1 /* Copyright (c) 2007-2019. 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. */
7 #include "simgrid/Exception.hpp"
8 #include "simgrid/s4u/Actor.hpp"
9 #include "simgrid/s4u/Exec.hpp"
10 #include "smx_private.hpp"
11 #include "src/kernel/activity/CommImpl.hpp"
12 #include "src/kernel/activity/ExecImpl.hpp"
13 #include "src/kernel/activity/IoImpl.hpp"
14 #include "src/kernel/activity/SleepImpl.hpp"
15 #include "src/kernel/activity/SynchroRaw.hpp"
16 #include "src/mc/mc_replay.hpp"
17 #include "src/mc/remote/Client.hpp"
18 #include "src/simix/smx_host_private.hpp"
19 #include "src/surf/HostImpl.hpp"
20 #include "src/surf/cpu_interface.hpp"
22 #include <boost/range/algorithm.hpp>
24 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_process, simix, "Logging specific to SIMIX (process)");
26 static unsigned long simix_process_maxpid = 0;
29 * @brief Returns the current agent.
31 * This functions returns the currently running SIMIX process.
33 * @return The SIMIX process
35 smx_actor_t SIMIX_process_self()
37 smx_context_t self_context = simgrid::kernel::context::Context::self();
39 return (self_context != nullptr) ? self_context->get_actor() : nullptr;
43 * @brief Returns whether a process has pending asynchronous communications.
44 * @return true if there are asynchronous communications in this process
46 int SIMIX_process_has_pending_comms(smx_actor_t process) {
48 return process->comms.size() > 0;
55 ActorImpl::ActorImpl(simgrid::xbt::string name, s4u::Host* host) : host_(host), name_(name), piface_(this)
57 pid_ = simix_process_maxpid++;
58 simcall.issuer = this;
61 ActorImpl::~ActorImpl()
63 delete this->context_;
65 /* Become an actor in the simulation
67 * Currently this can only be called by the main thread (once) and only work with some thread factories
68 * (currently ThreadContextFactory).
70 * In the future, it might be extended in order to attach other threads created by a third party library.
73 ActorImplPtr ActorImpl::attach(std::string name, void* data, s4u::Host* host,
74 std::unordered_map<std::string, std::string>* properties)
76 // This is mostly a copy/paste from create(), it'd be nice to share some code between those two functions.
78 XBT_DEBUG("Attach process %s on host '%s'", name.c_str(), host->get_cname());
80 if (not host->is_on()) {
81 XBT_WARN("Cannot launch process '%s' on failed host '%s'", name.c_str(), host->get_cname());
82 std::rethrow_exception(
83 std::make_exception_ptr(simgrid::HostFailureException(XBT_THROW_POINT, "Cannot attach actor on failed host.")));
87 ActorImpl* actor = new ActorImpl(xbt::string(name), host);
89 actor->set_user_data(data);
90 actor->code = nullptr;
92 XBT_VERB("Create context %s", actor->get_cname());
93 xbt_assert(simix_global != nullptr, "simix is not initialized, please call MSG_init first");
94 actor->context_ = simix_global->context_factory->attach(actor);
97 if (properties != nullptr)
98 for (auto const& kv : *properties)
99 actor->set_property(kv.first, kv.second);
101 /* Add the process to it's host process list */
102 host->pimpl_->process_list_.push_back(*actor);
104 /* Now insert it in the global process list and in the process to run list */
105 simix_global->process_list[actor->get_pid()] = actor;
106 XBT_DEBUG("Inserting [%p] %s(%s) in the to_run list", actor, actor->get_cname(), host->get_cname());
107 simix_global->actors_to_run.push_back(actor);
108 intrusive_ptr_add_ref(actor);
110 auto* context = dynamic_cast<simgrid::kernel::context::AttachContext*>(actor->context_);
111 xbt_assert(nullptr != context, "Not a suitable context");
112 context->attach_start();
114 /* The on_creation() signal must be delayed until there, where the pid and everything is set */
115 simgrid::s4u::ActorPtr tmp = actor->iface(); // Passing this directly to on_creation will lead to crashes
116 simgrid::s4u::Actor::on_creation(tmp);
118 return ActorImplPtr(actor);
120 /** @brief Detach an actor attached with `attach()`
122 * This is called when the current actor has finished its job.
