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/simix/smx_host_private.hpp"
19 #include "src/simix/smx_io_private.hpp"
20 #include "src/simix/smx_synchro_private.hpp"
21 #include "src/surf/HostImpl.hpp"
22 #include "src/surf/cpu_interface.hpp"
24 #include <boost/range/algorithm.hpp>
26 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_process, simix, "Logging specific to SIMIX (process)");
28 static unsigned long simix_process_maxpid = 0;
31 * @brief Returns the current agent.
33 * This functions returns the currently running SIMIX process.
35 * @return The SIMIX process
37 smx_actor_t SIMIX_process_self()
39 smx_context_t self_context = simgrid::kernel::context::Context::self();
41 return (self_context != nullptr) ? self_context->get_actor() : nullptr;
45 * @brief Returns whether a process has pending asynchronous communications.
46 * @return true if there are asynchronous communications in this process
48 int SIMIX_process_has_pending_comms(smx_actor_t process) {
50 return process->comms.size() > 0;
54 * @brief Moves a process to the list of processes to destroy.
56 void SIMIX_process_cleanup(smx_actor_t process)
58 XBT_DEBUG("Cleanup process %s (%p), waiting synchro %p", process->get_cname(), process,
59 process->waiting_synchro.get());
61 simix_global->mutex.lock();
63 simix_global->process_list.erase(process->pid_);
64 if (process->host_ && process->host_process_list_hook.is_linked())
65 simgrid::xbt::intrusive_erase(process->host_->pimpl_->process_list_, *process);
66 if (not process->smx_destroy_list_hook.is_linked()) {
68 xbt_dynar_push_as(simix_global->dead_actors_vector, smx_actor_t, process);
70 simix_global->actors_to_destroy.push_back(*process);
72 process->context_->iwannadie = false;
74 simix_global->mutex.unlock();
81 ActorImpl::ActorImpl(simgrid::xbt::string name, simgrid::s4u::Host* host) : name_(name), host_(host), piface_(this)
83 pid_ = simix_process_maxpid++;
84 simcall.issuer = this;
87 ActorImpl::~ActorImpl()
89 delete this->context_;
92 void ActorImpl::exit()
94 context_->iwannadie = true;
99 // Forcefully kill the actor if its host is turned off. Not a HostFailureException because you should not survive that
100 if (not host_->is_on())
101 this->throw_exception(std::make_exception_ptr(simgrid::kernel::context::StopRequest("host failed")));
103 /* destroy the blocking synchro if any */
104 if (waiting_synchro != nullptr) {
106 activity::ExecImplPtr exec = boost::dynamic_pointer_cast<activity::ExecImpl>(waiting_synchro);
107 activity::CommImplPtr comm = boost::dynamic_pointer_cast<activity::CommImpl>(waiting_synchro);
108 activity::SleepImplPtr sleep = boost::dynamic_pointer_cast<activity::SleepImpl>(waiting_synchro);
109 activity::RawImplPtr raw = boost::dynamic_pointer_cast<activity::RawImpl>(waiting_synchro);
110 activity::IoImplPtr io = boost::dynamic_pointer_cast<activity::IoImpl>(waiting_synchro);
112 if (exec != nullptr && exec->surf_action_) {
114 exec->surf_action_->unref();
115 exec->surf_action_ = nullptr;
116 } else if (comm != nullptr) {
117 comms.remove(waiting_synchro);
119 // Remove first occurrence of &process->simcall:
120 auto i = boost::range::find(waiting_synchro->simcalls_, &simcall);
121 if (i != waiting_synchro->simcalls_.end())
122 waiting_synchro->simcalls_.remove(&simcall);
123 } else if (sleep != nullptr) {
124 if (sleep->surf_action_)
125 sleep->surf_action_->cancel();
127 } else if (raw != nullptr) {
