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 const 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 actor->set_properties(*properties);
98 /* Add the process to it's host process list */
99 host->pimpl_->process_list_.push_back(*actor);
101 /* Now insert it in the global process list and in the process to run list */
102 simix_global->process_list[actor->get_pid()] = actor;
103 XBT_DEBUG("Inserting [%p] %s(%s) in the to_run list", actor, actor->get_cname(), host->get_cname());
104 simix_global->actors_to_run.push_back(actor);
105 intrusive_ptr_add_ref(actor);
107 auto* context = dynamic_cast<simgrid::kernel::context::AttachContext*>(actor->context_.get());
108 xbt_assert(nullptr != context, "Not a suitable context");
109 context->attach_start();
111 /* The on_creation() signal must be delayed until there, where the pid and everything is set */
112 simgrid::s4u::Actor::on_creation(*actor->ciface());
114 return ActorImplPtr(actor);
116 /** @brief Detach an actor attached with `attach()`
118 * This is called when the current actor has finished its job.
119 * Used in the main thread, it waits for the simulation to finish before returning. When it returns, the other
120 * simulated actors and the maestro are destroyed.
122 void ActorImpl::detach()
124 auto* context = dynamic_cast<context::AttachContext*>(context::Context::self());
125 if (context == nullptr)
126 xbt_die("Not a suitable context");
128 context->get_actor()->cleanup();
129 context->attach_stop();
132 void ActorImpl::cleanup()
136 if (has_to_auto_restart() && not get_host()->is_on()) {
137 XBT_DEBUG("Insert host %s to watched_hosts because it's off and %s needs to restart", get_host()->get_cname(),
139 watched_hosts.insert(get_host()->get_name());
142 // Execute the termination callbacks
143 bool failed = context_->iwannadie;
144 for (auto exit_fun = on_exit->crbegin(); exit_fun != on_exit->crend(); ++exit_fun)
146 if (not has_to_auto_restart())
149 /* cancel non-blocking activities */
150 for (auto activity : comms)
151 boost::static_pointer_cast<activity::CommImpl>(activity)->cancel();
154 XBT_DEBUG("%s@%s(%ld) should not run anymore", get_cname(), get_host()->get_cname(), get_pid());
156 if (this == simix_global->maestro_process) /* Do not cleanup maestro */
159 XBT_DEBUG("Cleanup actor %s (%p), waiting synchro %p", get_cname(), this, waiting_synchro.get());
161 /* Unregister from the kill timer if any */
162 if (kill_timer != nullptr) {
163 kill_timer->remove();
164 kill_timer = nullptr;
167 simix_global->mutex.lock();
169 simix_global->process_list.erase(pid_);
170 if (host_ && host_process_list_hook.is_linked())
171 simgrid::xbt::intrusive_erase(host_->pimpl_->process_list_, *this);
172 if (not smx_destroy_list_hook.is_linked()) {
174 xbt_dynar_push_as(simix_global->dead_actors_vector, ActorImpl*, this);
176 simix_global->actors_to_destroy.push_back(*this);
179 simix_global->mutex.unlock();
181 context_->iwannadie = false; // don't let the simcall's yield() do a Context::stop(), to avoid infinite loops
182 simgrid::simix::simcall([this] { simgrid::s4u::Actor::on_destruction(*ciface()); });
183 context_->iwannadie = true;
186 void ActorImpl::exit()
188 context_->iwannadie = true;
190 exception_ = nullptr;
192 // Forcefully kill the actor if its host is turned off. Not a HostFailureException because you should not survive that
193 if (not host_->is_on())
194 this->throw_exception(std::make_exception_ptr(ForcefulKillException("host failed")));
196 /* destroy the blocking synchro if any */
197 if (waiting_synchro != nullptr) {
198 waiting_synchro->cancel();
199 waiting_synchro->state_ = SIMIX_FAILED;
201 activity::ExecImplPtr exec = boost::dynamic_pointer_cast<activity::ExecImpl>(waiting_synchro);
202 activity::CommImplPtr comm = boost::dynamic_pointer_cast<activity::CommImpl>(waiting_synchro);
204 if (exec != nullptr) {
205 exec->clean_action();
206 } else if (comm != nullptr) {
207 comms.remove(waiting_synchro);
208 // Remove first occurrence of &actor->simcall:
209 auto i = boost::range::find(waiting_synchro->simcalls_, &simcall);
210 if (i != waiting_synchro->simcalls_.end())
211 waiting_synchro->simcalls_.remove(&simcall);
213 activity::ActivityImplPtr(waiting_synchro)->finish();
216 waiting_synchro = nullptr;
220 void ActorImpl::kill(ActorImpl* actor)
