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 simgrid::kernel::context::Context* 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::Actor::on_creation(*actor->ciface());
115 return ActorImplPtr(actor);
117 /** @brief Detach an actor attached with `attach()`
119 * This is called when the current actor has finished its job.
120 * Used in the main thread, it waits for the simulation to finish before returning. When it returns, the other
121 * simulated actors and the maestro are destroyed.
123 void ActorImpl::detach()
125 auto* context = dynamic_cast<context::AttachContext*>(context::Context::self());
126 if (context == nullptr)
127 xbt_die("Not a suitable context");
129 context->get_actor()->cleanup();
130 context->attach_stop();
133 void ActorImpl::cleanup()
137 if (has_to_auto_restart() && not get_host()->is_on()) {
138 XBT_DEBUG("Insert host %s to watched_hosts because it's off and %s needs to restart", get_host()->get_cname(),
140 watched_hosts.insert(get_host()->get_name());
143 // Execute the termination callbacks
144 bool failed = context_->iwannadie;
145 while (not on_exit.empty()) {
146 auto exit_fun = on_exit.back();
151 /* cancel non-blocking activities */
152 for (auto activity : comms)
153 boost::static_pointer_cast<activity::CommImpl>(activity)->cancel();
156 XBT_DEBUG("%s@%s(%ld) should not run anymore", get_cname(), get_host()->get_cname(), get_pid());
158 if (this == simix_global->maestro_process) /* Do not cleanup maestro */
161 XBT_DEBUG("Cleanup actor %s (%p), waiting synchro %p", get_cname(), this, waiting_synchro.get());
163 /* Unregister from the kill timer if any */
164 if (kill_timer != nullptr) {
165 kill_timer->remove();
166 kill_timer = nullptr;
169 simix_global->mutex.lock();
171 simix_global->process_list.erase(pid_);
172 if (host_ && host_process_list_hook.is_linked())
173 simgrid::xbt::intrusive_erase(host_->pimpl_->process_list_, *this);
174 if (not smx_destroy_list_hook.is_linked()) {
176 xbt_dynar_push_as(simix_global->dead_actors_vector, ActorImpl*, this);
178 simix_global->actors_to_destroy.push_back(*this);
181 simix_global->mutex.unlock();
183 context_->iwannadie = false; // don't let the simcall's yield() do a Context::stop(), to avoid infinite loops
184 simgrid::simix::simcall([this] { simgrid::s4u::Actor::on_destruction(*ciface()); });
185 context_->iwannadie = true;
188 void ActorImpl::exit()
190 context_->iwannadie = true;
193 exception_ = nullptr;
195 // Forcefully kill the actor if its host is turned off. Not a HostFailureException because you should not survive that
196 if (not host_->is_on())
197 this->throw_exception(std::make_exception_ptr(ForcefulKillException("host failed")));
199 /* destroy the blocking synchro if any */
200 if (waiting_synchro != nullptr) {
201 waiting_synchro->cancel();
202 waiting_synchro->state_ = SIMIX_FAILED;
204 activity::ExecImplPtr exec = boost::dynamic_pointer_cast<activity::ExecImpl>(waiting_synchro);
205 activity::CommImplPtr comm = boost::dynamic_pointer_cast<activity::CommImpl>(waiting_synchro);
207 if (exec != nullptr) {
208 exec->clean_action();
209 } else if (comm != nullptr) {
210 comms.remove(waiting_synchro);
211 // Remove first occurrence of &actor->simcall:
212 auto i = boost::range::find(waiting_synchro->simcalls_, &simcall);
213 if (i != waiting_synchro->simcalls_.end())
214 waiting_synchro->simcalls_.remove(&simcall);
216 activity::ActivityImplPtr(waiting_synchro)->finish();
219 waiting_synchro = nullptr;
223 void ActorImpl::kill(ActorImpl* actor)
225 if (actor->finished_) {
226 XBT_DEBUG("Ignoring request to kill actor %s@%s that is already dead", actor->get_cname(),
