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
45 int SIMIX_process_has_pending_comms(smx_actor_t process) {
47 return process->comms.size() > 0;
54 ActorImpl::ActorImpl(simgrid::xbt::string name, s4u::Host* host) : host_(host), name_(name), piface_(this)
56 pid_ = simix_process_maxpid++;
57 simcall.issuer = this;
60 ActorImpl::~ActorImpl()
62 delete this->context_;
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(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_ = 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_);
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, smx_actor_t, 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(smx_actor_t 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<simgrid::kernel::activity::SleepImpl*>(arg);
377 if (sleep->surf_action_)
378 sleep->surf_action_->finish(simgrid::kernel::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 simgrid::kernel::activity::SleepImplPtr(new simgrid::kernel::activity::SleepImpl("sleep", host_))
395 void ActorImpl::throw_exception(std::exception_ptr e)
402 /* cancel the blocking synchro if any */
403 if (waiting_synchro) {
405 activity::ExecImplPtr exec = boost::dynamic_pointer_cast<activity::ExecImpl>(waiting_synchro);
409 activity::CommImplPtr comm = boost::dynamic_pointer_cast<activity::CommImpl>(waiting_synchro);
410 if (comm != nullptr) {
415 activity::SleepImplPtr sleep = boost::dynamic_pointer_cast<activity::SleepImpl>(waiting_synchro);
416 if (sleep != nullptr) {
417 SIMIX_process_sleep_destroy(waiting_synchro);
418 if (std::find(begin(simix_global->actors_to_run), end(simix_global->actors_to_run), this) ==
419 end(simix_global->actors_to_run) &&
420 this != SIMIX_process_self()) {
421 XBT_DEBUG("Inserting [%p] %s in the to_run list", this, get_cname());
422 simix_global->actors_to_run.push_back(this);
426 activity::RawImplPtr raw = boost::dynamic_pointer_cast<activity::RawImpl>(waiting_synchro);
427 if (raw != nullptr) {
431 activity::IoImplPtr io = boost::dynamic_pointer_cast<activity::IoImpl>(waiting_synchro);
436 waiting_synchro = nullptr;
439 void ActorImpl::set_host(s4u::Host* dest)
441 simgrid::xbt::intrusive_erase(host_->pimpl_->process_list_, *this);
443 dest->pimpl_->process_list_.push_back(*this);
446 ActorImplPtr ActorImpl::create(std::string name, simix::ActorCode code, void* data, s4u::Host* host,
447 std::unordered_map<std::string, std::string>* properties, ActorImpl* parent_actor)
450 XBT_DEBUG("Start actor %s@'%s'", name.c_str(), host->get_cname());
452 if (not host->is_on()) {
453 XBT_WARN("Cannot launch actor '%s' on failed host '%s'", name.c_str(), host->get_cname());
454 std::rethrow_exception(
455 std::make_exception_ptr(simgrid::HostFailureException(XBT_THROW_POINT, "Cannot create actor on failed host.")));
458 ActorImpl* actor = new ActorImpl(simgrid::xbt::string(name), host);
460 xbt_assert(code && host != nullptr, "Invalid parameters");
462 actor->set_user_data(data);
465 if (parent_actor != nullptr)
466 actor->set_ppid(parent_actor->get_pid());
468 XBT_VERB("Create context %s", actor->get_cname());
469 actor->context_ = simix_global->context_factory->create_context(std::move(code), actor);
472 if (properties != nullptr)
473 for (auto const& kv : *properties)
474 actor->set_property(kv.first, kv.second);
476 /* Add the actor to its host's actor list */
477 host->pimpl_->process_list_.push_back(*actor);
479 XBT_DEBUG("Start context '%s'", actor->get_cname());
481 /* Now insert it in the global actor list and in the actor to run list */
482 simix_global->process_list[actor->get_pid()] = actor;
483 XBT_DEBUG("Inserting [%p] %s(%s) in the to_run list", actor, actor->get_cname(), host->get_cname());
