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 synchro->wait_for(simcall->issuer_, timeout);
25 void simcall_HANDLER_execution_waitany_for(smx_simcall_t simcall, simgrid::kernel::activity::ExecImpl* execs[],
26 size_t count, double timeout)
29 simcall->timeout_cb_ = nullptr;
31 simcall->timeout_cb_ = simgrid::simix::Timer::set(SIMIX_get_clock() + timeout, [simcall, execs, count]() {
32 for (size_t i = 0; i < count; i++) {
33 // Remove the first occurrence of simcall:
34 auto* exec = execs[i];
35 auto j = boost::range::find(exec->simcalls_, simcall);
36 if (j != exec->simcalls_.end())
37 exec->simcalls_.erase(j);
39 simcall_execution_waitany_for__set__result(simcall, -1);
40 simcall->issuer_->simcall_answer();
44 for (size_t i = 0; i < count; i++) {
45 /* associate this simcall to the the synchro */
46 auto* exec = execs[i];
47 exec->simcalls_.push_back(simcall);
49 /* see if the synchro is already finished */
50 if (exec->state_ != simgrid::kernel::activity::State::WAITING &&
51 exec->state_ != simgrid::kernel::activity::State::RUNNING) {
62 ExecImpl& ExecImpl::set_host(s4u::Host* host)
64 hosts_.assign(1, host);
68 ExecImpl& ExecImpl::set_hosts(const std::vector<s4u::Host*>& hosts)
74 ExecImpl& ExecImpl::set_timeout(double timeout)
76 if (timeout > 0 && not MC_is_active() && not MC_record_replay_is_active()) {
77 timeout_detector_.reset(hosts_.front()->pimpl_cpu->sleep(timeout));
78 timeout_detector_->set_activity(this);
83 ExecImpl& ExecImpl::set_flops_amount(double flops_amount)
85 flops_amounts_.assign(1, flops_amount);
89 ExecImpl& ExecImpl::set_flops_amounts(const std::vector<double>& flops_amounts)
91 flops_amounts_ = flops_amounts;
95 ExecImpl& ExecImpl::set_bytes_amounts(const std::vector<double>& bytes_amounts)
97 bytes_amounts_ = bytes_amounts;
102 ExecImpl* ExecImpl::start()
104 state_ = State::RUNNING;
105 if (not MC_is_active() && not MC_record_replay_is_active()) {
106 if (hosts_.size() == 1) {
107 surf_action_ = hosts_.front()->pimpl_cpu->execution_start(flops_amounts_.front());
108 surf_action_->set_sharing_penalty(sharing_penalty_);
109 surf_action_->set_category(get_tracing_category());
112 surf_action_->set_bound(bound_);
114 surf_action_ = surf_host_model->execute_parallel(hosts_, flops_amounts_.data(), bytes_amounts_.data(), -1);
116 surf_action_->set_activity(this);
119 XBT_DEBUG("Create execute synchro %p: %s", this, get_cname());
123 void ExecImpl::wait_for(actor::ActorImpl* issuer, double timeout)
125 XBT_DEBUG("Wait for execution of synchro %p, state %d", this, (int)state_);
126 xbt_assert(std::isfinite(timeout), "timeout is not finite!");
128 /* Associate this simcall to the synchro */
129 register_simcall(&issuer->simcall);
131 if (MC_is_active() || MC_record_replay_is_active()) {
132 int idx = SIMCALL_GET_MC_VALUE(issuer->simcall);
134 state_ = simgrid::kernel::activity::State::DONE;
136 /* If we reached this point, the wait simcall must have a timeout */
137 /* Otherwise it shouldn't be enabled and executed by the MC */
140 state_ = simgrid::kernel::activity::State::TIMEOUT;
146 /* If the synchro is already finished then perform the error handling */
147 if (state_ != simgrid::kernel::activity::State::RUNNING) {
149 } else { /* we need a sleep action (even when there is no timeout) to be notified of host failures */
150 set_timeout(timeout);
154 double ExecImpl::get_seq_remaining_ratio()
156 return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains() / surf_action_->get_cost();
159 double ExecImpl::get_par_remaining_ratio()
161 // parallel task: their remain is already between 0 and 1
162 return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains();
165 ExecImpl& ExecImpl::set_bound(double bound)
171 ExecImpl& ExecImpl::set_sharing_penalty(double sharing_penalty)
173 sharing_penalty_ = sharing_penalty;
177 void ExecImpl::post()
179 if (hosts_.size() == 1 && not hosts_.front()->is_on()) { /* FIXME: handle resource failure for parallel tasks too */
180 /* If the host running the synchro failed, notice it. This way, the asking