123 * Used in the main thread, it waits for the simulation to finish before returning. When it returns, the other
124 * simulated actors and the maestro are destroyed.
126 void ActorImpl::detach()
128 auto* context = dynamic_cast<context::AttachContext*>(context::Context::self());
129 if (context == nullptr)
130 xbt_die("Not a suitable context");
132 context->get_actor()->cleanup();
133 context->attach_stop();
136 void ActorImpl::cleanup()
138 if (this == simix_global->maestro_process) /* Do not cleanup maestro */
141 XBT_DEBUG("Cleanup actor %s (%p), waiting synchro %p", get_cname(), this, waiting_synchro.get());
143 simix_global->mutex.lock();
145 simix_global->process_list.erase(pid_);
146 if (host_ && host_process_list_hook.is_linked())
147 simgrid::xbt::intrusive_erase(host_->pimpl_->process_list_, *this);
148 if (not smx_destroy_list_hook.is_linked()) {
150 xbt_dynar_push_as(simix_global->dead_actors_vector, smx_actor_t, this);
152 simix_global->actors_to_destroy.push_back(*this);
154 context_->iwannadie = false;
156 simix_global->mutex.unlock();
159 void ActorImpl::exit()
161 context_->iwannadie = true;
164 exception_ = nullptr;
166 // Forcefully kill the actor if its host is turned off. Not a HostFailureException because you should not survive that
167 if (not host_->is_on())
168 this->throw_exception(std::make_exception_ptr(simgrid::kernel::context::StopRequest("host failed")));
170 /* destroy the blocking synchro if any */
171 if (waiting_synchro != nullptr) {
173 activity::ExecImplPtr exec = boost::dynamic_pointer_cast<activity::ExecImpl>(waiting_synchro);
174 activity::CommImplPtr comm = boost::dynamic_pointer_cast<activity::CommImpl>(waiting_synchro);
175 activity::SleepImplPtr sleep = boost::dynamic_pointer_cast<activity::SleepImpl>(waiting_synchro);
176 activity::RawImplPtr raw = boost::dynamic_pointer_cast<activity::RawImpl>(waiting_synchro);
177 activity::IoImplPtr io = boost::dynamic_pointer_cast<activity::IoImpl>(waiting_synchro);
179 if (exec != nullptr && exec->surf_action_) {
181 exec->surf_action_->unref();
182 exec->surf_action_ = nullptr;
183 } else if (comm != nullptr) {
184 comms.remove(waiting_synchro);
186 // Remove first occurrence of &actor->simcall:
187 auto i = boost::range::find(waiting_synchro->simcalls_, &simcall);
188 if (i != waiting_synchro->simcalls_.end())
189 waiting_synchro->simcalls_.remove(&simcall);
190 } else if (sleep != nullptr) {
191 if (sleep->surf_action_)
192 sleep->surf_action_->cancel();
194 } else if (raw != nullptr) {
196 } else if (io != nullptr) {
199 simgrid::kernel::activity::ActivityImplPtr activity = waiting_synchro;
200 xbt_die("Activity %s is of unknown type %s", activity->get_cname(),
201 simgrid::xbt::demangle(typeid(activity).name()).get());
204 waiting_synchro = nullptr;
208 void ActorImpl::kill(ActorImpl* actor)
210 if (actor->finished_) {
211 XBT_DEBUG("Ignoring request to kill actor %s@%s that is already dead", actor->get_cname(),
212 actor->host_->get_cname());
216 XBT_DEBUG("Actor '%s'@%s is killing actor '%s'@%s", get_cname(), host_ ? host_->get_cname() : "", actor->get_cname(),
217 actor->host_ ? actor->host_->get_cname() : "");
221 if (std::find(begin(simix_global->actors_to_run), end(simix_global->actors_to_run), actor) ==
222 end(simix_global->actors_to_run) &&
224 XBT_DEBUG("Inserting %s in the to_run list", actor->get_cname());
225 simix_global->actors_to_run.push_back(actor);
229 void ActorImpl::kill_all()
231 for (auto const& kv : simix_global->process_list)
232 if (kv.second != this)