222 if (actor->finished_) {
223 XBT_DEBUG("Ignoring request to kill actor %s@%s that is already dead", actor->get_cname(),
224 actor->host_->get_cname());
228 XBT_DEBUG("Actor '%s'@%s is killing actor '%s'@%s", get_cname(), host_ ? host_->get_cname() : "", actor->get_cname(),
229 actor->host_ ? actor->host_->get_cname() : "");
233 if (std::find(begin(simix_global->actors_to_run), end(simix_global->actors_to_run), actor) ==
234 end(simix_global->actors_to_run) &&
236 XBT_DEBUG("Inserting %s in the to_run list", actor->get_cname());
237 simix_global->actors_to_run.push_back(actor);
241 void ActorImpl::kill_all()
243 for (auto const& kv : simix_global->process_list)
244 if (kv.second != this)
245 this->kill(kv.second);
248 void ActorImpl::set_kill_time(double kill_time)
250 if (kill_time <= SIMIX_get_clock())
252 XBT_DEBUG("Set kill time %f for actor %s@%s", kill_time, get_cname(), host_->get_cname());
253 kill_timer = simix::Timer::set(kill_time, [this] {
255 kill_timer = nullptr;
259 double ActorImpl::get_kill_time()
261 return kill_timer ? kill_timer->get_date() : 0;
264 void ActorImpl::yield()
266 XBT_DEBUG("Yield actor '%s'", get_cname());
268 /* Go into sleep and return control to maestro */
271 /* Ok, maestro returned control to us */
272 XBT_DEBUG("Control returned to me: '%s'", get_cname());
274 if (context_->iwannadie) {
275 XBT_DEBUG("Actor %s@%s is dead", get_cname(), host_->get_cname());
276 // throw simgrid::kernel::context::ForcefulKillException(); Does not seem to properly kill the actor
282 XBT_DEBUG("Hey! I'm suspended.");
284 xbt_assert(exception_ == nullptr, "Gasp! This exception may be lost by subsequent calls.");
289 if (exception_ != nullptr) {
290 XBT_DEBUG("Wait, maestro left me an exception");
291 std::exception_ptr exception = std::move(exception_);
292 exception_ = nullptr;
293 std::rethrow_exception(std::move(exception));
296 if (SMPI_switch_data_segment && not finished_) {
297 SMPI_switch_data_segment(iface());
301 /** This actor will be terminated automatically when the last non-daemon actor finishes */
302 void ActorImpl::daemonize()
306 simix_global->daemons.push_back(this);
307 SIMIX_process_on_exit(this, [this](bool) {
308 auto& vect = simix_global->daemons;
309 auto it = std::find(vect.begin(), vect.end(), this);
310 xbt_assert(it != vect.end(), "The dying daemon is not a daemon after all. Please report that bug.");
312 /* Don't move the whole content since we don't really care about the order */
313 std::swap(*it, vect.back());
319 s4u::Actor* ActorImpl::restart()
321 xbt_assert(this != simix_global->maestro_process, "Restarting maestro is not supported");
323 XBT_DEBUG("Restarting actor %s on %s", get_cname(), host_->get_cname());
325 // retrieve the arguments of the old actor
326 ProcessArg arg = ProcessArg(host_, this);
328 // kill the old actor
329 context::Context::self()->get_actor()->kill(this);
331 // start the new actor
333 ActorImpl::create(arg.name, std::move(arg.code), arg.data, arg.host, arg.properties.get(), nullptr);
334 *actor->on_exit = std::move(*arg.on_exit);
335 actor->set_kill_time(arg.kill_time);
336 actor->set_auto_restart(arg.auto_restart);
338 return actor->ciface();
341 activity::ActivityImplPtr ActorImpl::suspend(ActorImpl* issuer)
344 XBT_DEBUG("Actor '%s' is already suspended", get_cname());
350 /* If we are suspending another actor that is waiting on a sync, suspend its synchronization. */
351 if (this != issuer) {
353 waiting_synchro->suspend();
354 /* If the other actor is not waiting, its suspension is delayed to when the actor is rescheduled. */
358 activity::ExecImpl* exec = new activity::ExecImpl();
359 (*exec).set_name("suspend").set_host(host_).set_flops_amount(0.0).start();
360 return activity::ExecImplPtr(exec);
364 void ActorImpl::resume()
366 XBT_IN("actor = %p", this);
368 if (context_->iwannadie) {
369 XBT_VERB("Ignoring request to suspend an actor that is currently dying.");
377 /* resume the synchronization that was blocking the resumed actor. */
379 waiting_synchro->resume();
384 activity::ActivityImplPtr ActorImpl::join(ActorImpl* actor, double timeout)
386 activity::ActivityImplPtr sleep = this->sleep(timeout);
387 SIMIX_process_on_exit(actor, [sleep](bool) {
388 if (sleep->surf_action_)
389 sleep->surf_action_->finish(resource::Action::State::FINISHED);
