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 simix_global->mutex.lock();
143 simix_global->process_list.erase(pid_);
144 if (host_ && host_process_list_hook.is_linked())
145 simgrid::xbt::intrusive_erase(host_->pimpl_->process_list_, *this);
146 if (not smx_destroy_list_hook.is_linked()) {
148 xbt_dynar_push_as(simix_global->dead_actors_vector, ActorImpl*, this);
150 simix_global->actors_to_destroy.push_back(*this);
152 context_->iwannadie = false;
154 simix_global->mutex.unlock();
157 void ActorImpl::exit()
159 context_->iwannadie = true;
162 exception_ = nullptr;
164 // Forcefully kill the actor if its host is turned off. Not a HostFailureException because you should not survive that
165 if (not host_->is_on())
166 this->throw_exception(std::make_exception_ptr(ForcefulKillException("host failed")));
168 /* destroy the blocking synchro if any */
169 if (waiting_synchro != nullptr) {
171 activity::ExecImplPtr exec = boost::dynamic_pointer_cast<activity::ExecImpl>(waiting_synchro);
172 activity::CommImplPtr comm = boost::dynamic_pointer_cast<activity::CommImpl>(waiting_synchro);
173 activity::SleepImplPtr sleep = boost::dynamic_pointer_cast<activity::SleepImpl>(waiting_synchro);
174 activity::RawImplPtr raw = boost::dynamic_pointer_cast<activity::RawImpl>(waiting_synchro);
175 activity::IoImplPtr io = boost::dynamic_pointer_cast<activity::IoImpl>(waiting_synchro);
177 if (exec != nullptr && exec->surf_action_) {
179 exec->surf_action_->unref();
180 exec->surf_action_ = nullptr;
181 } else if (comm != nullptr) {
182 comms.remove(waiting_synchro);
184 // Remove first occurrence of &actor->simcall:
185 auto i = boost::range::find(waiting_synchro->simcalls_, &simcall);
186 if (i != waiting_synchro->simcalls_.end())
187 waiting_synchro->simcalls_.remove(&simcall);
188 } else if (sleep != nullptr) {
189 if (sleep->surf_action_)
190 sleep->surf_action_->cancel();
192 } else if (raw != nullptr) {
194 } else if (io != nullptr) {
197 simgrid::kernel::activity::ActivityImplPtr activity = waiting_synchro;
198 xbt_die("Activity %s is of unknown type %s", activity->get_cname(),
199 simgrid::xbt::demangle(typeid(activity).name()).get());
202 waiting_synchro = nullptr;
206 void ActorImpl::kill(ActorImpl* actor)
208 if (actor->finished_) {
209 XBT_DEBUG("Ignoring request to kill actor %s@%s that is already dead", actor->get_cname(),
210 actor->host_->get_cname());
214 XBT_DEBUG("Actor '%s'@%s is killing actor '%s'@%s", get_cname(), host_ ? host_->get_cname() : "", actor->get_cname(),
215 actor->host_ ? actor->host_->get_cname() : "");
219 if (std::find(begin(simix_global->actors_to_run), end(simix_global->actors_to_run), actor) ==
220 end(simix_global->actors_to_run) &&
222 XBT_DEBUG("Inserting %s in the to_run list", actor->get_cname());
223 simix_global->actors_to_run.push_back(actor);
227 void ActorImpl::kill_all()
229 for (auto const& kv : simix_global->process_list)
230 if (kv.second != this)
231 this->kill(kv.second);
234 void ActorImpl::set_kill_time(double kill_time)
236 if (kill_time <= SIMIX_get_clock())
238 XBT_DEBUG("Set kill time %f for actor %s@%s", kill_time, get_cname(), host_->get_cname());
239 kill_timer = simix::Timer::set(kill_time, [this] {
241 kill_timer = nullptr;
245 double ActorImpl::get_kill_time()
247 return kill_timer ? kill_timer->get_date() : 0;
250 static void dying_daemon(int /*exit_status*/, void* data)
252 std::vector<ActorImpl*>* vect = &simix_global->daemons;
