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 "src/kernel/activity/CommImpl.hpp"
11 #include "src/kernel/activity/ExecImpl.hpp"
12 #include "src/kernel/activity/IoImpl.hpp"
13 #include "src/kernel/activity/SleepImpl.hpp"
14 #include "src/kernel/activity/SynchroRaw.hpp"
15 #include "src/mc/mc_replay.hpp"
16 #include "src/mc/remote/Client.hpp"
17 #include "src/simix/smx_private.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)
49 return process->comms.size() > 0;
56 ActorImpl::ActorImpl(const simgrid::xbt::string& name, s4u::Host* host) : host_(host), name_(name), piface_(this)
58 pid_ = simix_process_maxpid++;
59 simcall.issuer = this;
62 ActorImpl::~ActorImpl() = default;
64 /* Become an actor in the simulation
66 * Currently this can only be called by the main thread (once) and only work with some thread factories
67 * (currently ThreadContextFactory).
69 * In the future, it might be extended in order to attach other threads created by a third party library.
72 ActorImplPtr ActorImpl::attach(const std::string& name, void* data, s4u::Host* host,
73 std::unordered_map<std::string, std::string>* properties)
75 // This is mostly a copy/paste from create(), it'd be nice to share some code between those two functions.
77 XBT_DEBUG("Attach process %s on host '%s'", name.c_str(), host->get_cname());
79 if (not host->is_on()) {
80 XBT_WARN("Cannot launch process '%s' on failed host '%s'", name.c_str(), host->get_cname());
81 std::rethrow_exception(
82 std::make_exception_ptr(simgrid::HostFailureException(XBT_THROW_POINT, "Cannot attach actor on failed host.")));
85 ActorImpl* actor = new ActorImpl(xbt::string(name), host);
87 actor->set_user_data(data);
88 actor->code = nullptr;
90 XBT_VERB("Create context %s", actor->get_cname());
91 xbt_assert(simix_global != nullptr, "simix is not initialized, please call MSG_init first");
92 actor->context_.reset(simix_global->context_factory->attach(actor));
95 if (properties != nullptr)
96 for (auto const& kv : *properties)
97 actor->set_property(kv.first, kv.second);
99 /* Add the process to it's host process list */
100 host->pimpl_->process_list_.push_back(*actor);
102 /* Now insert it in the global process list and in the process to run list */
103 simix_global->process_list[actor->get_pid()] = actor;
104 XBT_DEBUG("Inserting [%p] %s(%s) in the to_run list", actor, actor->get_cname(), host->get_cname());
105 simix_global->actors_to_run.push_back(actor);
106 intrusive_ptr_add_ref(actor);
108 auto* context = dynamic_cast<simgrid::kernel::context::AttachContext*>(actor->context_.get());
109 xbt_assert(nullptr != context, "Not a suitable context");
110 context->attach_start();
112 /* The on_creation() signal must be delayed until there, where the pid and everything is set */
113 simgrid::s4u::ActorPtr tmp = actor->iface(); // Passing this directly to on_creation will lead to crashes
114 simgrid::s4u::Actor::on_creation(tmp);
116 return ActorImplPtr(actor);
118 /** @brief Detach an actor attached with `attach()`
120 * This is called when the current actor has finished its job.
121 * Used in the main thread, it waits for the simulation to finish before returning. When it returns, the other
122 * simulated actors and the maestro are destroyed.