128 SIMIX_synchro_stop_waiting(this, &simcall);
129 } else if (io != nullptr) {
132 simgrid::kernel::activity::ActivityImplPtr activity = waiting_synchro;
133 xbt_die("Activity %s is of unknown type %s", activity->get_cname(),
134 simgrid::xbt::demangle(typeid(activity).name()).get());
137 waiting_synchro = nullptr;
141 void ActorImpl::kill(smx_actor_t actor)
143 if (actor->finished_) {
144 XBT_DEBUG("Ignoring request to kill process %s@%s that is already dead", actor->get_cname(),
145 actor->host_->get_cname());
149 XBT_DEBUG("Actor '%s'@%s is killing actor '%s'@%s", get_cname(), host_->get_cname(), actor->get_cname(),
150 actor->host_->get_cname());
154 if (std::find(begin(simix_global->actors_to_run), end(simix_global->actors_to_run), actor) ==
155 end(simix_global->actors_to_run) &&
157 XBT_DEBUG("Inserting %s in the to_run list", actor->get_cname());
158 simix_global->actors_to_run.push_back(actor);
162 void ActorImpl::kill_all()
164 for (auto const& kv : simix_global->process_list)
165 if (kv.second != this)
166 this->kill(kv.second);
169 void ActorImpl::set_kill_time(double kill_time)
171 if (kill_time <= SIMIX_get_clock())
173 XBT_DEBUG("Set kill time %f for process %s@%s", kill_time, get_cname(), host_->get_cname());
174 kill_timer = SIMIX_timer_set(kill_time, [this] {
176 kill_timer = nullptr;
180 static void dying_daemon(int /*exit_status*/, void* data)
182 std::vector<ActorImpl*>* vect = &simix_global->daemons;
184 auto it = std::find(vect->begin(), vect->end(), static_cast<ActorImpl*>(data));
185 xbt_assert(it != vect->end(), "The dying daemon is not a daemon after all. Please report that bug.");
187 /* Don't move the whole content since we don't really care about the order */
188 std::swap(*it, vect->back());
192 void ActorImpl::yield()
194 XBT_DEBUG("Yield actor '%s'", get_cname());
196 /* Go into sleep and return control to maestro */
199 /* Ok, maestro returned control to us */
200 XBT_DEBUG("Control returned to me: '%s'", get_cname());
202 if (context_->iwannadie) {
204 XBT_DEBUG("Actor %s@%s is dead", get_cname(), host_->get_cname());
205 // throw simgrid::kernel::context::StopRequest(); Does not seem to properly kill the actor
211 XBT_DEBUG("Hey! I'm suspended.");
212 xbt_assert(exception_ != nullptr, "Gasp! This exception may be lost by subsequent calls.");
217 if (exception_ != nullptr) {
218 XBT_DEBUG("Wait, maestro left me an exception");
219 std::exception_ptr exception = std::move(exception_);
220 exception_ = nullptr;
221 std::rethrow_exception(std::move(exception));
224 if (SMPI_switch_data_segment && not finished_) {
225 SMPI_switch_data_segment(iface());
229 /** This process will be terminated automatically when the last non-daemon process finishes */
230 void ActorImpl::daemonize()
234 simix_global->daemons.push_back(this);
235 SIMIX_process_on_exit(this, dying_daemon, this);
239 simgrid::s4u::Actor* ActorImpl::restart()
241 XBT_DEBUG("Restarting process %s on %s", get_cname(), host_->get_cname());
243 // retrieve the arguments of the old process
244 simgrid::kernel::actor::ProcessArg arg = ProcessArg(host_, this);
246 // kill the old process
247 (this == simix_global->maestro_process) ? this->exit() : SIMIX_process_self()->kill(this);
249 // start the new process
251 ActorImpl::create(arg.name, std::move(arg.code), arg.data, arg.host, arg.properties.get(), nullptr);
252 actor->set_kill_time(arg.kill_time);
253 actor->set_auto_restart(arg.auto_restart);