233 this->kill(kv.second);
236 void ActorImpl::set_kill_time(double kill_time)
238 if (kill_time <= SIMIX_get_clock())
240 XBT_DEBUG("Set kill time %f for actor %s@%s", kill_time, get_cname(), host_->get_cname());
241 kill_timer = simix::Timer::set(kill_time, [this] {
243 kill_timer = nullptr;
247 double ActorImpl::get_kill_time()
249 return kill_timer ? kill_timer->get_date() : 0;
252 static void dying_daemon(int /*exit_status*/, void* data)
254 std::vector<ActorImpl*>* vect = &simix_global->daemons;
256 auto it = std::find(vect->begin(), vect->end(), static_cast<ActorImpl*>(data));
257 xbt_assert(it != vect->end(), "The dying daemon is not a daemon after all. Please report that bug.");
259 /* Don't move the whole content since we don't really care about the order */
260 std::swap(*it, vect->back());
264 void ActorImpl::yield()
266 XBT_DEBUG("Yield actor '%s'", get_cname());
268 /* Go into sleep and return control to maestro */
271 /* Ok, maestro returned control to us */
272 XBT_DEBUG("Control returned to me: '%s'", get_cname());
274 if (context_->iwannadie) {
276 XBT_DEBUG("Actor %s@%s is dead", get_cname(), host_->get_cname());
277 // throw simgrid::kernel::context::StopRequest(); Does not seem to properly kill the actor
283 XBT_DEBUG("Hey! I'm suspended.");
284 xbt_assert(exception_ != nullptr, "Gasp! This exception may be lost by subsequent calls.");
289 if (exception_ != nullptr) {
290 XBT_DEBUG("Wait, maestro left me an exception");
291 std::exception_ptr exception = std::move(exception_);
292 exception_ = nullptr;
293 std::rethrow_exception(std::move(exception));
296 if (SMPI_switch_data_segment && not finished_) {
297 SMPI_switch_data_segment(iface());
301 /** This actor will be terminated automatically when the last non-daemon actor finishes */
302 void ActorImpl::daemonize()
306 simix_global->daemons.push_back(this);
307 SIMIX_process_on_exit(this, dying_daemon, this);
311 s4u::Actor* ActorImpl::restart()
313 XBT_DEBUG("Restarting actor %s on %s", get_cname(), host_->get_cname());
315 // retrieve the arguments of the old actor
316 ProcessArg arg = ProcessArg(host_, this);
318 // kill the old actor
319 (this == simix_global->maestro_process) ? this->exit() : SIMIX_process_self()->kill(this);
321 // start the new actor
323 ActorImpl::create(arg.name, std::move(arg.code), arg.data, arg.host, arg.properties.get(), nullptr);
324 actor->set_kill_time(arg.kill_time);
325 actor->set_auto_restart(arg.auto_restart);
327 return actor->ciface();
330 activity::ActivityImplPtr ActorImpl::suspend(ActorImpl* issuer)
333 XBT_DEBUG("Actor '%s' is already suspended", get_cname());
339 /* If we are suspending another actor that is waiting on a sync, suspend its synchronization. */
340 if (this != issuer) {
342 waiting_synchro->suspend();
343 /* If the other actor is not waiting, its suspension is delayed to when the actor is rescheduled. */
347 return activity::ExecImplPtr(new activity::ExecImpl("suspend", "", this->host_))->start(0.0, 1.0, 0.0);
351 void ActorImpl::resume()
353 XBT_IN("actor = %p", this);
355 if (context_->iwannadie) {
356 XBT_VERB("Ignoring request to suspend an actor that is currently dying.");
364 /* resume the synchronization that was blocking the resumed actor. */
366 waiting_synchro->resume();
371 activity::ActivityImplPtr ActorImpl::join(smx_actor_t actor, double timeout)
373 activity::ActivityImplPtr res = this->sleep(timeout);
374 intrusive_ptr_add_ref(res.get());
375 SIMIX_process_on_exit(actor,
377 auto sleep = static_cast<simgrid::kernel::activity::SleepImpl*>(arg);
