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/surf/HostImpl.hpp"
19 #include "src/surf/cpu_interface.hpp"
21 #include <boost/range/algorithm.hpp>
23 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_process, simix, "Logging specific to SIMIX (process)");
25 static unsigned long simix_process_maxpid = 0;
28 * @brief Returns the current agent.
30 * This functions returns the currently running SIMIX process.
32 * @return The SIMIX process
34 smx_actor_t SIMIX_process_self()
36 smx_context_t self_context = simgrid::kernel::context::Context::self();
38 return (self_context != nullptr) ? self_context->get_actor() : nullptr;
42 * @brief Returns whether a process has pending asynchronous communications.
43 * @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.")));
86 ActorImpl* actor = new ActorImpl(xbt::string(name), host);
88 actor->set_user_data(data);
89 actor->code = nullptr;
91 XBT_VERB("Create context %s", actor->get_cname());
92 xbt_assert(simix_global != nullptr, "simix is not initialized, please call MSG_init first");
93 actor->context_ = simix_global->context_factory->attach(actor);
96 if (properties != nullptr)
97 for (auto const& kv : *properties)
98 actor->set_property(kv.first, kv.second);
100 /* Add the process to it's host process list */
101 host->pimpl_->process_list_.push_back(*actor);
103 /* Now insert it in the global process list and in the process to run list */
104 simix_global->process_list[actor->get_pid()] = actor;
105 XBT_DEBUG("Inserting [%p] %s(%s) in the to_run list", actor, actor->get_cname(), host->get_cname());
106 simix_global->actors_to_run.push_back(actor);
107 intrusive_ptr_add_ref(actor);
109 auto* context = dynamic_cast<simgrid::kernel::context::AttachContext*>(actor->context_);
110 xbt_assert(nullptr != context, "Not a suitable context");
111 context->attach_start();
113 /* The on_creation() signal must be delayed until there, where the pid and everything is set */
114 simgrid::s4u::ActorPtr tmp = actor->iface(); // Passing this directly to on_creation will lead to crashes
115 simgrid::s4u::Actor::on_creation(tmp);
117 return ActorImplPtr(actor);
119 /** @brief Detach an actor attached with `attach()`
121 * This is called when the current actor has finished its job.
122 * Used in the main thread, it waits for the simulation to finish before returning. When it returns, the other
123 * simulated actors and the maestro are destroyed.
125 void ActorImpl::detach()
127 auto* context = dynamic_cast<context::AttachContext*>(context::Context::self());
128 if (context == nullptr)
129 xbt_die("Not a suitable context");
131 context->get_actor()->cleanup();
132 context->attach_stop();
135 void ActorImpl::cleanup()
137 if (this == simix_global->maestro_process) /* Do not cleanup maestro */
140 XBT_DEBUG("Cleanup actor %s (%p), waiting synchro %p", get_cname(), this, waiting_synchro.get());
142 simix_global->mutex.lock();
144 simix_global->process_list.erase(pid_);
145 if (host_ && host_process_list_hook.is_linked())
146 simgrid::xbt::intrusive_erase(host_->pimpl_->process_list_, *this);
147 if (not smx_destroy_list_hook.is_linked()) {
149 xbt_dynar_push_as(simix_global->dead_actors_vector, smx_actor_t, this);
151 simix_global->actors_to_destroy.push_back(*this);
153 context_->iwannadie = false;
155 simix_global->mutex.unlock();
158 void ActorImpl::exit()
160 context_->iwannadie = true;
163 exception_ = nullptr;
165 // Forcefully kill the actor if its host is turned off. Not a HostFailureException because you should not survive that
166 if (not host_->is_on())
167 this->throw_exception(std::make_exception_ptr(ForcefulKillException("host failed")));
169 /* destroy the blocking synchro if any */
170 if (waiting_synchro != nullptr) {
172 activity::ExecImplPtr exec = boost::dynamic_pointer_cast<activity::ExecImpl>(waiting_synchro);
173 activity::CommImplPtr comm = boost::dynamic_pointer_cast<activity::CommImpl>(waiting_synchro);
174 activity::SleepImplPtr sleep = boost::dynamic_pointer_cast<activity::SleepImpl>(waiting_synchro);