394 activity::ActivityImplPtr ActorImpl::sleep(double duration)
396 if (not host_->is_on())
397 throw_exception(std::make_exception_ptr(simgrid::HostFailureException(
398 XBT_THROW_POINT, std::string("Host ") + host_->get_cname() + " failed, you cannot sleep there.")));
400 activity::SleepImpl* sleep = new activity::SleepImpl();
401 (*sleep).set_name("sleep").set_host(host_).set_duration(duration).start();
402 return activity::SleepImplPtr(sleep);
405 void ActorImpl::throw_exception(std::exception_ptr e)
412 /* cancel the blocking synchro if any */
413 if (waiting_synchro) {
414 waiting_synchro->cancel();
416 activity::CommImplPtr comm = boost::dynamic_pointer_cast<activity::CommImpl>(waiting_synchro);
421 waiting_synchro = nullptr;
425 void ActorImpl::set_host(s4u::Host* dest)
427 xbt::intrusive_erase(host_->pimpl_->process_list_, *this);
429 dest->pimpl_->process_list_.push_back(*this);
432 ActorImplPtr ActorImpl::init(const std::string& name, s4u::Host* host)
434 ActorImpl* actor = new ActorImpl(xbt::string(name), host);
435 actor->set_ppid(this->pid_);
437 intrusive_ptr_add_ref(actor);
438 /* The on_creation() signal must be delayed until there, where the pid and everything is set */
439 s4u::Actor::on_creation(*actor->ciface());
441 return ActorImplPtr(actor);
444 ActorImpl* ActorImpl::start(const simix::ActorCode& code)
446 xbt_assert(code && host_ != nullptr, "Invalid parameters");
448 if (not host_->is_on()) {
449 XBT_WARN("Cannot launch actor '%s' on failed host '%s'", name_.c_str(), host_->get_cname());
450 intrusive_ptr_release(this);
451 std::rethrow_exception(
452 std::make_exception_ptr(simgrid::HostFailureException(XBT_THROW_POINT, "Cannot start actor on failed host.")));
456 XBT_VERB("Create context %s", get_cname());
457 context_.reset(simix_global->context_factory->create_context(simix::ActorCode(code), this));
459 XBT_DEBUG("Start context '%s'", get_cname());
461 /* Add the actor to its host's actor list */
462 host_->pimpl_->process_list_.push_back(*this);
463 simix_global->process_list[pid_] = this;
465 /* Now insert it in the global actor list and in the actor to run list */
466 XBT_DEBUG("Inserting [%p] %s(%s) in the to_run list", this, get_cname(), host_->get_cname());
467 simix_global->actors_to_run.push_back(this);
472 ActorImplPtr ActorImpl::create(const std::string& name, const simix::ActorCode& code, void* data, s4u::Host* host,
473 const std::unordered_map<std::string, std::string>* properties, ActorImpl* parent_actor)
475 XBT_DEBUG("Start actor %s@'%s'", name.c_str(), host->get_cname());
478 if (parent_actor != nullptr)
479 actor = parent_actor->init(xbt::string(name), host);
481 actor = SIMIX_process_self()->init(xbt::string(name), host);
484 actor->set_user_data(data);
487 if (properties != nullptr)
488 actor->set_properties(*properties);
495 void create_maestro(const std::function<void()>& code)
497 /* Create maestro actor and initialize it */
498 ActorImpl* maestro = new ActorImpl(xbt::string(""), /*host*/ nullptr);
501 maestro->context_.reset(simix_global->context_factory->create_context(simix::ActorCode(), maestro));
503 maestro->context_.reset(simix_global->context_factory->create_maestro(simix::ActorCode(code), maestro));
506 maestro->simcall.issuer = maestro;
507 simix_global->maestro_process = maestro;
511 } // namespace kernel
512 } // namespace simgrid
514 void SIMIX_process_detach()
516 simgrid::kernel::actor::ActorImpl::detach();
519 smx_actor_t SIMIX_process_attach(const char* name, void* data, const char* hostname,
520 std::unordered_map<std::string, std::string>* properties,
521 smx_actor_t /*parent_process*/)
523 return simgrid::kernel::actor::ActorImpl::attach(name, data, sg_host_by_name(hostname), properties).get();
526 /** @deprecated When this function gets removed, also remove the xbt_ex class, that is only there to help users to
528 void SIMIX_process_throw(smx_actor_t actor, xbt_errcat_t cat, int value, const char* msg)
530 xbt_ex e(XBT_THROW_POINT, msg);
533 actor->throw_exception(std::make_exception_ptr(e));
536 void simcall_HANDLER_process_suspend(smx_simcall_t simcall, smx_actor_t actor)
538 smx_activity_t sync_suspend = actor->suspend(simcall->issuer);
540 if (actor != simcall->issuer) {
541 SIMIX_simcall_answer(simcall);
543 sync_suspend->simcalls_.push_back(simcall);
544 actor->waiting_synchro = sync_suspend;
