1 /* Copyright (c) 2007-2021. 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/EngineImpl.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/AppSide.hpp"
18 #include "src/simix/smx_private.hpp"
19 #include "src/surf/HostImpl.hpp"
20 #include "src/surf/cpu_interface.hpp"
22 #include <boost/core/demangle.hpp>
26 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_process, simix, "Logging specific to SIMIX (process)");
29 * @brief Returns the current agent.
31 * This functions returns the currently running SIMIX process.
33 * @return The SIMIX process
35 smx_actor_t SIMIX_process_self()
37 return simgrid::kernel::actor::ActorImpl::self();
44 static unsigned long maxpid = 0;
45 unsigned long get_maxpid()
49 unsigned long* get_maxpid_addr()
53 ActorImpl* ActorImpl::by_pid(aid_t pid)
55 auto item = simix_global->process_list.find(pid);
56 if (item != simix_global->process_list.end())
60 for (auto& a : simix_global->actors_to_destroy)
61 if (a.get_pid() == pid)
63 return nullptr; // Not found, even in the trash
66 ActorImpl* ActorImpl::self()
68 const context::Context* self_context = context::Context::self();
70 return (self_context != nullptr) ? self_context->get_actor() : nullptr;
73 ActorImpl::ActorImpl(xbt::string name, s4u::Host* host) : host_(host), name_(std::move(name)), piface_(this)
76 simcall_.issuer_ = this;
77 stacksize_ = smx_context_stack_size;
80 ActorImpl::~ActorImpl()
82 if (simix_global != nullptr && this != simix_global->maestro_)
83 s4u::Actor::on_destruction(*get_ciface());
86 /* Become an actor in the simulation
88 * Currently this can only be called by the main thread (once) and only work with some thread factories
89 * (currently ThreadContextFactory).
91 * In the future, it might be extended in order to attach other threads created by a third party library.
94 ActorImplPtr ActorImpl::attach(const std::string& name, void* data, s4u::Host* host)
96 // This is mostly a copy/paste from create(), it'd be nice to share some code between those two functions.
98 XBT_DEBUG("Attach actor %s on host '%s'", name.c_str(), host->get_cname());
100 if (not host->is_on()) {
101 XBT_WARN("Cannot attach actor '%s' on failed host '%s'", name.c_str(), host->get_cname());
102 throw HostFailureException(XBT_THROW_POINT, "Cannot attach actor on failed host.");
105 auto* actor = new ActorImpl(xbt::string(name), host);
107 actor->set_user_data(data);
108 actor->code_ = nullptr;
110 XBT_VERB("Create context %s", actor->get_cname());
111 xbt_assert(simix_global != nullptr, "simix is not initialized, please call MSG_init first");
112 actor->context_.reset(simix_global->context_factory->attach(actor));
114 /* Add the actor to it's host actor list */
115 host->get_impl()->add_actor(actor);
117 /* Now insert it in the global actor list and in the actors to run list */
118 simix_global->process_list[actor->get_pid()] = actor;
119 EngineImpl::get_instance()->add_actor_to_run_list_no_check(actor);
120 intrusive_ptr_add_ref(actor);
122 auto* context = dynamic_cast<context::AttachContext*>(actor->context_.get());
123 xbt_assert(nullptr != context, "Not a suitable context");
124 context->attach_start();
126 /* The on_creation() signal must be delayed until there, where the pid and everything is set */
127 s4u::Actor::on_creation(*actor->get_ciface());
129 return ActorImplPtr(actor);
131 /** @brief Detach an actor attached with `attach()`
133 * This is called when the current actor has finished its job.
134 * Used in the main thread, it waits for the simulation to finish before returning. When it returns, the other
135 * simulated actors and the maestro are destroyed.
