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()
64 if (simix_global != nullptr && this != simix_global->maestro_process) {
65 if (context_.get() != nullptr) /* the actor was not start()ed yet. This happens if its host was initially off */
66 context_->iwannadie = false; // don't let the simcall's yield() do a Context::stop(), to avoid infinite loops
67 simgrid::simix::simcall([this] { simgrid::s4u::Actor::on_destruction(*ciface()); });
68 if (context_.get() != nullptr)
69 context_->iwannadie = true;
73 /* Become an actor in the simulation
75 * Currently this can only be called by the main thread (once) and only work with some thread factories
76 * (currently ThreadContextFactory).
78 * In the future, it might be extended in order to attach other threads created by a third party library.
81 ActorImplPtr ActorImpl::attach(const std::string& name, void* data, s4u::Host* host,
82 const std::unordered_map<std::string, std::string>* properties)
84 // This is mostly a copy/paste from create(), it'd be nice to share some code between those two functions.
86 XBT_DEBUG("Attach process %s on host '%s'", name.c_str(), host->get_cname());
88 if (not host->is_on()) {
89 XBT_WARN("Cannot launch process '%s' on failed host '%s'", name.c_str(), host->get_cname());
90 throw simgrid::HostFailureException(XBT_THROW_POINT, "Cannot attach actor on failed host.");
93 ActorImpl* actor = new ActorImpl(xbt::string(name), host);
95 actor->set_user_data(data);
96 actor->code_ = nullptr;
98 XBT_VERB("Create context %s", actor->get_cname());
99 xbt_assert(simix_global != nullptr, "simix is not initialized, please call MSG_init first");
100 actor->context_.reset(simix_global->context_factory->attach(actor));
103 if (properties != nullptr)
104 actor->set_properties(*properties);
106 /* Add the process to it's host process list */
107 host->pimpl_->process_list_.push_back(*actor);
109 /* Now insert it in the global process list and in the process to run list */
110 simix_global->process_list[actor->get_pid()] = actor;
111 XBT_DEBUG("Inserting [%p] %s(%s) in the to_run list", actor, actor->get_cname(), host->get_cname());
112 simix_global->actors_to_run.push_back(actor);
113 intrusive_ptr_add_ref(actor);
115 auto* context = dynamic_cast<simgrid::kernel::context::AttachContext*>(actor->context_.get());
116 xbt_assert(nullptr != context, "Not a suitable context");
117 context->attach_start();
119 /* The on_creation() signal must be delayed until there, where the pid and everything is set */
120 simgrid::s4u::Actor::on_creation(*actor->ciface());
122 return ActorImplPtr(actor);
124 /** @brief Detach an actor attached with `attach()`
126 * This is called when the current actor has finished its job.
127 * Used in the main thread, it waits for the simulation to finish before returning. When it returns, the other
128 * simulated actors and the maestro are destroyed.