378 if (sleep->surf_action_)
379 sleep->surf_action_->finish(simgrid::kernel::resource::Action::State::FINISHED);
380 intrusive_ptr_release(sleep);
386 activity::ActivityImplPtr ActorImpl::sleep(double duration)
388 if (not host_->is_on())
389 throw_exception(std::make_exception_ptr(simgrid::HostFailureException(
390 XBT_THROW_POINT, std::string("Host ") + host_->get_cname() + " failed, you cannot sleep there.")));
392 return simgrid::kernel::activity::SleepImplPtr(new simgrid::kernel::activity::SleepImpl("sleep", host_))
396 void ActorImpl::throw_exception(std::exception_ptr e)
403 /* cancel the blocking synchro if any */
404 if (waiting_synchro) {
406 activity::ExecImplPtr exec = boost::dynamic_pointer_cast<activity::ExecImpl>(waiting_synchro);
410 activity::CommImplPtr comm = boost::dynamic_pointer_cast<activity::CommImpl>(waiting_synchro);
411 if (comm != nullptr) {
416 activity::SleepImplPtr sleep = boost::dynamic_pointer_cast<activity::SleepImpl>(waiting_synchro);
417 if (sleep != nullptr) {
418 SIMIX_process_sleep_destroy(waiting_synchro);
419 if (std::find(begin(simix_global->actors_to_run), end(simix_global->actors_to_run), this) ==
420 end(simix_global->actors_to_run) &&
421 this != SIMIX_process_self()) {
422 XBT_DEBUG("Inserting [%p] %s in the to_run list", this, get_cname());
423 simix_global->actors_to_run.push_back(this);
427 activity::RawImplPtr raw = boost::dynamic_pointer_cast<activity::RawImpl>(waiting_synchro);
428 if (raw != nullptr) {
432 activity::IoImplPtr io = boost::dynamic_pointer_cast<activity::IoImpl>(waiting_synchro);
437 waiting_synchro = nullptr;
440 void ActorImpl::set_host(s4u::Host* dest)
442 simgrid::xbt::intrusive_erase(host_->pimpl_->process_list_, *this);
444 dest->pimpl_->process_list_.push_back(*this);
447 ActorImplPtr ActorImpl::create(std::string name, simix::ActorCode code, void* data, s4u::Host* host,
448 std::unordered_map<std::string, std::string>* properties, ActorImpl* parent_actor)
451 XBT_DEBUG("Start actor %s@'%s'", name.c_str(), host->get_cname());
453 if (not host->is_on()) {
454 XBT_WARN("Cannot launch actor '%s' on failed host '%s'", name.c_str(), host->get_cname());
455 std::rethrow_exception(
456 std::make_exception_ptr(simgrid::HostFailureException(XBT_THROW_POINT, "Cannot create actor on failed host.")));
460 ActorImpl* actor = new ActorImpl(simgrid::xbt::string(name), host);
462 xbt_assert(code && host != nullptr, "Invalid parameters");
464 actor->set_user_data(data);
467 if (parent_actor != nullptr)
468 actor->set_ppid(parent_actor->get_pid());
470 XBT_VERB("Create context %s", actor->get_cname());
471 actor->context_ = simix_global->context_factory->create_context(std::move(code), actor);
474 if (properties != nullptr)
475 for (auto const& kv : *properties)
476 actor->set_property(kv.first, kv.second);
478 /* Add the actor to its host's actor list */
479 host->pimpl_->process_list_.push_back(*actor);
481 XBT_DEBUG("Start context '%s'", actor->get_cname());
483 /* Now insert it in the global actor list and in the actor to run list */
484 simix_global->process_list[actor->get_pid()] = actor;
485 XBT_DEBUG("Inserting [%p] %s(%s) in the to_run list", actor, actor->get_cname(), host->get_cname());
486 simix_global->actors_to_run.push_back(actor);
487 intrusive_ptr_add_ref(actor);
489 /* The on_creation() signal must be delayed until there, where the pid and everything is set */
490 s4u::Actor::on_creation(actor->iface());
492 return ActorImplPtr(actor);
495 void create_maestro(simix::ActorCode code)