181 * process can be killed if it runs on that host itself */
182 state_ = State::FAILED;
183 } else if (surf_action_ && surf_action_->get_state() == resource::Action::State::FAILED) {
184 /* If the host running the synchro didn't fail, then the synchro was canceled */
185 state_ = State::CANCELED;
186 } else if (timeout_detector_ && timeout_detector_->get_state() == resource::Action::State::FINISHED) {
187 state_ = State::TIMEOUT;
189 state_ = State::DONE;
193 timeout_detector_.reset();
194 /* Answer all simcalls associated with the synchro */
198 void ExecImpl::finish()
200 while (not simcalls_.empty()) {
201 smx_simcall_t simcall = simcalls_.front();
202 simcalls_.pop_front();
204 /* If a waitany simcall is waiting for this synchro to finish, then remove it from the other synchros in the waitany
205 * list. Afterwards, get the position of the actual synchro in the waitany list and return it as the result of the
208 if (simcall->call_ == SIMCALL_NONE) // FIXME: maybe a better way to handle this case
209 continue; // if process handling comm is killed
210 if (simcall->call_ == SIMCALL_EXECUTION_WAITANY_FOR) {
211 simgrid::kernel::activity::ExecImpl** execs = simcall_execution_waitany_for__get__execs(simcall);
212 size_t count = simcall_execution_waitany_for__get__count(simcall);
214 for (size_t i = 0; i < count; i++) {
215 // Remove the first occurrence of simcall:
216 auto* exec = execs[i];
217 auto j = boost::range::find(exec->simcalls_, simcall);
218 if (j != exec->simcalls_.end())
219 exec->simcalls_.erase(j);
221 if (simcall->timeout_cb_) {
222 simcall->timeout_cb_->remove();
223 simcall->timeout_cb_ = nullptr;
227 if (not MC_is_active() && not MC_record_replay_is_active()) {
228 ExecImpl** element = std::find(execs, execs + count, this);
229 int rank = (element != execs + count) ? element - execs : -1;
230 simcall_execution_waitany_for__set__result(simcall, rank);
236 /* do nothing, synchro done */
237 XBT_DEBUG("ExecImpl::finish(): execution successful");
241 XBT_DEBUG("ExecImpl::finish(): host '%s' failed", simcall->issuer_->get_host()->get_cname());
242 simcall->issuer_->context_->set_wannadie();
243 if (simcall->issuer_->get_host()->is_on())
244 simcall->issuer_->exception_ =
245 std::make_exception_ptr(simgrid::HostFailureException(XBT_THROW_POINT, "Host failed"));
246 /* else, the actor will be killed with no possibility to survive */
249 case State::CANCELED:
250 XBT_DEBUG("ExecImpl::finish(): execution canceled");
251 simcall->issuer_->exception_ =
252 std::make_exception_ptr(simgrid::CancelException(XBT_THROW_POINT, "Execution Canceled"));
256 XBT_DEBUG("ExecImpl::finish(): execution timeouted");
257 simcall->issuer_->exception_ = std::make_exception_ptr(simgrid::TimeoutException(XBT_THROW_POINT, "Timeouted"));
261 xbt_die("Internal error in ExecImpl::finish(): unexpected synchro state %d", static_cast<int>(state_));
264 simcall->issuer_->waiting_synchro = nullptr;
265 /* Fail the process if the host is down */
266 if (simcall->issuer_->get_host()->is_on())
267 simcall->issuer_->simcall_answer();
269 simcall->issuer_->context_->set_wannadie();
273 ActivityImpl* ExecImpl::migrate(s4u::Host* to)
275 if (not MC_is_active() && not MC_record_replay_is_active()) {
276 resource::Action* old_action = this->surf_action_;
277 resource::Action* new_action = to->pimpl_cpu->execution_start(old_action->get_cost());
278 new_action->set_remains(old_action->get_remains());
279 new_action->set_activity(this);
280 new_action->set_sharing_penalty(old_action->get_sharing_penalty());
282 // FIXME: the user-defined bound seem to not be kept by LMM, that seem to overwrite it for the multi-core modeling.
283 // I hope that the user did not provide any.
285 old_action->set_activity(nullptr);
286 old_action->cancel();
288 this->surf_action_ = new_action;
291 on_migration(*this, to);
298 xbt::signal<void(ExecImpl const&, s4u::Host*)> ExecImpl::on_migration;
300 } // namespace activity
301 } // namespace kernel
302 } // namespace simgrid