227 actor->host_->get_cname());
231 XBT_DEBUG("Actor '%s'@%s is killing actor '%s'@%s", get_cname(), host_ ? host_->get_cname() : "", actor->get_cname(),
232 actor->host_ ? actor->host_->get_cname() : "");
236 if (std::find(begin(simix_global->actors_to_run), end(simix_global->actors_to_run), actor) ==
237 end(simix_global->actors_to_run) &&
239 XBT_DEBUG("Inserting %s in the to_run list", actor->get_cname());
240 simix_global->actors_to_run.push_back(actor);
244 void ActorImpl::kill_all()
246 for (auto const& kv : simix_global->process_list)
247 if (kv.second != this)
248 this->kill(kv.second);
251 void ActorImpl::set_kill_time(double kill_time)
253 if (kill_time <= SIMIX_get_clock())
255 XBT_DEBUG("Set kill time %f for actor %s@%s", kill_time, get_cname(), host_->get_cname());
256 kill_timer = simix::Timer::set(kill_time, [this] {
258 kill_timer = nullptr;
262 double ActorImpl::get_kill_time()
264 return kill_timer ? kill_timer->get_date() : 0;
267 void ActorImpl::yield()
269 XBT_DEBUG("Yield actor '%s'", get_cname());
271 /* Go into sleep and return control to maestro */
274 /* Ok, maestro returned control to us */
275 XBT_DEBUG("Control returned to me: '%s'", get_cname());
277 if (context_->iwannadie) {
278 XBT_DEBUG("Actor %s@%s is dead", get_cname(), host_->get_cname());
279 // throw simgrid::kernel::context::ForcefulKillException(); Does not seem to properly kill the actor
285 XBT_DEBUG("Hey! I'm suspended.");
287 xbt_assert(exception_ == nullptr, "Gasp! This exception may be lost by subsequent calls.");
292 if (exception_ != nullptr) {
293 XBT_DEBUG("Wait, maestro left me an exception");
294 std::exception_ptr exception = std::move(exception_);
295 exception_ = nullptr;
296 std::rethrow_exception(std::move(exception));
299 if (SMPI_switch_data_segment && not finished_) {
300 SMPI_switch_data_segment(iface());
304 /** This actor will be terminated automatically when the last non-daemon actor finishes */
305 void ActorImpl::daemonize()
309 simix_global->daemons.push_back(this);
310 SIMIX_process_on_exit(this, [this](bool) {
311 auto& vect = simix_global->daemons;
312 auto it = std::find(vect.begin(), vect.end(), this);
313 xbt_assert(it != vect.end(), "The dying daemon is not a daemon after all. Please report that bug.");
315 /* Don't move the whole content since we don't really care about the order */
316 std::swap(*it, vect.back());
322 s4u::Actor* ActorImpl::restart()
324 xbt_assert(this != simix_global->maestro_process, "Restarting maestro is not supported");
326 XBT_DEBUG("Restarting actor %s on %s", get_cname(), host_->get_cname());
328 // retrieve the arguments of the old actor
329 ProcessArg arg = ProcessArg(host_, this);
331 // kill the old actor
332 context::Context::self()->get_actor()->kill(this);
334 // start the new actor
336 ActorImpl::create(arg.name, std::move(arg.code), arg.data, arg.host, arg.properties.get(), nullptr);
337 actor->set_kill_time(arg.kill_time);
338 actor->set_auto_restart(arg.auto_restart);
340 return actor->ciface();
343 activity::ActivityImplPtr ActorImpl::suspend(ActorImpl* issuer)
346 XBT_DEBUG("Actor '%s' is already suspended", get_cname());
352 /* If we are suspending another actor that is waiting on a sync, suspend its synchronization. */
353 if (this != issuer) {
355 waiting_synchro->suspend();
356 /* If the other actor is not waiting, its suspension is delayed to when the actor is rescheduled. */
360 activity::ExecImpl* exec = new activity::ExecImpl();
361 (*exec).set_name("suspend").set_host(host_).set_flops_amount(0.0).start();
362 return activity::ExecImplPtr(exec);
366 void ActorImpl::resume()
368 XBT_IN("actor = %p", this);