497 /* Create maestro actor and initialize it */
498 ActorImpl* maestro = new ActorImpl(xbt::string(""), /*host*/ nullptr);
501 maestro->context_ = simix_global->context_factory->create_context(simix::ActorCode(), maestro);
503 maestro->context_ = simix_global->context_factory->create_maestro(code, maestro);
506 maestro->simcall.issuer = maestro;
507 simix_global->maestro_process = maestro;
511 } // namespace kernel
514 void SIMIX_process_detach()
516 simgrid::kernel::actor::ActorImpl::detach();
519 smx_actor_t SIMIX_process_attach(const char* name, void* data, const char* hostname,
520 std::unordered_map<std::string, std::string>* properties,
521 smx_actor_t /*parent_process*/)
523 return simgrid::kernel::actor::ActorImpl::attach(name, data, sg_host_by_name(hostname), properties).get();
526 /** @deprecated When this function gets removed, also remove the xbt_ex class, that is only there to help users to
528 void SIMIX_process_throw(smx_actor_t actor, xbt_errcat_t cat, int value, const char* msg)
530 SMX_EXCEPTION(actor, cat, value, msg);
532 if (actor->is_suspended())
535 /* cancel the blocking synchro if any */
536 if (actor->waiting_synchro) {
538 simgrid::kernel::activity::ExecImplPtr exec =
539 boost::dynamic_pointer_cast<simgrid::kernel::activity::ExecImpl>(actor->waiting_synchro);
543 simgrid::kernel::activity::CommImplPtr comm =
544 boost::dynamic_pointer_cast<simgrid::kernel::activity::CommImpl>(actor->waiting_synchro);
545 if (comm != nullptr) {
546 actor->comms.remove(comm);
550 simgrid::kernel::activity::SleepImplPtr sleep =
551 boost::dynamic_pointer_cast<simgrid::kernel::activity::SleepImpl>(actor->waiting_synchro);
552 if (sleep != nullptr) {
553 SIMIX_process_sleep_destroy(actor->waiting_synchro);
554 if (std::find(begin(simix_global->actors_to_run), end(simix_global->actors_to_run), actor) ==
555 end(simix_global->actors_to_run) &&
556 actor != SIMIX_process_self()) {
557 XBT_DEBUG("Inserting [%p] %s in the to_run list", actor, actor->get_cname());
558 simix_global->actors_to_run.push_back(actor);
562 simgrid::kernel::activity::RawImplPtr raw =
563 boost::dynamic_pointer_cast<simgrid::kernel::activity::RawImpl>(actor->waiting_synchro);
564 if (raw != nullptr) {
568 simgrid::kernel::activity::IoImplPtr io =
569 boost::dynamic_pointer_cast<simgrid::kernel::activity::IoImpl>(actor->waiting_synchro);
574 actor->waiting_synchro = nullptr;
577 void simcall_HANDLER_process_suspend(smx_simcall_t simcall, smx_actor_t actor)
579 smx_activity_t sync_suspend = actor->suspend(simcall->issuer);
581 if (actor != simcall->issuer) {
582 SIMIX_simcall_answer(simcall);
584 sync_suspend->simcalls_.push_back(simcall);
585 actor->waiting_synchro = sync_suspend;
586 actor->waiting_synchro->suspend();
588 /* If we are suspending ourselves, then just do not finish the simcall now */
591 int SIMIX_process_get_maxpid() {
592 return simix_process_maxpid;
595 int SIMIX_process_count()
597 return simix_global->process_list.size();
600 void* SIMIX_process_self_get_data()
602 smx_actor_t self = SIMIX_process_self();
604 if (self == nullptr) {
607 return self->get_user_data();
610 void SIMIX_process_self_set_data(void *data)
612 SIMIX_process_self()->set_user_data(data);
616 /* needs to be public and without simcall because it is called
617 by exceptions and logging events */
618 const char* SIMIX_process_self_get_name() {
620 smx_actor_t process = SIMIX_process_self();
621 if (process == nullptr || process == simix_global->maestro_process)
624 return process->get_cname();