484 simix_global->actors_to_run.push_back(actor);
485 intrusive_ptr_add_ref(actor);
487 /* The on_creation() signal must be delayed until there, where the pid and everything is set */
488 s4u::Actor::on_creation(actor->iface());
490 return ActorImplPtr(actor);
493 void create_maestro(simix::ActorCode code)
495 /* Create maestro actor and initialize it */
496 ActorImpl* maestro = new ActorImpl(xbt::string(""), /*host*/ nullptr);
499 maestro->context_ = simix_global->context_factory->create_context(simix::ActorCode(), maestro);
501 maestro->context_ = simix_global->context_factory->create_maestro(code, maestro);
504 maestro->simcall.issuer = maestro;
505 simix_global->maestro_process = maestro;
509 } // namespace kernel
512 void SIMIX_process_detach()
514 simgrid::kernel::actor::ActorImpl::detach();
517 smx_actor_t SIMIX_process_attach(const char* name, void* data, const char* hostname,
518 std::unordered_map<std::string, std::string>* properties,
519 smx_actor_t /*parent_process*/)
521 return simgrid::kernel::actor::ActorImpl::attach(name, data, sg_host_by_name(hostname), properties).get();
524 /** @deprecated When this function gets removed, also remove the xbt_ex class, that is only there to help users to
526 void SIMIX_process_throw(smx_actor_t actor, xbt_errcat_t cat, int value, const char* msg)
528 SMX_EXCEPTION(actor, cat, value, msg);
530 if (actor->is_suspended())
533 /* cancel the blocking synchro if any */
534 if (actor->waiting_synchro) {
536 simgrid::kernel::activity::ExecImplPtr exec =
537 boost::dynamic_pointer_cast<simgrid::kernel::activity::ExecImpl>(actor->waiting_synchro);
541 simgrid::kernel::activity::CommImplPtr comm =
542 boost::dynamic_pointer_cast<simgrid::kernel::activity::CommImpl>(actor->waiting_synchro);
543 if (comm != nullptr) {
544 actor->comms.remove(comm);
548 simgrid::kernel::activity::SleepImplPtr sleep =
549 boost::dynamic_pointer_cast<simgrid::kernel::activity::SleepImpl>(actor->waiting_synchro);
550 if (sleep != nullptr) {
551 SIMIX_process_sleep_destroy(actor->waiting_synchro);
552 if (std::find(begin(simix_global->actors_to_run), end(simix_global->actors_to_run), actor) ==
553 end(simix_global->actors_to_run) &&
554 actor != SIMIX_process_self()) {
555 XBT_DEBUG("Inserting [%p] %s in the to_run list", actor, actor->get_cname());
556 simix_global->actors_to_run.push_back(actor);
560 simgrid::kernel::activity::RawImplPtr raw =
561 boost::dynamic_pointer_cast<simgrid::kernel::activity::RawImpl>(actor->waiting_synchro);
562 if (raw != nullptr) {
566 simgrid::kernel::activity::IoImplPtr io =
567 boost::dynamic_pointer_cast<simgrid::kernel::activity::IoImpl>(actor->waiting_synchro);
572 actor->waiting_synchro = nullptr;
575 void simcall_HANDLER_process_suspend(smx_simcall_t simcall, smx_actor_t actor)
577 smx_activity_t sync_suspend = actor->suspend(simcall->issuer);
579 if (actor != simcall->issuer) {
580 SIMIX_simcall_answer(simcall);
582 sync_suspend->simcalls_.push_back(simcall);
583 actor->waiting_synchro = sync_suspend;
584 actor->waiting_synchro->suspend();
586 /* If we are suspending ourselves, then just do not finish the simcall now */
589 int SIMIX_process_get_maxpid() {
590 return simix_process_maxpid;
593 int SIMIX_process_count()
595 return simix_global->process_list.size();
598 void* SIMIX_process_self_get_data()
600 smx_actor_t self = SIMIX_process_self();
602 if (self == nullptr) {
605 return self->get_user_data();
608 void SIMIX_process_self_set_data(void *data)
610 SIMIX_process_self()->set_user_data(data);
614 /* needs to be public and without simcall because it is called
615 by exceptions and logging events */
616 const char* SIMIX_process_self_get_name() {