545 actor->waiting_synchro->suspend();
547 /* If we are suspending ourselves, then just do not finish the simcall now */
550 int SIMIX_process_get_maxpid()
552 return simix_process_maxpid;
555 int SIMIX_process_count()
557 return simix_global->process_list.size();
560 void* SIMIX_process_self_get_data() // deprecated
562 smx_actor_t self = SIMIX_process_self();
564 if (self == nullptr) {
567 return self->get_user_data();
570 void SIMIX_process_self_set_data(void* data) // deprecated
572 SIMIX_process_self()->set_user_data(data);
575 /* needs to be public and without simcall because it is called
576 by exceptions and logging events */
577 const char* SIMIX_process_self_get_name()
580 smx_actor_t process = SIMIX_process_self();
581 if (process == nullptr || process == simix_global->maestro_process)
584 return process->get_cname();
587 void simcall_HANDLER_process_join(smx_simcall_t simcall, smx_actor_t process, double timeout)
589 if (process->finished_) {
590 // The joined process is already finished, just wake up the issuer process right away
591 simcall_process_sleep__set__result(simcall, SIMIX_DONE);
592 SIMIX_simcall_answer(simcall);
595 smx_activity_t sync = simcall->issuer->join(process, timeout);
596 sync->simcalls_.push_back(simcall);
597 simcall->issuer->waiting_synchro = sync;
600 void simcall_HANDLER_process_sleep(smx_simcall_t simcall, double duration)
602 if (MC_is_active() || MC_record_replay_is_active()) {
603 MC_process_clock_add(simcall->issuer, duration);
604 simcall_process_sleep__set__result(simcall, SIMIX_DONE);
605 SIMIX_simcall_answer(simcall);
608 smx_activity_t sync = simcall->issuer->sleep(duration);
609 sync->simcalls_.push_back(simcall);
610 simcall->issuer->waiting_synchro = sync;
614 * @brief Calling this function makes the process to yield.
616 * Only the current process can call this function, giving back the control to maestro.
618 * @param self the current process
621 /** @brief Returns the list of processes to run.
624 const std::vector<smx_actor_t>& simgrid::simix::process_get_runnable()
626 return simix_global->actors_to_run;
629 /** @brief Returns the process from PID. */
630 smx_actor_t SIMIX_process_from_PID(aid_t PID)
632 auto actor = simix_global->process_list.find(PID);
633 return actor == simix_global->process_list.end() ? nullptr : actor->second;
636 void SIMIX_process_on_exit(smx_actor_t actor, int_f_pvoid_pvoid_t fun, void* data)
638 SIMIX_process_on_exit(actor, [fun, data](bool failed) {
639 intptr_t status = failed ? SMX_EXIT_FAILURE : SMX_EXIT_SUCCESS;
640 fun(reinterpret_cast<void*>(status), data);
644 void SIMIX_process_on_exit(smx_actor_t actor, const std::function<void(int, void*)>& fun, void* data)
646 SIMIX_process_on_exit(actor, [fun, data](bool failed) { fun(failed ? SMX_EXIT_FAILURE : SMX_EXIT_SUCCESS, data); });
649 void SIMIX_process_on_exit(smx_actor_t actor, const std::function<void(bool /*failed*/)>& fun)
651 xbt_assert(actor, "current process not found: are you in maestro context ?");
652 actor->on_exit->emplace_back(fun);
655 /** @brief Restart a process, starting it again from the beginning. */
657 * @ingroup simix_process_management
658 * @brief Creates and runs a new SIMIX process.
660 * The structure and the corresponding thread are created and put in the list of ready processes.
662 * @param name a name for the process. It is for user-level information and can be nullptr.
663 * @param code the main function of the process
664 * @param data a pointer to any data one may want to attach to the new object. It is for user-level information and can
666 * It can be retrieved with the method ActorImpl::getUserData().
667 * @param host where the new agent is executed.
668 * @param properties the properties of the process
670 smx_actor_t simcall_process_create(const std::string& name, const simgrid::simix::ActorCode& code, void* data,
671 sg_host_t host, std::unordered_map<std::string, std::string>* properties)
673 smx_actor_t self = SIMIX_process_self();
674 return simgrid::simix::simcall([&name, &code, data, host, properties, self] {
675 return simgrid::kernel::actor::ActorImpl::create(name, code, data, host, properties, self).get();
679 void simcall_process_set_data(smx_actor_t process, void* data)
681 simgrid::simix::simcall([process, data] { process->set_user_data(data); });