255 return actor->ciface();
258 smx_activity_t ActorImpl::suspend(ActorImpl* issuer)
261 XBT_DEBUG("Actor '%s' is already suspended", get_cname());
267 /* If we are suspending another actor that is waiting on a sync, suspend its synchronization. */
268 if (this != issuer) {
270 waiting_synchro->suspend();
271 /* If the other actor is not waiting, its suspension is delayed to when the actor is rescheduled. */
275 return activity::ExecImplPtr(new activity::ExecImpl("suspend", "", nullptr, this->host_))->start(0.0, 1.0, 0.0);
279 void ActorImpl::resume()
281 XBT_IN("actor = %p", this);
283 if (context_->iwannadie) {
284 XBT_VERB("Ignoring request to suspend an actor that is currently dying.");
292 /* resume the synchronization that was blocking the resumed actor. */
294 waiting_synchro->resume();
299 smx_activity_t ActorImpl::join(smx_actor_t actor, double timeout)
301 smx_activity_t res = this->sleep(timeout);
302 intrusive_ptr_add_ref(res.get());
303 SIMIX_process_on_exit(actor,
305 auto sleep = static_cast<simgrid::kernel::activity::SleepImpl*>(arg);
306 if (sleep->surf_action_)
307 sleep->surf_action_->finish(simgrid::kernel::resource::Action::State::FINISHED);
308 intrusive_ptr_release(sleep);
313 smx_activity_t ActorImpl::sleep(double duration)
315 if (not host_->is_on())
316 throw_exception(std::make_exception_ptr(simgrid::HostFailureException(
317 XBT_THROW_POINT, std::string("Host ") + host_->get_cname() + " failed, you cannot sleep there.")));
319 return simgrid::kernel::activity::SleepImplPtr(new simgrid::kernel::activity::SleepImpl("sleep", host_))
323 void ActorImpl::throw_exception(std::exception_ptr e)
330 /* cancel the blocking synchro if any */
331 if (waiting_synchro) {
333 simgrid::kernel::activity::ExecImplPtr exec =
334 boost::dynamic_pointer_cast<simgrid::kernel::activity::ExecImpl>(waiting_synchro);
338 simgrid::kernel::activity::CommImplPtr comm =
339 boost::dynamic_pointer_cast<simgrid::kernel::activity::CommImpl>(waiting_synchro);
340 if (comm != nullptr) {
345 simgrid::kernel::activity::SleepImplPtr sleep =
346 boost::dynamic_pointer_cast<simgrid::kernel::activity::SleepImpl>(waiting_synchro);
347 if (sleep != nullptr) {
348 SIMIX_process_sleep_destroy(waiting_synchro);
349 if (std::find(begin(simix_global->actors_to_run), end(simix_global->actors_to_run), this) ==
350 end(simix_global->actors_to_run) &&
351 this != SIMIX_process_self()) {
352 XBT_DEBUG("Inserting [%p] %s in the to_run list", this, get_cname());
353 simix_global->actors_to_run.push_back(this);
357 simgrid::kernel::activity::RawImplPtr raw =
358 boost::dynamic_pointer_cast<simgrid::kernel::activity::RawImpl>(waiting_synchro);
359 if (raw != nullptr) {
360 SIMIX_synchro_stop_waiting(this, &simcall);
363 simgrid::kernel::activity::IoImplPtr io =
364 boost::dynamic_pointer_cast<simgrid::kernel::activity::IoImpl>(waiting_synchro);
369 waiting_synchro = nullptr;
372 void ActorImpl::set_host(sg_host_t dest)
374 simgrid::xbt::intrusive_erase(host_->pimpl_->process_list_, *this);
376 dest->pimpl_->process_list_.push_back(*this);
379 ActorImplPtr ActorImpl::create(std::string name, simgrid::simix::ActorCode code, void* data, simgrid::s4u::Host* host,
380 std::unordered_map<std::string, std::string>* properties, smx_actor_t parent_actor)
383 XBT_DEBUG("Start actor %s@'%s'", name.c_str(), host->get_cname());
385 if (not host->is_on()) {
386 XBT_WARN("Cannot launch actor '%s' on failed host '%s'", name.c_str(), host->get_cname());