370 if (context_->iwannadie) {
371 XBT_VERB("Ignoring request to suspend an actor that is currently dying.");
379 /* resume the synchronization that was blocking the resumed actor. */
381 waiting_synchro->resume();
386 activity::ActivityImplPtr ActorImpl::join(ActorImpl* actor, double timeout)
388 activity::ActivityImplPtr sleep = this->sleep(timeout);
389 SIMIX_process_on_exit(actor, [sleep](bool) {
390 if (sleep->surf_action_)
391 sleep->surf_action_->finish(resource::Action::State::FINISHED);
396 activity::ActivityImplPtr ActorImpl::sleep(double duration)
398 if (not host_->is_on())
399 throw_exception(std::make_exception_ptr(simgrid::HostFailureException(
400 XBT_THROW_POINT, std::string("Host ") + host_->get_cname() + " failed, you cannot sleep there.")));
402 activity::SleepImpl* sleep = new activity::SleepImpl();
403 (*sleep).set_name("sleep").set_host(host_).set_duration(duration).start();
404 return activity::SleepImplPtr(sleep);
407 void ActorImpl::throw_exception(std::exception_ptr e)
414 /* cancel the blocking synchro if any */
415 if (waiting_synchro) {
416 waiting_synchro->cancel();
418 activity::CommImplPtr comm = boost::dynamic_pointer_cast<activity::CommImpl>(waiting_synchro);
419 activity::SleepImplPtr sleep = boost::dynamic_pointer_cast<activity::SleepImpl>(waiting_synchro);
421 if (comm != nullptr) {
425 if (sleep != nullptr) {
426 if (std::find(begin(simix_global->actors_to_run), end(simix_global->actors_to_run), this) ==
427 end(simix_global->actors_to_run) &&
428 this != SIMIX_process_self()) {
429 XBT_DEBUG("Inserting [%p] %s in the to_run list", this, get_cname());
430 simix_global->actors_to_run.push_back(this);
435 waiting_synchro = nullptr;
438 void ActorImpl::set_host(s4u::Host* dest)
440 xbt::intrusive_erase(host_->pimpl_->process_list_, *this);
442 dest->pimpl_->process_list_.push_back(*this);
445 ActorImplPtr ActorImpl::init(const std::string& name, s4u::Host* host)
447 ActorImpl* actor = new ActorImpl(xbt::string(name), host);
448 actor->set_ppid(this->pid_);
450 intrusive_ptr_add_ref(actor);
451 /* The on_creation() signal must be delayed until there, where the pid and everything is set */
452 s4u::Actor::on_creation(*actor->ciface());
454 return ActorImplPtr(actor);
457 ActorImpl* ActorImpl::start(const simix::ActorCode& code)
459 xbt_assert(code && host_ != nullptr, "Invalid parameters");
461 if (not host_->is_on()) {
462 XBT_WARN("Cannot launch actor '%s' on failed host '%s'", name_.c_str(), host_->get_cname());
463 intrusive_ptr_release(this);
464 std::rethrow_exception(
465 std::make_exception_ptr(simgrid::HostFailureException(XBT_THROW_POINT, "Cannot start actor on failed host.")));
469 XBT_VERB("Create context %s", get_cname());
470 context_.reset(simix_global->context_factory->create_context(simix::ActorCode(code), this));
472 XBT_DEBUG("Start context '%s'", get_cname());
474 /* Add the actor to its host's actor list */
475 host_->pimpl_->process_list_.push_back(*this);
476 simix_global->process_list[pid_] = this;
478 /* Now insert it in the global actor list and in the actor to run list */
479 XBT_DEBUG("Inserting [%p] %s(%s) in the to_run list", this, get_cname(), host_->get_cname());
480 simix_global->actors_to_run.push_back(this);
485 ActorImplPtr ActorImpl::create(const std::string& name, const simix::ActorCode& code, void* data, s4u::Host* host,
486 std::unordered_map<std::string, std::string>* properties, ActorImpl* parent_actor)
488 XBT_DEBUG("Start actor %s@'%s'", name.c_str(), host->get_cname());
491 if (parent_actor != nullptr)
492 actor = parent_actor->init(xbt::string(name), host);
494 actor = SIMIX_process_self()->init(xbt::string(name), host);