627 void simcall_HANDLER_process_join(smx_simcall_t simcall, smx_actor_t process, double timeout)
629 if (process->finished_) {
630 // The joined process is already finished, just wake up the issuer process right away
631 simcall_process_sleep__set__result(simcall, SIMIX_DONE);
632 SIMIX_simcall_answer(simcall);
635 smx_activity_t sync = simcall->issuer->join(process, timeout);
636 sync->simcalls_.push_back(simcall);
637 simcall->issuer->waiting_synchro = sync;
640 void simcall_HANDLER_process_sleep(smx_simcall_t simcall, double duration)
642 if (MC_is_active() || MC_record_replay_is_active()) {
643 MC_process_clock_add(simcall->issuer, duration);
644 simcall_process_sleep__set__result(simcall, SIMIX_DONE);
645 SIMIX_simcall_answer(simcall);
648 smx_activity_t sync = simcall->issuer->sleep(duration);
649 sync->simcalls_.push_back(simcall);
650 simcall->issuer->waiting_synchro = sync;
653 void SIMIX_process_sleep_destroy(smx_activity_t synchro)
655 XBT_DEBUG("Destroy sleep synchro %p", synchro.get());
656 simgrid::kernel::activity::SleepImplPtr sleep =
657 boost::static_pointer_cast<simgrid::kernel::activity::SleepImpl>(synchro);
659 if (sleep->surf_action_) {
660 sleep->surf_action_->unref();
661 sleep->surf_action_ = nullptr;
666 * @brief Calling this function makes the process to yield.
668 * Only the current process can call this function, giving back the control to maestro.
670 * @param self the current process
673 /** @brief Returns the list of processes to run.
676 const std::vector<smx_actor_t>& simgrid::simix::process_get_runnable()
678 return simix_global->actors_to_run;
681 /** @brief Returns the process from PID. */
682 smx_actor_t SIMIX_process_from_PID(aid_t PID)
684 auto actor = simix_global->process_list.find(PID);
685 return actor == simix_global->process_list.end() ? nullptr : actor->second;
688 void SIMIX_process_on_exit(smx_actor_t actor, int_f_pvoid_pvoid_t fun, void* data)
690 SIMIX_process_on_exit(actor, [fun](int a, void* b) { fun((void*)(intptr_t)a, b); }, data);
693 void SIMIX_process_on_exit(smx_actor_t actor, std::function<void(int, void*)> fun, void* data)
695 xbt_assert(actor, "current process not found: are you in maestro context ?");
697 actor->on_exit.emplace_back(s_smx_process_exit_fun_t{fun, data});
700 /** @brief Restart a process, starting it again from the beginning. */
702 * @ingroup simix_process_management
703 * @brief Creates and runs a new SIMIX process.
705 * The structure and the corresponding thread are created and put in the list of ready processes.
707 * @param name a name for the process. It is for user-level information and can be nullptr.
708 * @param code the main function of the process
709 * @param data a pointer to any data one may want to attach to the new object. It is for user-level information and can
711 * It can be retrieved with the method ActorImpl::getUserData().
712 * @param host where the new agent is executed.
713 * @param properties the properties of the process
715 smx_actor_t simcall_process_create(std::string name, simgrid::simix::ActorCode code, void* data, sg_host_t host,
716 std::unordered_map<std::string, std::string>* properties)
718 smx_actor_t self = SIMIX_process_self();
719 return simgrid::simix::simcall([name, code, data, host, properties, self] {
720 return simgrid::kernel::actor::ActorImpl::create(std::move(name), std::move(code), data, host, properties, self)
725 void simcall_process_set_data(smx_actor_t process, void* data)
727 simgrid::simix::simcall([process, data] { process->set_user_data(data); });