124 void ActorImpl::detach()
126 auto* context = dynamic_cast<context::AttachContext*>(context::Context::self());
127 if (context == nullptr)
128 xbt_die("Not a suitable context");
130 context->get_actor()->cleanup();
131 context->attach_stop();
134 void ActorImpl::cleanup()
136 if (this == simix_global->maestro_process) /* Do not cleanup maestro */
139 XBT_DEBUG("Cleanup actor %s (%p), waiting synchro %p", get_cname(), this, waiting_synchro.get());
141 /* Unregister from the kill timer if any */
142 if (kill_timer != nullptr) {
143 kill_timer->remove();
144 kill_timer = nullptr;
147 simix_global->mutex.lock();
149 simix_global->process_list.erase(pid_);
150 if (host_ && host_process_list_hook.is_linked())
151 simgrid::xbt::intrusive_erase(host_->pimpl_->process_list_, *this);
152 if (not smx_destroy_list_hook.is_linked()) {
154 xbt_dynar_push_as(simix_global->dead_actors_vector, ActorImpl*, this);
156 simix_global->actors_to_destroy.push_back(*this);
158 context_->iwannadie = false;
160 simix_global->mutex.unlock();
163 void ActorImpl::exit()
165 context_->iwannadie = true;
168 exception_ = nullptr;
170 // Forcefully kill the actor if its host is turned off. Not a HostFailureException because you should not survive that
171 if (not host_->is_on())
172 this->throw_exception(std::make_exception_ptr(ForcefulKillException("host failed")));
174 /* destroy the blocking synchro if any */
175 if (waiting_synchro != nullptr) {
177 activity::ExecImplPtr exec = boost::dynamic_pointer_cast<activity::ExecImpl>(waiting_synchro);
178 activity::CommImplPtr comm = boost::dynamic_pointer_cast<activity::CommImpl>(waiting_synchro);
179 activity::SleepImplPtr sleep = boost::dynamic_pointer_cast<activity::SleepImpl>(waiting_synchro);
180 activity::RawImplPtr raw = boost::dynamic_pointer_cast<activity::RawImpl>(waiting_synchro);
181 activity::IoImplPtr io = boost::dynamic_pointer_cast<activity::IoImpl>(waiting_synchro);
183 if (exec != nullptr && exec->surf_action_) {
185 exec->surf_action_->unref();
186 exec->surf_action_ = nullptr;
187 } else if (comm != nullptr) {
188 comms.remove(waiting_synchro);
190 // Remove first occurrence of &actor->simcall:
191 auto i = boost::range::find(waiting_synchro->simcalls_, &simcall);
192 if (i != waiting_synchro->simcalls_.end())
193 waiting_synchro->simcalls_.remove(&simcall);
194 } else if (sleep != nullptr) {
195 if (sleep->surf_action_)
196 sleep->surf_action_->cancel();
198 } else if (raw != nullptr) {
200 } else if (io != nullptr) {
203 simgrid::kernel::activity::ActivityImplPtr activity = waiting_synchro;
204 xbt_die("Activity %s is of unknown type %s", activity->get_cname(),
205 simgrid::xbt::demangle(typeid(activity).name()).get());
208 waiting_synchro = nullptr;
212 void ActorImpl::kill(ActorImpl* actor)
214 if (actor->finished_) {
215 XBT_DEBUG("Ignoring request to kill actor %s@%s that is already dead", actor->get_cname(),
216 actor->host_->get_cname());
220 XBT_DEBUG("Actor '%s'@%s is killing actor '%s'@%s", get_cname(), host_ ? host_->get_cname() : "", actor->get_cname(),
221 actor->host_ ? actor->host_->get_cname() : "");
225 if (std::find(begin(simix_global->actors_to_run), end(simix_global->actors_to_run), actor) ==
226 end(simix_global->actors_to_run) &&
228 XBT_DEBUG("Inserting %s in the to_run list", actor->get_cname());
229 simix_global->actors_to_run.push_back(actor);
233 void ActorImpl::kill_all()
235 for (auto const& kv : simix_global->process_list)
236 if (kv.second != this)
237 this->kill(kv.second);
240 void ActorImpl::set_kill_time(double kill_time)
