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 while (not on_exit.empty()) {
145 auto exit_fun = on_exit.back();
150 /* cancel non-blocking activities */
151 for (auto activity : comms)
152 boost::static_pointer_cast<activity::CommImpl>(activity)->cancel();
155 XBT_DEBUG("%s@%s(%ld) should not run anymore", get_cname(), get_host()->get_cname(), get_pid());
157 if (this == simix_global->maestro_process) /* Do not cleanup maestro */
160 XBT_DEBUG("Cleanup actor %s (%p), waiting synchro %p", get_cname(), this, waiting_synchro.get());
162 /* Unregister from the kill timer if any */
163 if (kill_timer != nullptr) {
164 kill_timer->remove();
165 kill_timer = nullptr;
168 simix_global->mutex.lock();
170 simix_global->process_list.erase(pid_);
171 if (host_ && host_process_list_hook.is_linked())
172 simgrid::xbt::intrusive_erase(host_->pimpl_->process_list_, *this);
173 if (not smx_destroy_list_hook.is_linked()) {
175 xbt_dynar_push_as(simix_global->dead_actors_vector, ActorImpl*, this);
177 simix_global->actors_to_destroy.push_back(*this);
180 simix_global->mutex.unlock();
182 context_->iwannadie = false; // don't let the simcall's yield() do a Context::stop(), to avoid infinite loops
183 simgrid::simix::simcall([this] { simgrid::s4u::Actor::on_destruction(*ciface()); });
184 context_->iwannadie = true;
187 void ActorImpl::exit()
189 context_->iwannadie = true;
192 exception_ = nullptr;
194 // Forcefully kill the actor if its host is turned off. Not a HostFailureException because you should not survive that
195 if (not host_->is_on())
196 this->throw_exception(std::make_exception_ptr(ForcefulKillException("host failed")));
198 /* destroy the blocking synchro if any */
199 if (waiting_synchro != nullptr) {
200 waiting_synchro->cancel();
201 waiting_synchro->state_ = SIMIX_FAILED;
203 activity::ExecImplPtr exec = boost::dynamic_pointer_cast<activity::ExecImpl>(waiting_synchro);
204 activity::CommImplPtr comm = boost::dynamic_pointer_cast<activity::CommImpl>(waiting_synchro);
206 if (exec != nullptr) {
207 exec->clean_action();
208 } else if (comm != nullptr) {
209 comms.remove(waiting_synchro);
210 // Remove first occurrence of &actor->simcall:
211 auto i = boost::range::find(waiting_synchro->simcalls_, &simcall);
212 if (i != waiting_synchro->simcalls_.end())
213 waiting_synchro->simcalls_.remove(&simcall);
215 activity::ActivityImplPtr(waiting_synchro)->finish();
218 waiting_synchro = nullptr;
222 void ActorImpl::kill(ActorImpl* actor)
224 if (actor->finished_) {
225 XBT_DEBUG("Ignoring request to kill actor %s@%s that is already dead", actor->get_cname(),
226 actor->host_->get_cname());
230 XBT_DEBUG("Actor '%s'@%s is killing actor '%s'@%s", get_cname(), host_ ? host_->get_cname() : "", actor->get_cname(),
231 actor->host_ ? actor->host_->get_cname() : "");
235 if (std::find(begin(simix_global->actors_to_run), end(simix_global->actors_to_run), actor) ==
236 end(simix_global->actors_to_run) &&
238 XBT_DEBUG("Inserting %s in the to_run list", actor->get_cname());
239 simix_global->actors_to_run.push_back(actor);
243 void ActorImpl::kill_all()
245 for (auto const& kv : simix_global->process_list)
246 if (kv.second != this)
247 this->kill(kv.second);
250 void ActorImpl::set_kill_time(double kill_time)
252 if (kill_time <= SIMIX_get_clock())
254 XBT_DEBUG("Set kill time %f for actor %s@%s", kill_time, get_cname(), host_->get_cname());
255 kill_timer = simix::Timer::set(kill_time, [this] {
257 kill_timer = nullptr;
261 double ActorImpl::get_kill_time()
263 return kill_timer ? kill_timer->get_date() : 0;
266 void ActorImpl::yield()
268 XBT_DEBUG("Yield actor '%s'", get_cname());
270 /* Go into sleep and return control to maestro */
273 /* Ok, maestro returned control to us */
274 XBT_DEBUG("Control returned to me: '%s'", get_cname());
276 if (context_->iwannadie) {
277 XBT_DEBUG("Actor %s@%s is dead", get_cname(), host_->get_cname());
278 // throw simgrid::kernel::context::ForcefulKillException(); Does not seem to properly kill the actor
284 XBT_DEBUG("Hey! I'm suspended.");
286 xbt_assert(exception_ == nullptr, "Gasp! This exception may be lost by subsequent calls.");
291 if (exception_ != nullptr) {
292 XBT_DEBUG("Wait, maestro left me an exception");
293 std::exception_ptr exception = std::move(exception_);
294 exception_ = nullptr;
295 std::rethrow_exception(std::move(exception));
298 if (SMPI_switch_data_segment && not finished_) {
299 SMPI_switch_data_segment(iface());
303 /** This actor will be terminated automatically when the last non-daemon actor finishes */