254 auto it = std::find(vect->begin(), vect->end(), static_cast<ActorImpl*>(data));
255 xbt_assert(it != vect->end(), "The dying daemon is not a daemon after all. Please report that bug.");
257 /* Don't move the whole content since we don't really care about the order */
258 std::swap(*it, vect->back());
262 void ActorImpl::yield()
264 XBT_DEBUG("Yield actor '%s'", get_cname());
266 /* Go into sleep and return control to maestro */
269 /* Ok, maestro returned control to us */
270 XBT_DEBUG("Control returned to me: '%s'", get_cname());
272 if (context_->iwannadie) {
274 XBT_DEBUG("Actor %s@%s is dead", get_cname(), host_->get_cname());
275 // throw simgrid::kernel::context::ForcefulKillException(); Does not seem to properly kill the actor
281 XBT_DEBUG("Hey! I'm suspended.");
283 xbt_assert(exception_ == nullptr, "Gasp! This exception may be lost by subsequent calls.");
288 if (exception_ != nullptr) {
289 XBT_DEBUG("Wait, maestro left me an exception");
290 std::exception_ptr exception = std::move(exception_);
291 exception_ = nullptr;
292 std::rethrow_exception(std::move(exception));
295 if (SMPI_switch_data_segment && not finished_) {
296 SMPI_switch_data_segment(iface());
300 /** This actor will be terminated automatically when the last non-daemon actor finishes */
301 void ActorImpl::daemonize()
305 simix_global->daemons.push_back(this);
306 SIMIX_process_on_exit(this, dying_daemon, this);
310 s4u::Actor* ActorImpl::restart()
312 XBT_DEBUG("Restarting actor %s on %s", get_cname(), host_->get_cname());
314 // retrieve the arguments of the old actor
315 ProcessArg arg = ProcessArg(host_, this);
317 // kill the old actor
318 (this == simix_global->maestro_process) ? this->exit() : SIMIX_process_self()->kill(this);
320 // start the new actor
322 ActorImpl::create(arg.name, std::move(arg.code), arg.data, arg.host, arg.properties.get(), nullptr);
323 actor->set_kill_time(arg.kill_time);
324 actor->set_auto_restart(arg.auto_restart);
326 return actor->ciface();
329 activity::ActivityImplPtr ActorImpl::suspend(ActorImpl* issuer)
332 XBT_DEBUG("Actor '%s' is already suspended", get_cname());
338 /* If we are suspending another actor that is waiting on a sync, suspend its synchronization. */
339 if (this != issuer) {
341 waiting_synchro->suspend();
342 /* If the other actor is not waiting, its suspension is delayed to when the actor is rescheduled. */
346 return activity::ExecImplPtr(new activity::ExecImpl("suspend", ""))->set_host(host_)->start(0.0, 1.0, 0.0);
350 void ActorImpl::resume()
352 XBT_IN("actor = %p", this);
354 if (context_->iwannadie) {
355 XBT_VERB("Ignoring request to suspend an actor that is currently dying.");
363 /* resume the synchronization that was blocking the resumed actor. */
365 waiting_synchro->resume();
370 activity::ActivityImplPtr ActorImpl::join(ActorImpl* actor, double timeout)
372 activity::ActivityImplPtr res = this->sleep(timeout);
373 intrusive_ptr_add_ref(res.get());
374 SIMIX_process_on_exit(actor,
376 auto sleep = static_cast<activity::SleepImpl*>(arg);
377 if (sleep->surf_action_)
378 sleep->surf_action_->finish(resource::Action::State::FINISHED);
379 intrusive_ptr_release(sleep);
385 activity::ActivityImplPtr ActorImpl::sleep(double duration)
387 if (not host_->is_on())
388 throw_exception(std::make_exception_ptr(simgrid::HostFailureException(
389 XBT_THROW_POINT, std::string("Host ") + host_->get_cname() + " failed, you cannot sleep there.")));