137 void ActorImpl::detach()
139 auto* context = dynamic_cast<context::AttachContext*>(context::Context::self());
140 xbt_assert(context != nullptr, "Not a suitable context");
142 context->get_actor()->cleanup();
143 context->attach_stop();
146 void ActorImpl::cleanup_from_simix()
148 const std::lock_guard<std::mutex> lock(simix_global->mutex);
149 simix_global->process_list.erase(pid_);
150 if (host_ && host_actor_list_hook.is_linked())
151 host_->get_impl()->remove_actor(this);
152 if (not smx_destroy_list_hook.is_linked()) {
154 xbt_dynar_push_as(simix_global->dead_actors_vector, ActorImpl*, this);
156 simix_global->actors_to_destroy.push_back(*this);
160 void ActorImpl::cleanup()
164 if (has_to_auto_restart() && not get_host()->is_on()) {
165 XBT_DEBUG("Insert host %s to watched_hosts because it's off and %s needs to restart", get_host()->get_cname(),
167 watched_hosts().insert(get_host()->get_name());
171 // Execute the termination callbacks
172 bool failed = context_->wannadie();
173 for (auto exit_fun = on_exit->crbegin(); exit_fun != on_exit->crend(); ++exit_fun)
179 /* cancel non-blocking activities */
180 for (auto activity : activities_)
184 XBT_DEBUG("%s@%s(%ld) should not run anymore", get_cname(), get_host()->get_cname(), get_pid());
186 if (this == simix_global->maestro_) /* Do not cleanup maestro */
189 XBT_DEBUG("Cleanup actor %s (%p), waiting synchro %p", get_cname(), this, waiting_synchro_.get());
191 /* Unregister associated timers if any */
192 if (kill_timer_ != nullptr) {
193 kill_timer_->remove();
194 kill_timer_ = nullptr;
196 if (simcall_.timeout_cb_) {
197 simcall_.timeout_cb_->remove();
198 simcall_.timeout_cb_ = nullptr;
201 cleanup_from_simix();
203 context_->set_wannadie(false); // don't let the simcall's yield() do a Context::stop(), to avoid infinite loops
204 actor::simcall([this] { s4u::Actor::on_termination(*get_ciface()); });
205 context_->set_wannadie();
208 void ActorImpl::exit()
210 context_->set_wannadie();
212 exception_ = nullptr;
214 /* destroy the blocking synchro if any */
215 if (waiting_synchro_ != nullptr) {
216 waiting_synchro_->cancel();
217 waiting_synchro_->state_ = activity::State::FAILED;
219 activity::ExecImplPtr exec = boost::dynamic_pointer_cast<activity::ExecImpl>(waiting_synchro_);
220 activity::CommImplPtr comm = boost::dynamic_pointer_cast<activity::CommImpl>(waiting_synchro_);
222 if (exec != nullptr) {
223 exec->clean_action();
224 } else if (comm != nullptr) {
225 comm->unregister_simcall(&simcall_);
227 activity::ActivityImplPtr(waiting_synchro_)->finish();
230 activities_.remove(waiting_synchro_);
231 waiting_synchro_ = nullptr;
233 for (auto const& activity : activities_)
237 // Forcefully kill the actor if its host is turned off. Not a HostFailureException because you should not survive that
238 this->throw_exception(std::make_exception_ptr(ForcefulKillException(host_->is_on() ? "exited" : "host failed")));
241 void ActorImpl::kill(ActorImpl* actor) const
243 xbt_assert(actor != simix_global->maestro_, "Killing maestro is a rather bad idea");
244 if (actor->finished_) {
245 XBT_DEBUG("Ignoring request to kill actor %s@%s that is already dead", actor->get_cname(),
246 actor->host_->get_cname());
250 XBT_DEBUG("Actor '%s'@%s is killing actor '%s'@%s", get_cname(), host_ ? host_->get_cname() : "", actor->get_cname(),
251 actor->host_ ? actor->host_->get_cname() : "");
256 XBT_DEBUG("Go on, this is a suicide,");
258 EngineImpl::get_instance()->add_actor_to_run_list(actor);
261 void ActorImpl::kill_all() const