304 void ActorImpl::daemonize()
308 simix_global->daemons.push_back(this);
309 SIMIX_process_on_exit(this, [this](bool) {
310 auto& vect = simix_global->daemons;
311 auto it = std::find(vect.begin(), vect.end(), this);
312 xbt_assert(it != vect.end(), "The dying daemon is not a daemon after all. Please report that bug.");
314 /* Don't move the whole content since we don't really care about the order */
315 std::swap(*it, vect.back());
321 s4u::Actor* ActorImpl::restart()
323 xbt_assert(this != simix_global->maestro_process, "Restarting maestro is not supported");
325 XBT_DEBUG("Restarting actor %s on %s", get_cname(), host_->get_cname());
327 // retrieve the arguments of the old actor
328 ProcessArg arg = ProcessArg(host_, this);
330 // kill the old actor
331 context::Context::self()->get_actor()->kill(this);
333 // start the new actor
335 ActorImpl::create(arg.name, std::move(arg.code), arg.data, arg.host, arg.properties.get(), nullptr);
336 actor->on_exit = std::move(arg.on_exit);
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);
423 waiting_synchro = nullptr;
427 void ActorImpl::set_host(s4u::Host* dest)
429 xbt::intrusive_erase(host_->pimpl_->process_list_, *this);
431 dest->pimpl_->process_list_.push_back(*this);
434 ActorImplPtr ActorImpl::init(const std::string& name, s4u::Host* host)
436 ActorImpl* actor = new ActorImpl(xbt::string(name), host);
437 actor->set_ppid(this->pid_);
439 intrusive_ptr_add_ref(actor);
440 /* The on_creation() signal must be delayed until there, where the pid and everything is set */
441 s4u::Actor::on_creation(*actor->ciface());
443 return ActorImplPtr(actor);
446 ActorImpl* ActorImpl::start(const simix::ActorCode& code)
448 xbt_assert(code && host_ != nullptr, "Invalid parameters");
450 if (not host_->is_on()) {
451 XBT_WARN("Cannot launch actor '%s' on failed host '%s'", name_.c_str(), host_->get_cname());
452 intrusive_ptr_release(this);
453 std::rethrow_exception(
454 std::make_exception_ptr(simgrid::HostFailureException(XBT_THROW_POINT, "Cannot start actor on failed host.")));
458 XBT_VERB("Create context %s", get_cname());
459 context_.reset(simix_global->context_factory->create_context(simix::ActorCode(code), this));
461 XBT_DEBUG("Start context '%s'", get_cname());
463 /* Add the actor to its host's actor list */
464 host_->pimpl_->process_list_.push_back(*this);
465 simix_global->process_list[pid_] = this;
467 /* Now insert it in the global actor list and in the actor to run list */
468 XBT_DEBUG("Inserting [%p] %s(%s) in the to_run list", this, get_cname(), host_->get_cname());
469 simix_global->actors_to_run.push_back(this);
474 ActorImplPtr ActorImpl::create(const std::string& name, const simix::ActorCode& code, void* data, s4u::Host* host,
475 const std::unordered_map<std::string, std::string>* properties, ActorImpl* parent_actor)
477 XBT_DEBUG("Start actor %s@'%s'", name.c_str(), host->get_cname());
480 if (parent_actor != nullptr)
481 actor = parent_actor->init(xbt::string(name), host);
483 actor = SIMIX_process_self()->init(xbt::string(name), host);
486 actor->set_user_data(data);
489 if (properties != nullptr)
490 actor->set_properties(*properties);
497 void create_maestro(const std::function<void()>& code)
499 /* Create maestro actor and initialize it */
500 ActorImpl* maestro = new ActorImpl(xbt::string(""), /*host*/ nullptr);
503 maestro->context_.reset(simix_global->context_factory->create_context(simix::ActorCode(), maestro));
505 maestro->context_.reset(simix_global->context_factory->create_maestro(simix::ActorCode(code), maestro));
508 maestro->simcall.issuer = maestro;
509 simix_global->maestro_process = maestro;
513 } // namespace kernel
514 } // namespace simgrid
516 void SIMIX_process_detach()
518 simgrid::kernel::actor::ActorImpl::detach();
521 smx_actor_t SIMIX_process_attach(const char* name, void* data, const char* hostname,
522 std::unordered_map<std::string, std::string>* properties,
523 smx_actor_t /*parent_process*/)
525 return simgrid::kernel::actor::ActorImpl::attach(name, data, sg_host_by_name(hostname), properties).get();
528 /** @deprecated When this function gets removed, also remove the xbt_ex class, that is only there to help users to
530 void SIMIX_process_throw(smx_actor_t actor, xbt_errcat_t cat, int value, const char* msg)
532 xbt_ex e(XBT_THROW_POINT, msg);
535 actor->throw_exception(std::make_exception_ptr(e));
538 void simcall_HANDLER_process_suspend(smx_simcall_t simcall, smx_actor_t actor)