497 actor->set_user_data(data);
500 if (properties != nullptr)
501 for (auto const& kv : *properties)
502 actor->set_property(kv.first, kv.second);
509 void create_maestro(const std::function<void()>& code)
511 /* Create maestro actor and initialize it */
512 ActorImpl* maestro = new ActorImpl(xbt::string(""), /*host*/ nullptr);
515 maestro->context_.reset(simix_global->context_factory->create_context(simix::ActorCode(), maestro));
517 maestro->context_.reset(simix_global->context_factory->create_maestro(simix::ActorCode(code), maestro));
520 maestro->simcall.issuer = maestro;
521 simix_global->maestro_process = maestro;
525 } // namespace kernel
526 } // namespace simgrid
528 void SIMIX_process_detach()
530 simgrid::kernel::actor::ActorImpl::detach();
533 smx_actor_t SIMIX_process_attach(const char* name, void* data, const char* hostname,
534 std::unordered_map<std::string, std::string>* properties,
535 smx_actor_t /*parent_process*/)
537 return simgrid::kernel::actor::ActorImpl::attach(name, data, sg_host_by_name(hostname), properties).get();
540 /** @deprecated When this function gets removed, also remove the xbt_ex class, that is only there to help users to
542 void SIMIX_process_throw(smx_actor_t actor, xbt_errcat_t cat, int value, const char* msg)
544 SMX_EXCEPTION(actor, cat, value, msg);
546 if (actor->is_suspended())
549 /* cancel the blocking synchro if any */
550 if (actor->waiting_synchro) {
552 simgrid::kernel::activity::ExecImplPtr exec =
553 boost::dynamic_pointer_cast<simgrid::kernel::activity::ExecImpl>(actor->waiting_synchro);
557 simgrid::kernel::activity::CommImplPtr comm =
558 boost::dynamic_pointer_cast<simgrid::kernel::activity::CommImpl>(actor->waiting_synchro);
559 if (comm != nullptr) {
560 actor->comms.remove(comm);
564 simgrid::kernel::activity::SleepImplPtr sleep =
565 boost::dynamic_pointer_cast<simgrid::kernel::activity::SleepImpl>(actor->waiting_synchro);
566 if (sleep != nullptr) {
567 sleep->clean_action();
568 if (std::find(begin(simix_global->actors_to_run), end(simix_global->actors_to_run), actor) ==
569 end(simix_global->actors_to_run) &&
570 actor != SIMIX_process_self()) {
571 XBT_DEBUG("Inserting [%p] %s in the to_run list", actor, actor->get_cname());
572 simix_global->actors_to_run.push_back(actor);
576 simgrid::kernel::activity::RawImplPtr raw =
577 boost::dynamic_pointer_cast<simgrid::kernel::activity::RawImpl>(actor->waiting_synchro);
578 if (raw != nullptr) {
582 simgrid::kernel::activity::IoImplPtr io =
583 boost::dynamic_pointer_cast<simgrid::kernel::activity::IoImpl>(actor->waiting_synchro);
588 actor->waiting_synchro = nullptr;
591 void simcall_HANDLER_process_suspend(smx_simcall_t simcall, smx_actor_t actor)
593 smx_activity_t sync_suspend = actor->suspend(simcall->issuer);
595 if (actor != simcall->issuer) {
596 SIMIX_simcall_answer(simcall);
598 sync_suspend->simcalls_.push_back(simcall);
599 actor->waiting_synchro = sync_suspend;
600 actor->waiting_synchro->suspend();
602 /* If we are suspending ourselves, then just do not finish the simcall now */
605 int SIMIX_process_get_maxpid()
607 return simix_process_maxpid;
610 int SIMIX_process_count()
612 return simix_global->process_list.size();
615 void* SIMIX_process_self_get_data() // deprecated
617 smx_actor_t self = SIMIX_process_self();
619 if (self == nullptr) {
622 return self->get_user_data();
625 void SIMIX_process_self_set_data(void* data) // deprecated
627 SIMIX_process_self()->set_user_data(data);
630 /* needs to be public and without simcall because it is called
631 by exceptions and logging events */
632 const char* SIMIX_process_self_get_name()