390 ActorImpl* actor = new simgrid::kernel::actor::ActorImpl(simgrid::xbt::string(name), host);
392 xbt_assert(code && host != nullptr, "Invalid parameters");
394 actor->set_user_data(data);
397 if (parent_actor != nullptr)
398 actor->ppid_ = parent_actor->pid_;
400 XBT_VERB("Create context %s", actor->get_cname());
401 actor->context_ = SIMIX_context_new(std::move(code), &SIMIX_process_cleanup, actor);
404 if (properties != nullptr)
405 for (auto const& kv : *properties)
406 actor->set_property(kv.first, kv.second);
408 /* Add the process to its host's process list */
409 host->pimpl_->process_list_.push_back(*actor);
411 XBT_DEBUG("Start context '%s'", actor->get_cname());
413 /* Now insert it in the global process list and in the process to run list */
414 simix_global->process_list[actor->pid_] = actor;
415 XBT_DEBUG("Inserting [%p] %s(%s) in the to_run list", actor, actor->get_cname(), host->get_cname());
416 simix_global->actors_to_run.push_back(actor);
417 intrusive_ptr_add_ref(actor);
419 /* The on_creation() signal must be delayed until there, where the pid and everything is set */
420 simgrid::s4u::Actor::on_creation(actor->iface());
422 return ActorImplPtr(actor);
425 void create_maestro(simgrid::simix::ActorCode code)
427 /* Create maestro process and initialize it */
428 smx_actor_t maestro = new simgrid::kernel::actor::ActorImpl(simgrid::xbt::string(""), /*host*/ nullptr);
431 maestro->context_ = SIMIX_context_new(simgrid::simix::ActorCode(), nullptr, maestro);
433 maestro->context_ = simix_global->context_factory->create_maestro(code, maestro);
436 maestro->simcall.issuer = maestro;
437 simix_global->maestro_process = maestro;
441 } // namespace kernel
444 smx_actor_t SIMIX_process_attach(const char* name, void* data, const char* hostname,
445 std::unordered_map<std::string, std::string>* properties, smx_actor_t parent_process)
447 // This is mostly a copy/paste from SIMIX_process_new(),
448 // it'd be nice to share some code between those two functions.
450 sg_host_t host = sg_host_by_name(hostname);
451 XBT_DEBUG("Attach process %s on host '%s'", name, hostname);
453 if (not host->is_on()) {
454 XBT_WARN("Cannot launch process '%s' on failed host '%s'", name, hostname);
458 smx_actor_t actor = new simgrid::kernel::actor::ActorImpl(simgrid::xbt::string(name), host);
460 actor->set_user_data(data);
461 actor->code = nullptr;
463 if (parent_process != nullptr)
464 actor->ppid_ = parent_process->pid_;
466 XBT_VERB("Create context %s", actor->get_cname());
467 xbt_assert(simix_global != nullptr, "simix is not initialized, please call MSG_init first");
468 actor->context_ = simix_global->context_factory->attach(&SIMIX_process_cleanup, actor);
471 if (properties != nullptr)
472 for (auto const& kv : *properties)
473 actor->set_property(kv.first, kv.second);
475 /* Add the process to it's host process list */
476 host->pimpl_->process_list_.push_back(*actor);
478 /* Now insert it in the global process list and in the process to run list */
479 simix_global->process_list[actor->pid_] = actor;
480 XBT_DEBUG("Inserting [%p] %s(%s) in the to_run list", actor, actor->get_cname(), host->get_cname());
481 simix_global->actors_to_run.push_back(actor);
482 intrusive_ptr_add_ref(actor);
484 auto* context = dynamic_cast<simgrid::kernel::context::AttachContext*>(actor->context_);
485 xbt_assert(nullptr != context, "Not a suitable context");
486 context->attach_start();