130 void ActorImpl::detach()
132 auto* context = dynamic_cast<context::AttachContext*>(context::Context::self());
133 if (context == nullptr)
134 xbt_die("Not a suitable context");
136 context->get_actor()->cleanup();
137 context->attach_stop();
140 void ActorImpl::cleanup()
144 if (has_to_auto_restart() && not get_host()->is_on()) {
145 XBT_DEBUG("Insert host %s to watched_hosts because it's off and %s needs to restart", get_host()->get_cname(),
147 watched_hosts.insert(get_host()->get_name());
151 // Execute the termination callbacks
152 bool failed = context_->iwannadie;
153 for (auto exit_fun = on_exit->crbegin(); exit_fun != on_exit->crend(); ++exit_fun)
159 /* cancel non-blocking activities */
160 for (auto activity : comms)
161 boost::static_pointer_cast<activity::CommImpl>(activity)->cancel();
164 XBT_DEBUG("%s@%s(%ld) should not run anymore", get_cname(), get_host()->get_cname(), get_pid());
166 if (this == simix_global->maestro_process) /* Do not cleanup maestro */
169 XBT_DEBUG("Cleanup actor %s (%p), waiting synchro %p", get_cname(), this, waiting_synchro.get());
171 /* Unregister from the kill timer if any */
172 if (kill_timer != nullptr) {
173 kill_timer->remove();
174 kill_timer = nullptr;
177 simix_global->mutex.lock();
179 simix_global->process_list.erase(pid_);
180 if (host_ && host_process_list_hook.is_linked())
181 simgrid::xbt::intrusive_erase(host_->pimpl_->process_list_, *this);
182 if (not smx_destroy_list_hook.is_linked()) {
184 xbt_dynar_push_as(simix_global->dead_actors_vector, ActorImpl*, this);
186 simix_global->actors_to_destroy.push_back(*this);
189 simix_global->mutex.unlock();
191 context_->iwannadie = false; // don't let the simcall's yield() do a Context::stop(), to avoid infinite loops
192 simgrid::simix::simcall([this] { simgrid::s4u::Actor::on_termination(*ciface()); });
193 context_->iwannadie = true;
196 void ActorImpl::exit()
198 context_->iwannadie = true;
200 exception_ = nullptr;
202 // Forcefully kill the actor if its host is turned off. Not a HostFailureException because you should not survive that
203 if (not host_->is_on())
204 this->throw_exception(std::make_exception_ptr(ForcefulKillException("host failed")));
206 /* destroy the blocking synchro if any */
207 if (waiting_synchro != nullptr) {
208 waiting_synchro->cancel();
209 waiting_synchro->state_ = SIMIX_FAILED;
211 activity::ExecImplPtr exec = boost::dynamic_pointer_cast<activity::ExecImpl>(waiting_synchro);
212 activity::CommImplPtr comm = boost::dynamic_pointer_cast<activity::CommImpl>(waiting_synchro);
214 if (exec != nullptr) {
215 exec->clean_action();
216 } else if (comm != nullptr) {
217 comms.remove(waiting_synchro);
218 // Remove first occurrence of &actor->simcall:
219 auto i = boost::range::find(waiting_synchro->simcalls_, &simcall);
220 if (i != waiting_synchro->simcalls_.end())
221 waiting_synchro->simcalls_.remove(&simcall);
223 activity::ActivityImplPtr(waiting_synchro)->finish();
226 waiting_synchro = nullptr;
230 void ActorImpl::kill(ActorImpl* actor)
232 if (actor->finished_) {
233 XBT_DEBUG("Ignoring request to kill actor %s@%s that is already dead", actor->get_cname(),
234 actor->host_->get_cname());
238 XBT_DEBUG("Actor '%s'@%s is killing actor '%s'@%s", get_cname(), host_ ? host_->get_cname() : "", actor->get_cname(),
239 actor->host_ ? actor->host_->get_cname() : "");
243 if (std::find(begin(simix_global->actors_to_run), end(simix_global->actors_to_run), actor) ==
244 end(simix_global->actors_to_run) &&
246 XBT_DEBUG("Inserting %s in the to_run list", actor->get_cname());
247 simix_global->actors_to_run.push_back(actor);
251 void ActorImpl::kill_all()
253 for (auto const& kv : simix_global->process_list)
254 if (kv.second != this)