635 smx_actor_t process = SIMIX_process_self();
636 if (process == nullptr || process == simix_global->maestro_process)
639 return process->get_cname();
642 void simcall_HANDLER_process_join(smx_simcall_t simcall, smx_actor_t process, double timeout)
644 if (process->finished_) {
645 // The joined process is already finished, just wake up the issuer process right away
646 simcall_process_sleep__set__result(simcall, SIMIX_DONE);
647 SIMIX_simcall_answer(simcall);
650 smx_activity_t sync = simcall->issuer->join(process, timeout);
651 sync->simcalls_.push_back(simcall);
652 simcall->issuer->waiting_synchro = sync;
655 void simcall_HANDLER_process_sleep(smx_simcall_t simcall, double duration)
657 if (MC_is_active() || MC_record_replay_is_active()) {
658 MC_process_clock_add(simcall->issuer, duration);
659 simcall_process_sleep__set__result(simcall, SIMIX_DONE);
660 SIMIX_simcall_answer(simcall);
663 smx_activity_t sync = simcall->issuer->sleep(duration);
664 sync->simcalls_.push_back(simcall);
665 simcall->issuer->waiting_synchro = sync;
669 * @brief Calling this function makes the process to yield.
671 * Only the current process can call this function, giving back the control to maestro.
673 * @param self the current process
676 /** @brief Returns the list of processes to run.
679 const std::vector<smx_actor_t>& simgrid::simix::process_get_runnable()
681 return simix_global->actors_to_run;
684 /** @brief Returns the process from PID. */
685 smx_actor_t SIMIX_process_from_PID(aid_t PID)
687 auto actor = simix_global->process_list.find(PID);
688 return actor == simix_global->process_list.end() ? nullptr : actor->second;
691 void SIMIX_process_on_exit(smx_actor_t actor, int_f_pvoid_pvoid_t fun, void* data)
693 SIMIX_process_on_exit(actor, [fun, data](bool failed) {
694 intptr_t status = failed ? SMX_EXIT_FAILURE : SMX_EXIT_SUCCESS;
695 fun(reinterpret_cast<void*>(status), data);
699 void SIMIX_process_on_exit(smx_actor_t actor, const std::function<void(int, void*)>& fun, void* data)
701 SIMIX_process_on_exit(actor, [fun, data](bool failed) { fun(failed ? SMX_EXIT_FAILURE : SMX_EXIT_SUCCESS, data); });
704 void SIMIX_process_on_exit(smx_actor_t actor, const std::function<void(bool /*failed*/)>& fun)
706 xbt_assert(actor, "current process not found: are you in maestro context ?");
707 actor->on_exit.emplace_back(fun);
710 /** @brief Restart a process, starting it again from the beginning. */
712 * @ingroup simix_process_management
713 * @brief Creates and runs a new SIMIX process.
715 * The structure and the corresponding thread are created and put in the list of ready processes.
717 * @param name a name for the process. It is for user-level information and can be nullptr.
718 * @param code the main function of the process
719 * @param data a pointer to any data one may want to attach to the new object. It is for user-level information and can
721 * It can be retrieved with the method ActorImpl::getUserData().
722 * @param host where the new agent is executed.
723 * @param properties the properties of the process
725 smx_actor_t simcall_process_create(const std::string& name, const simgrid::simix::ActorCode& code, void* data,
726 sg_host_t host, std::unordered_map<std::string, std::string>* properties)
728 smx_actor_t self = SIMIX_process_self();
729 return simgrid::simix::simcall([&name, &code, data, host, properties, self] {
730 return simgrid::kernel::actor::ActorImpl::create(name, code, data, host, properties, self).get();
734 void simcall_process_set_data(smx_actor_t process, void* data)
736 simgrid::simix::simcall([process, data] { process->set_user_data(data); });