391 return activity::SleepImplPtr(new activity::SleepImpl())
394 ->set_duration(duration)
398 void ActorImpl::throw_exception(std::exception_ptr e)
405 /* cancel the blocking synchro if any */
406 if (waiting_synchro) {
408 activity::ExecImplPtr exec = boost::dynamic_pointer_cast<activity::ExecImpl>(waiting_synchro);
412 activity::CommImplPtr comm = boost::dynamic_pointer_cast<activity::CommImpl>(waiting_synchro);
413 if (comm != nullptr) {
418 activity::SleepImplPtr sleep = boost::dynamic_pointer_cast<activity::SleepImpl>(waiting_synchro);
419 if (sleep != nullptr) {
420 SIMIX_process_sleep_destroy(sleep);
421 if (std::find(begin(simix_global->actors_to_run), end(simix_global->actors_to_run), this) ==
422 end(simix_global->actors_to_run) &&
423 this != SIMIX_process_self()) {
424 XBT_DEBUG("Inserting [%p] %s in the to_run list", this, get_cname());
425 simix_global->actors_to_run.push_back(this);
429 activity::RawImplPtr raw = boost::dynamic_pointer_cast<activity::RawImpl>(waiting_synchro);
430 if (raw != nullptr) {
434 activity::IoImplPtr io = boost::dynamic_pointer_cast<activity::IoImpl>(waiting_synchro);
439 waiting_synchro = nullptr;
442 void ActorImpl::set_host(s4u::Host* dest)
444 xbt::intrusive_erase(host_->pimpl_->process_list_, *this);
446 dest->pimpl_->process_list_.push_back(*this);
449 ActorImplPtr ActorImpl::init(const std::string& name, s4u::Host* host)
451 ActorImpl* actor = new ActorImpl(xbt::string(name), host);
452 actor->set_ppid(this->pid_);
454 intrusive_ptr_add_ref(actor);
455 /* The on_creation() signal must be delayed until there, where the pid and everything is set */
456 s4u::Actor::on_creation(actor->iface());
458 return ActorImplPtr(actor);
461 ActorImpl* ActorImpl::start(const simix::ActorCode& code)
463 xbt_assert(code && host_ != nullptr, "Invalid parameters");
465 if (not host_->is_on()) {
466 XBT_WARN("Cannot launch actor '%s' on failed host '%s'", name_.c_str(), host_->get_cname());
467 intrusive_ptr_release(this);
468 std::rethrow_exception(
469 std::make_exception_ptr(simgrid::HostFailureException(XBT_THROW_POINT, "Cannot start actor on failed host.")));
473 XBT_VERB("Create context %s", get_cname());
474 context_.reset(simix_global->context_factory->create_context(simix::ActorCode(code), this));
476 XBT_DEBUG("Start context '%s'", get_cname());
478 /* Add the actor to its host's actor list */
479 host_->pimpl_->process_list_.push_back(*this);
480 simix_global->process_list[pid_] = this;
482 /* Now insert it in the global actor list and in the actor to run list */
483 XBT_DEBUG("Inserting [%p] %s(%s) in the to_run list", this, get_cname(), host_->get_cname());
484 simix_global->actors_to_run.push_back(this);
489 ActorImplPtr ActorImpl::create(const std::string& name, const simix::ActorCode& code, void* data, s4u::Host* host,
490 std::unordered_map<std::string, std::string>* properties, ActorImpl* parent_actor)
492 XBT_DEBUG("Start actor %s@'%s'", name.c_str(), host->get_cname());
495 if (parent_actor != nullptr)
496 actor = parent_actor->init(xbt::string(name), host);
498 actor = SIMIX_process_self()->init(xbt::string(name), host);
501 actor->set_user_data(data);
504 if (properties != nullptr)
505 for (auto const& kv : *properties)
506 actor->set_property(kv.first, kv.second);
513 void create_maestro(const std::function<void()>& code)
515 /* Create maestro actor and initialize it */
516 ActorImpl* maestro = new ActorImpl(xbt::string(""), /*host*/ nullptr);