175 activity::RawImplPtr raw = boost::dynamic_pointer_cast<activity::RawImpl>(waiting_synchro);
176 activity::IoImplPtr io = boost::dynamic_pointer_cast<activity::IoImpl>(waiting_synchro);
178 if (exec != nullptr && exec->surf_action_) {
180 exec->surf_action_->unref();
181 exec->surf_action_ = nullptr;
182 } else if (comm != nullptr) {
183 comms.remove(waiting_synchro);
185 // Remove first occurrence of &actor->simcall:
186 auto i = boost::range::find(waiting_synchro->simcalls_, &simcall);
187 if (i != waiting_synchro->simcalls_.end())
188 waiting_synchro->simcalls_.remove(&simcall);
189 } else if (sleep != nullptr) {
190 if (sleep->surf_action_)
191 sleep->surf_action_->cancel();
193 } else if (raw != nullptr) {
195 } else if (io != nullptr) {
198 simgrid::kernel::activity::ActivityImplPtr activity = waiting_synchro;
199 xbt_die("Activity %s is of unknown type %s", activity->get_cname(),
200 simgrid::xbt::demangle(typeid(activity).name()).get());
203 waiting_synchro = nullptr;
207 void ActorImpl::kill(ActorImpl* actor)
209 if (actor->finished_) {
210 XBT_DEBUG("Ignoring request to kill actor %s@%s that is already dead", actor->get_cname(),
211 actor->host_->get_cname());
215 XBT_DEBUG("Actor '%s'@%s is killing actor '%s'@%s", get_cname(), host_ ? host_->get_cname() : "", actor->get_cname(),
216 actor->host_ ? actor->host_->get_cname() : "");
220 if (std::find(begin(simix_global->actors_to_run), end(simix_global->actors_to_run), actor) ==
221 end(simix_global->actors_to_run) &&
223 XBT_DEBUG("Inserting %s in the to_run list", actor->get_cname());
224 simix_global->actors_to_run.push_back(actor);
228 void ActorImpl::kill_all()
230 for (auto const& kv : simix_global->process_list)
231 if (kv.second != this)
232 this->kill(kv.second);
235 void ActorImpl::set_kill_time(double kill_time)
237 if (kill_time <= SIMIX_get_clock())
239 XBT_DEBUG("Set kill time %f for actor %s@%s", kill_time, get_cname(), host_->get_cname());
240 kill_timer = simix::Timer::set(kill_time, [this] {
242 kill_timer = nullptr;
246 double ActorImpl::get_kill_time()
248 return kill_timer ? kill_timer->get_date() : 0;
251 static void dying_daemon(int /*exit_status*/, void* data)
253 std::vector<ActorImpl*>* vect = &simix_global->daemons;
255 auto it = std::find(vect->begin(), vect->end(), static_cast<ActorImpl*>(data));
256 xbt_assert(it != vect->end(), "The dying daemon is not a daemon after all. Please report that bug.");
258 /* Don't move the whole content since we don't really care about the order */
259 std::swap(*it, vect->back());
263 void ActorImpl::yield()
265 XBT_DEBUG("Yield actor '%s'", get_cname());
267 /* Go into sleep and return control to maestro */
270 /* Ok, maestro returned control to us */
271 XBT_DEBUG("Control returned to me: '%s'", get_cname());
273 if (context_->iwannadie) {
275 XBT_DEBUG("Actor %s@%s is dead", get_cname(), host_->get_cname());
276 // throw simgrid::kernel::context::ForcefulKillException(); Does not seem to properly kill the actor
282 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", ""))->set_host(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::init(std::string name, s4u::Host* host)
449 ActorImpl* actor = new ActorImpl(simgrid::xbt::string(name), host);
450 actor->set_ppid(this->pid_);
452 intrusive_ptr_add_ref(actor);
453 /* The on_creation() signal must be delayed until there, where the pid and everything is set */
454 s4u::Actor::on_creation(actor->iface());
456 return ActorImplPtr(actor);
459 ActorImpl* ActorImpl::start(simix::ActorCode code)
461 xbt_assert(code && host_ != nullptr, "Invalid parameters");
463 if (not host_->is_on()) {
464 intrusive_ptr_release(this);
465 XBT_WARN("Cannot launch actor '%s' on failed host '%s'", name_.c_str(), host_->get_cname());
466 std::rethrow_exception(
467 std::make_exception_ptr(simgrid::HostFailureException(XBT_THROW_POINT, "Cannot start actor on failed host.")));
471 XBT_VERB("Create context %s", get_cname());
472 context_ = simix_global->context_factory->create_context(std::move(code), this);
474 XBT_DEBUG("Start context '%s'", get_cname());