618 smx_actor_t process = SIMIX_process_self();
619 if (process == nullptr || process == simix_global->maestro_process)
622 return process->get_cname();
625 void simcall_HANDLER_process_join(smx_simcall_t simcall, smx_actor_t process, double timeout)
627 if (process->finished_) {
628 // The joined process is already finished, just wake up the issuer process right away
629 simcall_process_sleep__set__result(simcall, SIMIX_DONE);
630 SIMIX_simcall_answer(simcall);
633 smx_activity_t sync = simcall->issuer->join(process, timeout);
634 sync->simcalls_.push_back(simcall);
635 simcall->issuer->waiting_synchro = sync;
638 void simcall_HANDLER_process_sleep(smx_simcall_t simcall, double duration)
640 if (MC_is_active() || MC_record_replay_is_active()) {
641 MC_process_clock_add(simcall->issuer, duration);
642 simcall_process_sleep__set__result(simcall, SIMIX_DONE);
643 SIMIX_simcall_answer(simcall);
646 smx_activity_t sync = simcall->issuer->sleep(duration);
647 sync->simcalls_.push_back(simcall);
648 simcall->issuer->waiting_synchro = sync;
651 void SIMIX_process_sleep_destroy(smx_activity_t synchro)
653 XBT_DEBUG("Destroy sleep synchro %p", synchro.get());
654 simgrid::kernel::activity::SleepImplPtr sleep =
655 boost::static_pointer_cast<simgrid::kernel::activity::SleepImpl>(synchro);
657 if (sleep->surf_action_) {
658 sleep->surf_action_->unref();
659 sleep->surf_action_ = nullptr;
664 * @brief Calling this function makes the process to yield.
666 * Only the current process can call this function, giving back the control to maestro.
668 * @param self the current process
671 /** @brief Returns the list of processes to run.
674 const std::vector<smx_actor_t>& simgrid::simix::process_get_runnable()
676 return simix_global->actors_to_run;
679 /** @brief Returns the process from PID. */
680 smx_actor_t SIMIX_process_from_PID(aid_t PID)
682 auto actor = simix_global->process_list.find(PID);
683 return actor == simix_global->process_list.end() ? nullptr : actor->second;
686 void SIMIX_process_on_exit(smx_actor_t actor, int_f_pvoid_pvoid_t fun, void* data)
688 SIMIX_process_on_exit(actor, [fun](int a, void* b) { fun((void*)(intptr_t)a, b); }, data);
691 void SIMIX_process_on_exit(smx_actor_t actor, std::function<void(bool, void*)> fun, void* data)
693 xbt_assert(actor, "current process not found: are you in maestro context ?");
695 actor->on_exit.emplace_back(s_smx_process_exit_fun_t{fun, data});
698 /** @brief Restart a process, starting it again from the beginning. */
700 * @ingroup simix_process_management
701 * @brief Creates and runs a new SIMIX process.
703 * The structure and the corresponding thread are created and put in the list of ready processes.
705 * @param name a name for the process. It is for user-level information and can be nullptr.
706 * @param code the main function of the process
707 * @param data a pointer to any data one may want to attach to the new object. It is for user-level information and can
709 * It can be retrieved with the method ActorImpl::getUserData().
710 * @param host where the new agent is executed.
711 * @param properties the properties of the process
713 smx_actor_t simcall_process_create(std::string name, simgrid::simix::ActorCode code, void* data, sg_host_t host,
714 std::unordered_map<std::string, std::string>* properties)
716 smx_actor_t self = SIMIX_process_self();
717 return simgrid::simix::simcall([name, code, data, host, properties, self] {
718 return simgrid::kernel::actor::ActorImpl::create(std::move(name), std::move(code), data, host, properties, self)
723 void simcall_process_set_data(smx_actor_t process, void* data)
725 simgrid::simix::simcall([process, data] { process->set_user_data(data); });