488 /* The on_creation() signal must be delayed until there, where the pid and everything is set */
489 simgrid::s4u::ActorPtr tmp = actor->iface(); // Passing this directly to on_creation will lead to crashes
490 simgrid::s4u::Actor::on_creation(tmp);
495 void SIMIX_process_detach()
497 auto* context = dynamic_cast<simgrid::kernel::context::AttachContext*>(simgrid::kernel::context::Context::self());
498 if (context == nullptr)
499 xbt_die("Not a suitable context");
501 SIMIX_process_cleanup(context->get_actor());
502 context->attach_stop();
506 /** @deprecated When this function gets removed, also remove the xbt_ex class, that is only there to help users to
508 void SIMIX_process_throw(smx_actor_t actor, xbt_errcat_t cat, int value, const char* msg)
510 SMX_EXCEPTION(actor, cat, value, msg);
512 if (actor->suspended_)
515 /* cancel the blocking synchro if any */
516 if (actor->waiting_synchro) {
518 simgrid::kernel::activity::ExecImplPtr exec =
519 boost::dynamic_pointer_cast<simgrid::kernel::activity::ExecImpl>(actor->waiting_synchro);
523 simgrid::kernel::activity::CommImplPtr comm =
524 boost::dynamic_pointer_cast<simgrid::kernel::activity::CommImpl>(actor->waiting_synchro);
525 if (comm != nullptr) {
526 actor->comms.remove(comm);
530 simgrid::kernel::activity::SleepImplPtr sleep =
531 boost::dynamic_pointer_cast<simgrid::kernel::activity::SleepImpl>(actor->waiting_synchro);
532 if (sleep != nullptr) {
533 SIMIX_process_sleep_destroy(actor->waiting_synchro);
534 if (std::find(begin(simix_global->actors_to_run), end(simix_global->actors_to_run), actor) ==
535 end(simix_global->actors_to_run) &&
536 actor != SIMIX_process_self()) {
537 XBT_DEBUG("Inserting [%p] %s in the to_run list", actor, actor->get_cname());
538 simix_global->actors_to_run.push_back(actor);
542 simgrid::kernel::activity::RawImplPtr raw =
543 boost::dynamic_pointer_cast<simgrid::kernel::activity::RawImpl>(actor->waiting_synchro);
544 if (raw != nullptr) {
545 SIMIX_synchro_stop_waiting(actor, &actor->simcall);
548 simgrid::kernel::activity::IoImplPtr io =
549 boost::dynamic_pointer_cast<simgrid::kernel::activity::IoImpl>(actor->waiting_synchro);
554 actor->waiting_synchro = nullptr;
557 void simcall_HANDLER_process_suspend(smx_simcall_t simcall, smx_actor_t actor)
559 smx_activity_t sync_suspend = actor->suspend(simcall->issuer);
561 if (actor != simcall->issuer) {
562 SIMIX_simcall_answer(simcall);
564 sync_suspend->simcalls_.push_back(simcall);
565 actor->waiting_synchro = sync_suspend;
566 actor->waiting_synchro->suspend();
568 /* If we are suspending ourselves, then just do not finish the simcall now */
571 int SIMIX_process_get_maxpid() {
572 return simix_process_maxpid;
575 int SIMIX_process_count()
577 return simix_global->process_list.size();
580 void* SIMIX_process_self_get_data()
582 smx_actor_t self = SIMIX_process_self();
584 if (self == nullptr) {
587 return self->get_user_data();
590 void SIMIX_process_self_set_data(void *data)
592 SIMIX_process_self()->set_user_data(data);
596 /* needs to be public and without simcall because it is called
597 by exceptions and logging events */
598 const char* SIMIX_process_self_get_name() {
600 smx_actor_t process = SIMIX_process_self();
601 if (process == nullptr || process == simix_global->maestro_process)
604 return process->get_cname();
607 void simcall_HANDLER_process_join(smx_simcall_t simcall, smx_actor_t process, double timeout)