476 /* Add the actor to its host's actor list */
477 host_->pimpl_->process_list_.push_back(*this);
478 simix_global->process_list[pid_] = this;
480 /* Now insert it in the global actor list and in the actor to run list */
481 XBT_DEBUG("Inserting [%p] %s(%s) in the to_run list", this, get_cname(), host_->get_cname());
482 simix_global->actors_to_run.push_back(this);
487 ActorImplPtr ActorImpl::create(std::string name, simix::ActorCode code, void* data, s4u::Host* host,
488 std::unordered_map<std::string, std::string>* properties, ActorImpl* parent_actor)
490 XBT_DEBUG("Start actor %s@'%s'", name.c_str(), host->get_cname());
493 if (parent_actor != nullptr)
494 actor = parent_actor->init(simgrid::xbt::string(name), host);
496 actor = SIMIX_process_self()->init(simgrid::xbt::string(name), host);
499 actor->set_user_data(data);
502 if (properties != nullptr)
503 for (auto const& kv : *properties)
504 actor->set_property(kv.first, kv.second);
506 actor->start(std::move(code));
511 void create_maestro(simix::ActorCode code)
513 /* Create maestro actor and initialize it */
514 ActorImpl* maestro = new ActorImpl(xbt::string(""), /*host*/ nullptr);
517 maestro->context_ = simix_global->context_factory->create_context(simix::ActorCode(), maestro);
519 maestro->context_ = simix_global->context_factory->create_maestro(code, maestro);
522 maestro->simcall.issuer = maestro;
523 simix_global->maestro_process = maestro;
527 } // namespace kernel
530 void SIMIX_process_detach()
532 simgrid::kernel::actor::ActorImpl::detach();
535 smx_actor_t SIMIX_process_attach(const char* name, void* data, const char* hostname,
536 std::unordered_map<std::string, std::string>* properties,
537 smx_actor_t /*parent_process*/)
539 return simgrid::kernel::actor::ActorImpl::attach(name, data, sg_host_by_name(hostname), properties).get();
542 /** @deprecated When this function gets removed, also remove the xbt_ex class, that is only there to help users to
544 void SIMIX_process_throw(smx_actor_t actor, xbt_errcat_t cat, int value, const char* msg)
546 SMX_EXCEPTION(actor, cat, value, msg);
548 if (actor->is_suspended())
551 /* cancel the blocking synchro if any */
552 if (actor->waiting_synchro) {
554 simgrid::kernel::activity::ExecImplPtr exec =
555 boost::dynamic_pointer_cast<simgrid::kernel::activity::ExecImpl>(actor->waiting_synchro);
559 simgrid::kernel::activity::CommImplPtr comm =
560 boost::dynamic_pointer_cast<simgrid::kernel::activity::CommImpl>(actor->waiting_synchro);
561 if (comm != nullptr) {
562 actor->comms.remove(comm);
566 simgrid::kernel::activity::SleepImplPtr sleep =
567 boost::dynamic_pointer_cast<simgrid::kernel::activity::SleepImpl>(actor->waiting_synchro);
568 if (sleep != nullptr) {
569 SIMIX_process_sleep_destroy(actor->waiting_synchro);
570 if (std::find(begin(simix_global->actors_to_run), end(simix_global->actors_to_run), actor) ==
571 end(simix_global->actors_to_run) &&
572 actor != SIMIX_process_self()) {
573 XBT_DEBUG("Inserting [%p] %s in the to_run list", actor, actor->get_cname());
574 simix_global->actors_to_run.push_back(actor);
578 simgrid::kernel::activity::RawImplPtr raw =
579 boost::dynamic_pointer_cast<simgrid::kernel::activity::RawImpl>(actor->waiting_synchro);
580 if (raw != nullptr) {
584 simgrid::kernel::activity::IoImplPtr io =
585 boost::dynamic_pointer_cast<simgrid::kernel::activity::IoImpl>(actor->waiting_synchro);
590 actor->waiting_synchro = nullptr;
593 void simcall_HANDLER_process_suspend(smx_simcall_t simcall, smx_actor_t actor)
595 smx_activity_t sync_suspend = actor->suspend(simcall->issuer);
597 if (actor != simcall->issuer) {
598 SIMIX_simcall_answer(simcall);
600 sync_suspend->simcalls_.push_back(simcall);
601 actor->waiting_synchro = sync_suspend;
602 actor->waiting_synchro->suspend();
604 /* If we are suspending ourselves, then just do not finish the simcall now */
607 int SIMIX_process_get_maxpid() {
608 return simix_process_maxpid;
611 int SIMIX_process_count()
613 return simix_global->process_list.size();
616 void* SIMIX_process_self_get_data() // deprecated