242 if (kill_time <= SIMIX_get_clock())
244 XBT_DEBUG("Set kill time %f for actor %s@%s", kill_time, get_cname(), host_->get_cname());
245 kill_timer = simix::Timer::set(kill_time, [this] {
247 kill_timer = nullptr;
251 double ActorImpl::get_kill_time()
253 return kill_timer ? kill_timer->get_date() : 0;
256 static void dying_daemon(int /*exit_status*/, void* data)
258 std::vector<ActorImpl*>* vect = &simix_global->daemons;
260 auto it = std::find(vect->begin(), vect->end(), static_cast<ActorImpl*>(data));
261 xbt_assert(it != vect->end(), "The dying daemon is not a daemon after all. Please report that bug.");
263 /* Don't move the whole content since we don't really care about the order */
264 std::swap(*it, vect->back());
268 void ActorImpl::yield()
270 XBT_DEBUG("Yield actor '%s'", get_cname());
272 /* Go into sleep and return control to maestro */
275 /* Ok, maestro returned control to us */
276 XBT_DEBUG("Control returned to me: '%s'", get_cname());
278 if (context_->iwannadie) {
280 XBT_DEBUG("Actor %s@%s is dead", get_cname(), host_->get_cname());
281 // throw simgrid::kernel::context::ForcefulKillException(); Does not seem to properly kill the actor
287 XBT_DEBUG("Hey! I'm suspended.");
289 xbt_assert(exception_ == nullptr, "Gasp! This exception may be lost by subsequent calls.");
294 if (exception_ != nullptr) {
295 XBT_DEBUG("Wait, maestro left me an exception");
296 std::exception_ptr exception = std::move(exception_);
297 exception_ = nullptr;
298 std::rethrow_exception(std::move(exception));
301 if (SMPI_switch_data_segment && not finished_) {
302 SMPI_switch_data_segment(iface());
306 /** This actor will be terminated automatically when the last non-daemon actor finishes */
307 void ActorImpl::daemonize()
311 simix_global->daemons.push_back(this);
312 SIMIX_process_on_exit(this, dying_daemon, this);
316 s4u::Actor* ActorImpl::restart()
318 XBT_DEBUG("Restarting actor %s on %s", get_cname(), host_->get_cname());
320 // retrieve the arguments of the old actor
321 ProcessArg arg = ProcessArg(host_, this);
323 // kill the old actor
324 (this == simix_global->maestro_process) ? this->exit() : SIMIX_process_self()->kill(this);
326 // start the new actor
328 ActorImpl::create(arg.name, std::move(arg.code), arg.data, arg.host, arg.properties.get(), nullptr);
329 actor->set_kill_time(arg.kill_time);
330 actor->set_auto_restart(arg.auto_restart);
332 return actor->ciface();
335 activity::ActivityImplPtr ActorImpl::suspend(ActorImpl* issuer)
338 XBT_DEBUG("Actor '%s' is already suspended", get_cname());
344 /* If we are suspending another actor that is waiting on a sync, suspend its synchronization. */
345 if (this != issuer) {
347 waiting_synchro->suspend();
348 /* If the other actor is not waiting, its suspension is delayed to when the actor is rescheduled. */
352 return activity::ExecImplPtr(new activity::ExecImpl("suspend", ""))->set_host(host_)->start(0.0, 1.0, 0.0);
356 void ActorImpl::resume()
358 XBT_IN("actor = %p", this);
360 if (context_->iwannadie) {
361 XBT_VERB("Ignoring request to suspend an actor that is currently dying.");
369 /* resume the synchronization that was blocking the resumed actor. */
371 waiting_synchro->resume();
376 activity::ActivityImplPtr ActorImpl::join(ActorImpl* actor, double timeout)
378 activity::ActivityImplPtr res = this->sleep(timeout);
379 intrusive_ptr_add_ref(res.get());
380 SIMIX_process_on_exit(actor,
382 auto sleep = static_cast<activity::SleepImpl*>(arg);
383 if (sleep->surf_action_)
384 sleep->surf_action_->finish(resource::Action::State::FINISHED);