255 this->kill(kv.second);
258 void ActorImpl::set_kill_time(double kill_time)
260 if (kill_time <= SIMIX_get_clock())
262 XBT_DEBUG("Set kill time %f for actor %s@%s", kill_time, get_cname(), host_->get_cname());
263 kill_timer = simix::Timer::set(kill_time, [this] {
265 kill_timer = nullptr;
269 double ActorImpl::get_kill_time()
271 return kill_timer ? kill_timer->get_date() : 0;
274 void ActorImpl::yield()
276 XBT_DEBUG("Yield actor '%s'", get_cname());
278 /* Go into sleep and return control to maestro */
281 /* Ok, maestro returned control to us */
282 XBT_DEBUG("Control returned to me: '%s'", get_cname());
284 if (context_->iwannadie) {
285 XBT_DEBUG("Actor %s@%s is dead", get_cname(), host_->get_cname());
286 // throw simgrid::kernel::context::ForcefulKillException(); Does not seem to properly kill the actor
292 XBT_DEBUG("Hey! I'm suspended.");
294 xbt_assert(exception_ == nullptr, "Gasp! This exception may be lost by subsequent calls.");
299 if (exception_ != nullptr) {
300 XBT_DEBUG("Wait, maestro left me an exception");
301 std::exception_ptr exception = std::move(exception_);
302 exception_ = nullptr;
303 std::rethrow_exception(std::move(exception));
306 if (SMPI_switch_data_segment && not finished_) {
307 SMPI_switch_data_segment(iface());
311 /** This actor will be terminated automatically when the last non-daemon actor finishes */
312 void ActorImpl::daemonize()
316 simix_global->daemons.push_back(this);
320 void ActorImpl::undaemonize()
323 auto& vect = simix_global->daemons;
324 auto it = std::find(vect.begin(), vect.end(), this);
325 xbt_assert(it != vect.end(), "The dying daemon is not a daemon after all. Please report that bug.");
326 /* Don't move the whole content since we don't really care about the order */
328 std::swap(*it, vect.back());
334 s4u::Actor* ActorImpl::restart()
336 xbt_assert(this != simix_global->maestro_process, "Restarting maestro is not supported");
338 XBT_DEBUG("Restarting actor %s on %s", get_cname(), host_->get_cname());
340 // retrieve the arguments of the old actor
341 ProcessArg arg = ProcessArg(host_, this);
343 // kill the old actor
344 context::Context::self()->get_actor()->kill(this);
346 // start the new actor
348 ActorImpl::create(arg.name, std::move(arg.code), arg.data, arg.host, arg.properties.get(), nullptr);
349 *actor->on_exit = std::move(*arg.on_exit);
350 actor->set_kill_time(arg.kill_time);
351 actor->set_auto_restart(arg.auto_restart);
353 return actor->ciface();
356 activity::ActivityImplPtr ActorImpl::suspend(ActorImpl* issuer)
359 XBT_DEBUG("Actor '%s' is already suspended", get_cname());
365 /* If we are suspending another actor that is waiting on a sync, suspend its synchronization. */
366 if (this != issuer) {
368 waiting_synchro->suspend();
369 /* If the other actor is not waiting, its suspension is delayed to when the actor is rescheduled. */
373 activity::ExecImpl* exec = new activity::ExecImpl();
374 (*exec).set_name("suspend").set_host(host_).set_flops_amount(0.0).start();
375 return activity::ExecImplPtr(exec);
379 void ActorImpl::resume()
381 XBT_IN("actor = %p", this);
383 if (context_->iwannadie) {
384 XBT_VERB("Ignoring request to suspend an actor that is currently dying.");
392 /* resume the synchronization that was blocking the resumed actor. */
394 waiting_synchro->resume();
399 activity::ActivityImplPtr ActorImpl::join(ActorImpl* actor, double timeout)
401 activity::ActivityImplPtr sleep = this->sleep(timeout);
402 SIMIX_process_on_exit(actor, [sleep](bool) {
403 if (sleep->surf_action_)
404 sleep->surf_action_->finish(resource::Action::State::FINISHED);
409 activity::ActivityImplPtr ActorImpl::sleep(double duration)
411 if (not host_->is_on())