519 maestro->context_.reset(simix_global->context_factory->create_context(simix::ActorCode(), maestro));
521 maestro->context_.reset(simix_global->context_factory->create_maestro(simix::ActorCode(code), maestro));
524 maestro->simcall.issuer = maestro;
525 simix_global->maestro_process = maestro;
529 } // namespace kernel
530 } // namespace simgrid
532 void SIMIX_process_detach()
534 simgrid::kernel::actor::ActorImpl::detach();
537 smx_actor_t SIMIX_process_attach(const char* name, void* data, const char* hostname,
538 std::unordered_map<std::string, std::string>* properties,
539 smx_actor_t /*parent_process*/)
541 return simgrid::kernel::actor::ActorImpl::attach(name, data, sg_host_by_name(hostname), properties).get();
544 /** @deprecated When this function gets removed, also remove the xbt_ex class, that is only there to help users to
546 void SIMIX_process_throw(smx_actor_t actor, xbt_errcat_t cat, int value, const char* msg)
548 SMX_EXCEPTION(actor, cat, value, msg);
550 if (actor->is_suspended())
553 /* cancel the blocking synchro if any */
554 if (actor->waiting_synchro) {
556 simgrid::kernel::activity::ExecImplPtr exec =
557 boost::dynamic_pointer_cast<simgrid::kernel::activity::ExecImpl>(actor->waiting_synchro);
561 simgrid::kernel::activity::CommImplPtr comm =
562 boost::dynamic_pointer_cast<simgrid::kernel::activity::CommImpl>(actor->waiting_synchro);
563 if (comm != nullptr) {
564 actor->comms.remove(comm);
568 simgrid::kernel::activity::SleepImplPtr sleep =
569 boost::dynamic_pointer_cast<simgrid::kernel::activity::SleepImpl>(actor->waiting_synchro);
570 if (sleep != nullptr) {
571 SIMIX_process_sleep_destroy(sleep);
572 if (std::find(begin(simix_global->actors_to_run), end(simix_global->actors_to_run), actor) ==
573 end(simix_global->actors_to_run) &&
574 actor != SIMIX_process_self()) {
575 XBT_DEBUG("Inserting [%p] %s in the to_run list", actor, actor->get_cname());
576 simix_global->actors_to_run.push_back(actor);
580 simgrid::kernel::activity::RawImplPtr raw =
581 boost::dynamic_pointer_cast<simgrid::kernel::activity::RawImpl>(actor->waiting_synchro);
582 if (raw != nullptr) {
586 simgrid::kernel::activity::IoImplPtr io =
587 boost::dynamic_pointer_cast<simgrid::kernel::activity::IoImpl>(actor->waiting_synchro);
592 actor->waiting_synchro = nullptr;
595 void simcall_HANDLER_process_suspend(smx_simcall_t simcall, smx_actor_t actor)
597 smx_activity_t sync_suspend = actor->suspend(simcall->issuer);
599 if (actor != simcall->issuer) {
600 SIMIX_simcall_answer(simcall);
602 sync_suspend->simcalls_.push_back(simcall);
603 actor->waiting_synchro = sync_suspend;
604 actor->waiting_synchro->suspend();
606 /* If we are suspending ourselves, then just do not finish the simcall now */
609 int SIMIX_process_get_maxpid()
611 return simix_process_maxpid;
614 int SIMIX_process_count()
616 return simix_global->process_list.size();
619 void* SIMIX_process_self_get_data() // deprecated
621 smx_actor_t self = SIMIX_process_self();
623 if (self == nullptr) {
626 return self->get_user_data();
629 void SIMIX_process_self_set_data(void* data) // deprecated
631 SIMIX_process_self()->set_user_data(data);
634 /* needs to be public and without simcall because it is called
635 by exceptions and logging events */
636 const char* SIMIX_process_self_get_name()
639 smx_actor_t process = SIMIX_process_self();
640 if (process == nullptr || process == simix_global->maestro_process)
643 return process->get_cname();
