUG("ExecImpl::finish(): execution timeouted");
253 simcall->issuer_->exception_ = std::make_exception_ptr(simgrid::TimeoutException(XBT_THROW_POINT, "Timeouted"));
257 xbt_die("Internal error in ExecImpl::finish(): unexpected synchro state %d", static_cast<int>(state_));
260 simcall->issuer_->waiting_synchro = nullptr;
261 /* Fail the process if the host is down */
262 if (simcall->issuer_->get_host()->is_on())
263 simcall->issuer_->simcall_answer();
265 simcall->issuer_->context_->set_wannadie();
269 ActivityImpl* ExecImpl::migrate(s4u::Host* to)
271 if (not MC_is_active() && not MC_record_replay_is_active()) {
272 resource::Action* old_action = this->surf_action_;
273 resource::Action* new_action = to->pimpl_cpu->execution_start(old_action->get_cost());
274 new_action->set_remains(old_action->get_remains());
275 new_action->set_activity(this);
276 new_action->set_sharing_penalty(old_action->get_sharing_penalty());
278 // FIXME: the user-defined bound seem to not be kept by LMM, that seem to overwrite it for the multi-core modeling.
279 // I hope that the user did not provide any.
281 old_action->set_activity(nullptr);
282 old_action->cancel();
284 this->surf_action_ = new_action;
287 on_migration(*this, to);
294 xbt::signal<void(ExecImpl const&, s4u::Host*)> ExecImpl::on_migration;
296 } // namespace activity
297 } // namespace kernel
298 } // namespace simgrid