540 smx_activity_t sync_suspend = actor->suspend(simcall->issuer);
542 if (actor != simcall->issuer) {
543 SIMIX_simcall_answer(simcall);
545 sync_suspend->simcalls_.push_back(simcall);
546 actor->waiting_synchro = sync_suspend;
547 actor->waiting_synchro->suspend();
549 /* If we are suspending ourselves, then just do not finish the simcall now */
552 int SIMIX_process_get_maxpid()
554 return simix_process_maxpid;
557 int SIMIX_process_count()
559 return simix_global->process_list.size();
562 void* SIMIX_process_self_get_data() // deprecated
564 smx_actor_t self = SIMIX_process_self();
566 if (self == nullptr) {
569 return self->get_user_data();
572 void SIMIX_process_self_set_data(void* data) // deprecated
574 SIMIX_process_self()->set_user_data(data);
577 /* needs to be public and without simcall because it is called
578 by exceptions and logging events */
579 const char* SIMIX_process_self_get_name()
582 smx_actor_t process = SIMIX_process_self();
583 if (process == nullptr || process == simix_global->maestro_process)
586 return process->get_cname();
589 void simcall_HANDLER_process_join(smx_simcall_t simcall, smx_actor_t process, double timeout)
591 if (process->finished_) {
592 // The joined process is already finished, just wake up the issuer process right away
593 simcall_process_sleep__set__result(simcall, SIMIX_DONE);
594 SIMIX_simcall_answer(simcall);
597 smx_activity_t sync = simcall->issuer->join(process, timeout);
598 sync->simcalls_.push_back(simcall);
599 simcall->issuer->waiting_synchro = sync;
602 void simcall_HANDLER_process_sleep(smx_simcall_t simcall, double duration)
604 if (MC_is_active() || MC_record_replay_is_active()) {
605 MC_process_clock_add(simcall->issuer, duration);
606 simcall_process_sleep__set__result(simcall, SIMIX_DONE);
607 SIMIX_simcall_answer(simcall);
610 smx_activity_t sync = simcall->issuer->sleep(duration);
611 sync->simcalls_.push_back(simcall);
612 simcall->issuer->waiting_synchro = sync;
616 * @brief Calling this function makes the process to yield.
618 * Only the current process can call this function, giving back the control to maestro.
620 * @param self the current process
623 /** @brief Returns the list of processes to run.
626 const std::vector<smx_actor_t>& simgrid::simix::process_get_runnable()
628 return simix_global->actors_to_run;
631 /** @brief Returns the process from PID. */
632 smx_actor_t SIMIX_process_from_PID(aid_t PID)
634 auto actor = simix_global->process_list.find(PID);
635 return actor == simix_global->process_list.end() ? nullptr : actor->second;
638 void SIMIX_process_on_exit(smx_actor_t actor, int_f_pvoid_pvoid_t fun, void* data)
640 SIMIX_process_on_exit(actor, [fun, data](bool failed) {
641 intptr_t status = failed ? SMX_EXIT_FAILURE : SMX_EXIT_SUCCESS;
642 fun(reinterpret_cast<void*>(status), data);
646 void SIMIX_process_on_exit(smx_actor_t actor, const std::function<void(int, void*)>& fun, void* data)
648 SIMIX_process_on_exit(actor, [fun, data](bool failed) { fun(failed ? SMX_EXIT_FAILURE : SMX_EXIT_SUCCESS, data); });
651 void SIMIX_process_on_exit(smx_actor_t actor, const std::function<void(bool /*failed*/)>& fun)
653 xbt_assert(actor, "current process not found: are you in maestro context ?");
654 actor->on_exit.emplace_back(fun);
657 /** @brief Restart a process, starting it again from the beginning. */
659 * @ingroup simix_process_management
660 * @brief Creates and runs a new SIMIX process.
662 * The structure and the corresponding thread are created and put in the list of ready processes.
664 * @param name a name for the process. It is for user-level information and can be nullptr.
665 * @param code the main function of the process
666 * @param data a pointer to any data one may want to attach to the new object. It is for user-level information and can
668 * It can be retrieved with the method ActorImpl::getUserData().
669 * @param host where the new agent is executed.
670 * @param properties the properties of the process
672 smx_actor_t simcall_process_create(const std::string& name, const simgrid::simix::ActorCode& code, void* data,
673 sg_host_t host, std::unordered_map<std::string, std::string>* properties)
675 smx_actor_t self = SIMIX_process_self();
676 return simgrid::simix::simcall([&name, &code, data, host, properties, self] {
677 return simgrid::kernel::actor::ActorImpl::create(name, code, data, host, properties, self).get();
681 void simcall_process_set_data(smx_actor_t process, void* data)
683 simgrid::simix::simcall([process, data] { process->set_user_data(data); });