412 throw_exception(std::make_exception_ptr(simgrid::HostFailureException(
413 XBT_THROW_POINT, std::string("Host ") + host_->get_cname() + " failed, you cannot sleep there.")));
415 activity::SleepImpl* sleep = new activity::SleepImpl();
416 (*sleep).set_name("sleep").set_host(host_).set_duration(duration).start();
417 return activity::SleepImplPtr(sleep);
420 void ActorImpl::throw_exception(std::exception_ptr e)
427 /* cancel the blocking synchro if any */
428 if (waiting_synchro) {
429 waiting_synchro->cancel();
431 activity::CommImplPtr comm = boost::dynamic_pointer_cast<activity::CommImpl>(waiting_synchro);
436 waiting_synchro = nullptr;
440 void ActorImpl::set_host(s4u::Host* dest)
442 xbt::intrusive_erase(host_->pimpl_->process_list_, *this);
444 dest->pimpl_->process_list_.push_back(*this);
447 ActorImplPtr ActorImpl::init(const std::string& name, s4u::Host* host)
449 ActorImpl* actor = new ActorImpl(xbt::string(name), host);
450 actor->set_ppid(this->pid_);
452 intrusive_ptr_add_ref(actor);
453 /* The on_creation() signal must be delayed until there, where the pid and everything is set */
454 s4u::Actor::on_creation(*actor->ciface());
456 return ActorImplPtr(actor);
459 ActorImpl* ActorImpl::start(const simix::ActorCode& code)
461 xbt_assert(code && host_ != nullptr, "Invalid parameters");
463 if (not host_->is_on()) {
464 XBT_WARN("Cannot launch actor '%s' on failed host '%s'", name_.c_str(), host_->get_cname());
465 intrusive_ptr_release(this);
466 throw simgrid::HostFailureException(XBT_THROW_POINT, "Cannot start actor on failed host.");
470 XBT_VERB("Create context %s", get_cname());
471 context_.reset(simix_global->context_factory->create_context(simix::ActorCode(code), this));
473 XBT_DEBUG("Start context '%s'", get_cname());
475 /* Add the actor to its host's actor list */
476 host_->pimpl_->process_list_.push_back(*this);
477 simix_global->process_list[pid_] = this;
479 /* Now insert it in the global actor list and in the actor to run list */
480 XBT_DEBUG("Inserting [%p] %s(%s) in the to_run list", this, get_cname(), host_->get_cname());
481 simix_global->actors_to_run.push_back(this);
486 ActorImplPtr ActorImpl::create(const std::string& name, const simix::ActorCode& code, void* data, s4u::Host* host,
487 const std::unordered_map<std::string, std::string>* properties, ActorImpl* parent_actor)
489 XBT_DEBUG("Start actor %s@'%s'", name.c_str(), host->get_cname());
492 if (parent_actor != nullptr)
493 actor = parent_actor->init(xbt::string(name), host);
495 actor = SIMIX_process_self()->init(xbt::string(name), host);
498 actor->set_user_data(data);
501 if (properties != nullptr)
502 actor->set_properties(*properties);
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 void simcall_HANDLER_process_suspend(smx_simcall_t simcall, smx_actor_t actor)
542 smx_activity_t sync_suspend = actor->suspend(simcall->issuer);
544 if (actor != simcall->issuer) {
545 SIMIX_simcall_answer(simcall);
547 sync_suspend->simcalls_.push_back(simcall);
548 actor->waiting_synchro = sync_suspend;
549 actor->waiting_synchro->suspend();
551 /* If we are suspending ourselves, then just do not finish the simcall now */
554 int SIMIX_process_get_maxpid()
556 return simix_process_maxpid;
559 int SIMIX_process_count()
561 return simix_global->process_list.size();
564 void* SIMIX_process_self_get_data() // deprecated
566 smx_actor_t self = SIMIX_process_self();
568 if (self == nullptr) {
571 return self->get_user_data();
574 void SIMIX_process_self_set_data(void* data) // deprecated
576 SIMIX_process_self()->set_user_data(data);
579 /* needs to be public and without simcall because it is called
580 by exceptions and logging events */
581 const char* SIMIX_process_self_get_name()