609 if (process->finished_) {
610 // The joined process is already finished, just wake up the issuer process right away
611 simcall_process_sleep__set__result(simcall, SIMIX_DONE);
612 SIMIX_simcall_answer(simcall);
615 smx_activity_t sync = simcall->issuer->join(process, timeout);
616 sync->simcalls_.push_back(simcall);
617 simcall->issuer->waiting_synchro = sync;
620 void simcall_HANDLER_process_sleep(smx_simcall_t simcall, double duration)
622 if (MC_is_active() || MC_record_replay_is_active()) {
623 MC_process_clock_add(simcall->issuer, duration);
624 simcall_process_sleep__set__result(simcall, SIMIX_DONE);
625 SIMIX_simcall_answer(simcall);
628 smx_activity_t sync = simcall->issuer->sleep(duration);
629 sync->simcalls_.push_back(simcall);
630 simcall->issuer->waiting_synchro = sync;
633 void SIMIX_process_sleep_destroy(smx_activity_t synchro)
635 XBT_DEBUG("Destroy sleep synchro %p", synchro.get());
636 simgrid::kernel::activity::SleepImplPtr sleep =
637 boost::static_pointer_cast<simgrid::kernel::activity::SleepImpl>(synchro);
639 if (sleep->surf_action_) {
640 sleep->surf_action_->unref();
641 sleep->surf_action_ = nullptr;
646 * @brief Calling this function makes the process to yield.
648 * Only the current process can call this function, giving back the control to maestro.
650 * @param self the current process
653 /** @brief Returns the list of processes to run.
656 const std::vector<smx_actor_t>& simgrid::simix::process_get_runnable()
658 return simix_global->actors_to_run;
661 /** @brief Returns the process from PID. */
662 smx_actor_t SIMIX_process_from_PID(aid_t PID)
664 auto actor = simix_global->process_list.find(PID);
665 return actor == simix_global->process_list.end() ? nullptr : actor->second;
668 void SIMIX_process_on_exit(smx_actor_t actor, int_f_pvoid_pvoid_t fun, void* data)
670 SIMIX_process_on_exit(actor, [fun](int a, void* b) { fun((void*)(intptr_t)a, b); }, data);
673 void SIMIX_process_on_exit(smx_actor_t actor, std::function<void(int, void*)> fun, void* data)
675 xbt_assert(actor, "current process not found: are you in maestro context ?");
677 actor->on_exit.emplace_back(s_smx_process_exit_fun_t{fun, data});
680 /** @brief Restart a process, starting it again from the beginning. */
682 * @ingroup simix_process_management
683 * @brief Creates and runs a new SIMIX process.
685 * The structure and the corresponding thread are created and put in the list of ready processes.
687 * @param name a name for the process. It is for user-level information and can be nullptr.
688 * @param code the main function of the process
689 * @param data a pointer to any data one may want to attach to the new object. It is for user-level information and can
691 * It can be retrieved with the method ActorImpl::getUserData().
692 * @param host where the new agent is executed.
693 * @param properties the properties of the process
695 smx_actor_t simcall_process_create(std::string name, simgrid::simix::ActorCode code, void* data, sg_host_t host,
696 std::unordered_map<std::string, std::string>* properties)
698 smx_actor_t self = SIMIX_process_self();
699 return simgrid::simix::simcall([name, code, data, host, properties, self] {
700 return simgrid::kernel::actor::ActorImpl::create(std::move(name), std::move(code), data, host, properties, self)
705 void simcall_process_set_data(smx_actor_t process, void* data)
707 simgrid::simix::simcall([process, data] { process->set_user_data(data); });