385 intrusive_ptr_release(sleep);
391 activity::ActivityImplPtr ActorImpl::sleep(double duration)
393 if (not host_->is_on())
394 throw_exception(std::make_exception_ptr(simgrid::HostFailureException(
395 XBT_THROW_POINT, std::string("Host ") + host_->get_cname() + " failed, you cannot sleep there.")));
397 return activity::SleepImplPtr(new activity::SleepImpl())
400 ->set_duration(duration)
404 void ActorImpl::throw_exception(std::exception_ptr e)
411 /* cancel the blocking synchro if any */
412 if (waiting_synchro) {
414 activity::ExecImplPtr exec = boost::dynamic_pointer_cast<activity::ExecImpl>(waiting_synchro);
418 activity::CommImplPtr comm = boost::dynamic_pointer_cast<activity::CommImpl>(waiting_synchro);
419 if (comm != nullptr) {
424 activity::SleepImplPtr sleep = boost::dynamic_pointer_cast<activity::SleepImpl>(waiting_synchro);
425 if (sleep != nullptr) {
426 SIMIX_process_sleep_destroy(sleep);
427 if (std::find(begin(simix_global->actors_to_run), end(simix_global->actors_to_run), this) ==
428 end(simix_global->actors_to_run) &&
429 this != SIMIX_process_self()) {
430 XBT_DEBUG("Inserting [%p] %s in the to_run list", this, get_cname());
431 simix_global->actors_to_run.push_back(this);
435 activity::RawImplPtr raw = boost::dynamic_pointer_cast<activity::RawImpl>(waiting_synchro);
436 if (raw != nullptr) {
440 activity::IoImplPtr io = boost::dynamic_pointer_cast<activity::IoImpl>(waiting_synchro);
445 waiting_synchro = nullptr;
448 void ActorImpl::set_host(s4u::Host* dest)
450 xbt::intrusive_erase(host_->pimpl_->process_list_, *this);
452 dest->pimpl_->process_list_.push_back(*this);
455 ActorImplPtr ActorImpl::init(const std::string& name, s4u::Host* host)
457 ActorImpl* actor = new ActorImpl(xbt::string(name), host);
458 actor->set_ppid(this->pid_);
460 intrusive_ptr_add_ref(actor);
461 /* The on_creation() signal must be delayed until there, where the pid and everything is set */
462 s4u::Actor::on_creation(actor->iface());
464 return ActorImplPtr(actor);
467 ActorImpl* ActorImpl::start(const simix::ActorCode& code)
469 xbt_assert(code && host_ != nullptr, "Invalid parameters");
471 if (not host_->is_on()) {
472 XBT_WARN("Cannot launch actor '%s' on failed host '%s'", name_.c_str(), host_->get_cname());
473 intrusive_ptr_release(this);
474 std::rethrow_exception(
475 std::make_exception_ptr(simgrid::HostFailureException(XBT_THROW_POINT, "Cannot start actor on failed host.")));
479 XBT_VERB("Create context %s", get_cname());
480 context_.reset(simix_global->context_factory->create_context(simix::ActorCode(code), this));
482 XBT_DEBUG("Start context '%s'", get_cname());
484 /* Add the actor to its host's actor list */
485 host_->pimpl_->process_list_.push_back(*this);
486 simix_global->process_list[pid_] = this;
488 /* Now insert it in the global actor list and in the actor to run list */
489 XBT_DEBUG("Inserting [%p] %s(%s) in the to_run list", this, get_cname(), host_->get_cname());
490 simix_global->actors_to_run.push_back(this);
495 ActorImplPtr ActorImpl::create(const std::string& name, const simix::ActorCode& code, void* data, s4u::Host* host,
496 std::unordered_map<std::string, std::string>* properties, ActorImpl* parent_actor)
498 XBT_DEBUG("Start actor %s@'%s'", name.c_str(), host->get_cname());
501 if (parent_actor != nullptr)
502 actor = parent_actor->init(xbt::string(name), host);
504 actor = SIMIX_process_self()->init(xbt::string(name), host);
507 actor->set_user_data(data);
510 if (properties != nullptr)
511 for (auto const& kv : *properties)