584 smx_actor_t process = SIMIX_process_self();
585 if (process == nullptr || process == simix_global->maestro_process)
588 return process->get_cname();
591 void simcall_HANDLER_process_join(smx_simcall_t simcall, smx_actor_t process, double timeout)
593 if (process->finished_) {
594 // The joined process is already finished, just wake up the issuer process right away
595 simcall_process_sleep__set__result(simcall, SIMIX_DONE);
596 SIMIX_simcall_answer(simcall);
599 smx_activity_t sync = simcall->issuer->join(process, timeout);
600 sync->simcalls_.push_back(simcall);
601 simcall->issuer->waiting_synchro = sync;
604 void simcall_HANDLER_process_sleep(smx_simcall_t simcall, double duration)
606 if (MC_is_active() || MC_record_replay_is_active()) {
607 MC_process_clock_add(simcall->issuer, duration);
608 simcall_process_sleep__set__result(simcall, SIMIX_DONE);
609 SIMIX_simcall_answer(simcall);
612 smx_activity_t sync = simcall->issuer->sleep(duration);
613 sync->simcalls_.push_back(simcall);
614 simcall->issuer->waiting_synchro = sync;
618 * @brief Calling this function makes the process to yield.
620 * Only the current process can call this function, giving back the control to maestro.
622 * @param self the current process
625 /** @brief Returns the list of processes to run.
628 const std::vector<smx_actor_t>& simgrid::simix::process_get_runnable()
630 return simix_global->actors_to_run;
633 /** @brief Returns the process from PID. */
634 smx_actor_t SIMIX_process_from_PID(aid_t PID)
636 auto item = simix_global->process_list.find(PID);
637 if (item == simix_global->process_list.end()) {
638 for (auto& a : simix_global->actors_to_destroy)
639 if (a.get_pid() == PID)
641 return nullptr; // Not found, even in the trash
646 void SIMIX_process_on_exit(smx_actor_t actor, int_f_pvoid_pvoid_t fun, void* data)
648 SIMIX_process_on_exit(actor, [fun, data](bool failed) {
649 intptr_t status = failed ? SMX_EXIT_FAILURE : SMX_EXIT_SUCCESS;
650 fun(reinterpret_cast<void*>(status), data);
654 void SIMIX_process_on_exit(smx_actor_t actor, const std::function<void(int, void*)>& fun, void* data)
656 SIMIX_process_on_exit(actor, [fun, data](bool failed) { fun(failed ? SMX_EXIT_FAILURE : SMX_EXIT_SUCCESS, data); });
659 void SIMIX_process_on_exit(smx_actor_t actor, const std::function<void(bool /*failed*/)>& fun)
661 xbt_assert(actor, "current process not found: are you in maestro context ?");
662 actor->on_exit->emplace_back(fun);
665 /** @brief Restart a process, starting it again from the beginning. */
667 * @ingroup simix_process_management
668 * @brief Creates and runs a new SIMIX process.
670 * The structure and the corresponding thread are created and put in the list of ready processes.
672 * @param name a name for the process. It is for user-level information and can be nullptr.
673 * @param code the main function of the process
674 * @param data a pointer to any data one may want to attach to the new object. It is for user-level information and can
676 * It can be retrieved with the method ActorImpl::getUserData().
677 * @param host where the new agent is executed.
678 * @param properties the properties of the process
680 smx_actor_t simcall_process_create(const std::string& name, const simgrid::simix::ActorCode& code, void* data,
681 sg_host_t host, std::unordered_map<std::string, std::string>* properties)
683 smx_actor_t self = SIMIX_process_self();
684 return simgrid::simix::simcall([&name, &code, data, host, properties, self] {
685 return simgrid::kernel::actor::ActorImpl::create(name, code, data, host, properties, self).get();
689 void simcall_process_set_data(smx_actor_t process, void* data)
691 simgrid::simix::simcall([process, data] { process->set_user_data(data); });