646 void simcall_HANDLER_process_join(smx_simcall_t simcall, smx_actor_t process, double timeout)
648 if (process->finished_) {
649 // The joined process is already finished, just wake up the issuer process right away
650 simcall_process_sleep__set__result(simcall, SIMIX_DONE);
651 SIMIX_simcall_answer(simcall);
654 smx_activity_t sync = simcall->issuer->join(process, timeout);
655 sync->simcalls_.push_back(simcall);
656 simcall->issuer->waiting_synchro = sync;
659 void simcall_HANDLER_process_sleep(smx_simcall_t simcall, double duration)
661 if (MC_is_active() || MC_record_replay_is_active()) {
662 MC_process_clock_add(simcall->issuer, duration);
663 simcall_process_sleep__set__result(simcall, SIMIX_DONE);
664 SIMIX_simcall_answer(simcall);
667 smx_activity_t sync = simcall->issuer->sleep(duration);
668 sync->simcalls_.push_back(simcall);
669 simcall->issuer->waiting_synchro = sync;
672 void SIMIX_process_sleep_destroy(simgrid::kernel::activity::SleepImplPtr sleep)
674 XBT_DEBUG("Destroy sleep synchro %p", sleep.get());
676 if (sleep->surf_action_) {
677 sleep->surf_action_->unref();
678 sleep->surf_action_ = nullptr;
683 * @brief Calling this function makes the process to yield.
685 * Only the current process can call this function, giving back the control to maestro.
687 * @param self the current process
690 /** @brief Returns the list of processes to run.
693 const std::vector<smx_actor_t>& simgrid::simix::process_get_runnable()
695 return simix_global->actors_to_run;
698 /** @brief Returns the process from PID. */
699 smx_actor_t SIMIX_process_from_PID(aid_t PID)
701 auto actor = simix_global->process_list.find(PID);
702 return actor == simix_global->process_list.end() ? nullptr : actor->second;
705 void SIMIX_process_on_exit(smx_actor_t actor, int_f_pvoid_pvoid_t fun, void* data)
707 SIMIX_process_on_exit(actor, [fun](int a, void* b) { fun((void*)(intptr_t)a, b); }, data);
710 void SIMIX_process_on_exit(smx_actor_t actor, const std::function<void(bool, void*)>& fun, void* data)
712 xbt_assert(actor, "current process not found: are you in maestro context ?");
714 actor->on_exit.emplace_back(s_smx_process_exit_fun_t{fun, data});
717 /** @brief Restart a process, starting it again from the beginning. */
719 * @ingroup simix_process_management
720 * @brief Creates and runs a new SIMIX process.
722 * The structure and the corresponding thread are created and put in the list of ready processes.
724 * @param name a name for the process. It is for user-level information and can be nullptr.
725 * @param code the main function of the process
726 * @param data a pointer to any data one may want to attach to the new object. It is for user-level information and can
728 * It can be retrieved with the method ActorImpl::getUserData().
729 * @param host where the new agent is executed.
730 * @param properties the properties of the process
732 smx_actor_t simcall_process_create(const std::string& name, const simgrid::simix::ActorCode& code, void* data,
733 sg_host_t host, std::unordered_map<std::string, std::string>* properties)
735 smx_actor_t self = SIMIX_process_self();
736 return simgrid::simix::simcall([&name, &code, data, host, properties, self] {
737 return simgrid::kernel::actor::ActorImpl::create(name, code, data, host, properties, self).get();
741 void simcall_process_set_data(smx_actor_t process, void* data)
743 simgrid::simix::simcall([process, data] { process->set_user_data(data); });