618 smx_actor_t self = SIMIX_process_self();
620 if (self == nullptr) {
623 return self->get_user_data();
626 void SIMIX_process_self_set_data(void* data) // deprecated
628 SIMIX_process_self()->set_user_data(data);
632 /* needs to be public and without simcall because it is called
633 by exceptions and logging events */
634 const char* SIMIX_process_self_get_name() {
636 smx_actor_t process = SIMIX_process_self();
637 if (process == nullptr || process == simix_global->maestro_process)
640 return process->get_cname();
643 void simcall_HANDLER_process_join(smx_simcall_t simcall, smx_actor_t process, double timeout)
645 if (process->finished_) {
646 // The joined process is already finished, just wake up the issuer process right away
647 simcall_process_sleep__set__result(simcall, SIMIX_DONE);
648 SIMIX_simcall_answer(simcall);
651 smx_activity_t sync = simcall->issuer->join(process, timeout);
652 sync->simcalls_.push_back(simcall);
653 simcall->issuer->waiting_synchro = sync;
656 void simcall_HANDLER_process_sleep(smx_simcall_t simcall, double duration)
658 if (MC_is_active() || MC_record_replay_is_active()) {
659 MC_process_clock_add(simcall->issuer, duration);
660 simcall_process_sleep__set__result(simcall, SIMIX_DONE);
661 SIMIX_simcall_answer(simcall);
664 smx_activity_t sync = simcall->issuer->sleep(duration);
665 sync->simcalls_.push_back(simcall);
666 simcall->issuer->waiting_synchro = sync;
669 void SIMIX_process_sleep_destroy(smx_activity_t synchro)
671 XBT_DEBUG("Destroy sleep synchro %p", synchro.get());
672 simgrid::kernel::activity::SleepImplPtr sleep =
673 boost::static_pointer_cast<simgrid::kernel::activity::SleepImpl>(synchro);
675 if (sleep->surf_action_) {
676 sleep->surf_action_->unref();
677 sleep->surf_action_ = nullptr;
682 * @brief Calling this function makes the process to yield.
684 * Only the current process can call this function, giving back the control to maestro.
686 * @param self the current process
689 /** @brief Returns the list of processes to run.
692 const std::vector<smx_actor_t>& simgrid::simix::process_get_runnable()
694 return simix_global->actors_to_run;
697 /** @brief Returns the process from PID. */
698 smx_actor_t SIMIX_process_from_PID(aid_t PID)
700 auto actor = simix_global->process_list.find(PID);
701 return actor == simix_global->process_list.end() ? nullptr : actor->second;
704 void SIMIX_process_on_exit(smx_actor_t actor, int_f_pvoid_pvoid_t fun, void* data)
706 SIMIX_process_on_exit(actor, [fun](int a, void* b) { fun((void*)(intptr_t)a, b); }, data);
709 void SIMIX_process_on_exit(smx_actor_t actor, std::function<void(bool, void*)> fun, void* data)
711 xbt_assert(actor, "current process not found: are you in maestro context ?");
713 actor->on_exit.emplace_back(s_smx_process_exit_fun_t{fun, data});
716 /** @brief Restart a process, starting it again from the beginning. */
718 * @ingroup simix_process_management
719 * @brief Creates and runs a new SIMIX process.
721 * The structure and the corresponding thread are created and put in the list of ready processes.
723 * @param name a name for the process. It is for user-level information and can be nullptr.
724 * @param code the main function of the process
725 * @param data a pointer to any data one may want to attach to the new object. It is for user-level information and can
727 * It can be retrieved with the method ActorImpl::getUserData().
728 * @param host where the new agent is executed.
729 * @param properties the properties of the process
731 smx_actor_t simcall_process_create(std::string name, simgrid::simix::ActorCode code, void* data, sg_host_t host,
732 std::unordered_map<std::string, std::string>* properties)
734 smx_actor_t self = SIMIX_process_self();
735 return simgrid::simix::simcall([name, code, data, host, properties, self] {
736 return simgrid::kernel::actor::ActorImpl::create(std::move(name), std::move(code), data, host, properties, self)
741 void simcall_process_set_data(smx_actor_t process, void* data)
743 simgrid::simix::simcall([process, data] { process->set_user_data(data); });