512 actor->set_property(kv.first, kv.second);
519 void create_maestro(const std::function<void()>& code)
521 /* Create maestro actor and initialize it */
522 ActorImpl* maestro = new ActorImpl(xbt::string(""), /*host*/ nullptr);
525 maestro->context_.reset(simix_global->context_factory->create_context(simix::ActorCode(), maestro));
527 maestro->context_.reset(simix_global->context_factory->create_maestro(simix::ActorCode(code), maestro));
530 maestro->simcall.issuer = maestro;
531 simix_global->maestro_process = maestro;
535 } // namespace kernel
536 } // namespace simgrid
538 void SIMIX_process_detach()
540 simgrid::kernel::actor::ActorImpl::detach();
543 smx_actor_t SIMIX_process_attach(const char* name, void* data, const char* hostname,
544 std::unordered_map<std::string, std::string>* properties,
545 smx_actor_t /*parent_process*/)
547 return simgrid::kernel::actor::ActorImpl::attach(name, data, sg_host_by_name(hostname), properties).get();
550 /** @deprecated When this function gets removed, also remove the xbt_ex class, that is only there to help users to
552 void SIMIX_process_throw(smx_actor_t actor, xbt_errcat_t cat, int value, const char* msg)
554 SMX_EXCEPTION(actor, cat, value, msg);
556 if (actor->is_suspended())
559 /* cancel the blocking synchro if any */
560 if (actor->waiting_synchro) {
562 simgrid::kernel::activity::ExecImplPtr exec =
563 boost::dynamic_pointer_cast<simgrid::kernel::activity::ExecImpl>(actor->waiting_synchro);
567 simgrid::kernel::activity::CommImplPtr comm =
568 boost::dynamic_pointer_cast<simgrid::kernel::activity::CommImpl>(actor->waiting_synchro);
569 if (comm != nullptr) {
570 actor->comms.remove(comm);
574 simgrid::kernel::activity::SleepImplPtr sleep =
575 boost::dynamic_pointer_cast<simgrid::kernel::activity::SleepImpl>(actor->waiting_synchro);
576 if (sleep != nullptr) {
577 SIMIX_process_sleep_destroy(sleep);
578 if (std::find(begin(simix_global->actors_to_run), end(simix_global->actors_to_run), actor) ==
579 end(simix_global->actors_to_run) &&
580 actor != SIMIX_process_self()) {
581 XBT_DEBUG("Inserting [%p] %s in the to_run list", actor, actor->get_cname());
582 simix_global->actors_to_run.push_back(actor);
586 simgrid::kernel::activity::RawImplPtr raw =
587 boost::dynamic_pointer_cast<simgrid::kernel::activity::RawImpl>(actor->waiting_synchro);
588 if (raw != nullptr) {
592 simgrid::kernel::activity::IoImplPtr io =
593 boost::dynamic_pointer_cast<simgrid::kernel::activity::IoImpl>(actor->waiting_synchro);
598 actor->waiting_synchro = nullptr;
601 void simcall_HANDLER_process_suspend(smx_simcall_t simcall, smx_actor_t actor)
603 smx_activity_t sync_suspend = actor->suspend(simcall->issuer);
605 if (actor != simcall->issuer) {
606 SIMIX_simcall_answer(simcall);
608 sync_suspend->simcalls_.push_back(simcall);
609 actor->waiting_synchro = sync_suspend;
610 actor->waiting_synchro->suspend();
612 /* If we are suspending ourselves, then just do not finish the simcall now */
615 int SIMIX_process_get_maxpid()
617 return simix_process_maxpid;
620 int SIMIX_process_count()
622 return simix_global->process_list.size();
625 void* SIMIX_process_self_get_data() // deprecated
627 smx_actor_t self = SIMIX_process_self();
629 if (self == nullptr) {
632 return self->get_user_data();
635 void SIMIX_process_self_set_data(void* data) // deprecated
637 SIMIX_process_self()->set_user_data(data);
640 /* needs to be public and without simcall because it is called
641 by exceptions and logging events */
642 const char* SIMIX_process_self_get_name()
645 smx_actor_t process = SIMIX_process_self();
646 if (process == nullptr || process == simix_global->maestro_process)
649 return process->get_cname();
652 void simcall_HANDLER_process_join(smx_simcall_t simcall, smx_actor_t process, double timeout)
654 if (process->finished_) {
655 // The joined process is already finished, just wake up the issuer process right away
656 simcall_process_sleep__set__result(simcall, SIMIX_DONE);
657 SIMIX_simcall_answer(simcall);
660 smx_activity_t sync = simcall->issuer->join(process, timeout);
661 sync->simcalls_.push_back(simcall);
662 simcall->issuer->waiting_synchro = sync;
665 void simcall_HANDLER_process_sleep(smx_simcall_t simcall, double duration)
667 if (MC_is_active() || MC_record_replay_is_active()) {
668 MC_process_clock_add(simcall->issuer, duration);
669 simcall_process_sleep__set__result(simcall, SIMIX_DONE);
670 SIMIX_simcall_answer(simcall);
673 smx_activity_t sync = simcall->issuer->sleep(duration);
674 sync->simcalls_.push_back(simcall);
675 simcall->issuer->waiting_synchro = sync;
678 void SIMIX_process_sleep_destroy(simgrid::kernel::activity::SleepImplPtr sleep)
680 XBT_DEBUG("Destroy sleep synchro %p", sleep.get());
682 if (sleep->surf_action_) {
683 sleep->surf_action_->unref();
684 sleep->surf_action_ = nullptr;
689 * @brief Calling this function makes the process to yield.
691 * Only the current process can call this function, giving back the control to maestro.
693 * @param self the current process
696 /** @brief Returns the list of processes to run.
699 const std::vector<smx_actor_t>& simgrid::simix::process_get_runnable()
701 return simix_global->actors_to_run;
704 /** @brief Returns the process from PID. */
705 smx_actor_t SIMIX_process_from_PID(aid_t PID)
707 auto actor = simix_global->process_list.find(PID);
708 return actor == simix_global->process_list.end() ? nullptr : actor->second;
711 void SIMIX_process_on_exit(smx_actor_t actor, int_f_pvoid_pvoid_t fun, void* data)
713 SIMIX_process_on_exit(actor, [fun](int a, void* b) { fun((void*)(intptr_t)a, b); }, data);
716 void SIMIX_process_on_exit(smx_actor_t actor, const std::function<void(bool, void*)>& fun, void* data)
718 xbt_assert(actor, "current process not found: are you in maestro context ?");
720 actor->on_exit.emplace_back(s_smx_process_exit_fun_t{fun, data});
723 /** @brief Restart a process, starting it again from the beginning. */
725 * @ingroup simix_process_management
726 * @brief Creates and runs a new SIMIX process.
728 * The structure and the corresponding thread are created and put in the list of ready processes.
730 * @param name a name for the process. It is for user-level information and can be nullptr.
731 * @param code the main function of the process
732 * @param data a pointer to any data one may want to attach to the new object. It is for user-level information and can
734 * It can be retrieved with the method ActorImpl::getUserData().
735 * @param host where the new agent is executed.
736 * @param properties the properties of the process
738 smx_actor_t simcall_process_create(const std::string& name, const simgrid::simix::ActorCode& code, void* data,
739 sg_host_t host, std::unordered_map<std::string, std::string>* properties)
741 smx_actor_t self = SIMIX_process_self();
742 return simgrid::simix::simcall([&name, &code, data, host, properties, self] {
743 return simgrid::kernel::actor::ActorImpl::create(name, code, data, host, properties, self).get();
747 void simcall_process_set_data(smx_actor_t process, void* data)
749 simgrid::simix::simcall([process, data] { process->set_user_data(data); });