263 for (auto const& kv : simix_global->process_list)
264 if (kv.second != this)
265 this->kill(kv.second);
268 void ActorImpl::set_kill_time(double kill_time)
270 if (kill_time <= SIMIX_get_clock())
272 XBT_DEBUG("Set kill time %f for actor %s@%s", kill_time, get_cname(), host_->get_cname());
273 kill_timer_ = timer::Timer::set(kill_time, [this] {
275 kill_timer_ = nullptr;
279 double ActorImpl::get_kill_time() const
281 return kill_timer_ ? kill_timer_->get_date() : 0.0;
284 void ActorImpl::yield()
286 XBT_DEBUG("Yield actor '%s'", get_cname());
288 /* Go into sleep and return control to maestro */
291 /* Ok, maestro returned control to us */
292 XBT_DEBUG("Control returned to me: '%s'", get_cname());
294 if (context_->wannadie()) {
295 XBT_DEBUG("Actor %s@%s is dead", get_cname(), host_->get_cname());
301 XBT_DEBUG("Hey! I'm suspended.");
302 xbt_assert(exception_ == nullptr, "Gasp! This exception may be lost by subsequent calls.");
303 yield(); // Yield back to maestro without proceeding with my execution. I'll get rescheduled by resume()
306 if (exception_ != nullptr) {
307 XBT_DEBUG("Wait, maestro left me an exception");
308 std::exception_ptr exception = std::move(exception_);
309 exception_ = nullptr;
311 std::rethrow_exception(std::move(exception));
312 } catch (const simgrid::Exception& e) {
313 e.rethrow_nested(XBT_THROW_POINT, boost::core::demangle(typeid(e).name()) + " raised in kernel mode.");
317 if (SMPI_switch_data_segment && not finished_) {
318 SMPI_switch_data_segment(get_iface());
322 /** This actor will be terminated automatically when the last non-daemon actor finishes */
323 void ActorImpl::daemonize()
327 EngineImpl::get_instance()->add_daemon(this);
331 void ActorImpl::undaemonize()
335 EngineImpl::get_instance()->rm_daemon(this);
339 s4u::Actor* ActorImpl::restart()
341 xbt_assert(this != simix_global->maestro_, "Restarting maestro is not supported");
343 XBT_DEBUG("Restarting actor %s on %s", get_cname(), host_->get_cname());
345 // retrieve the arguments of the old actor
346 ProcessArg arg(host_, this);
348 // kill the old actor
349 context::Context::self()->get_actor()->kill(this);
351 // start the new actor
352 ActorImplPtr actor = ActorImpl::create(arg.name, arg.code, arg.data, arg.host, nullptr);
353 actor->set_properties(arg.properties);
354 *actor->on_exit = std::move(*arg.on_exit);
355 actor->set_kill_time(arg.kill_time);
356 actor->set_auto_restart(arg.auto_restart);
358 return actor->get_ciface();
361 void ActorImpl::suspend()
364 XBT_DEBUG("Actor '%s' is already suspended", get_cname());
370 /* Suspend the activities associated with this actor. */
371 for (auto const& activity : activities_)
375 void ActorImpl::resume()
377 XBT_IN("actor = %p", this);
379 if (context_->wannadie()) {
380 XBT_VERB("Ignoring request to suspend an actor that is currently dying.");
388 /* resume the activities that were blocked when suspending the actor. */
389 for (auto const& activity : activities_)
391 if (not waiting_synchro_) // Reschedule the actor if it was forcefully unscheduled in yield()
392 EngineImpl::get_instance()->add_actor_to_run_list_no_check(this);
397 activity::ActivityImplPtr ActorImpl::join(const ActorImpl* actor, double timeout)
399 activity::ActivityImplPtr sleep = this->sleep(timeout);
400 actor->on_exit->emplace_back([sleep](bool) {
401 if (sleep->surf_action_)
402 sleep->surf_action_->finish(resource::Action::State::FINISHED);
407 activity::ActivityImplPtr ActorImpl::sleep(double duration)
409 if (not host_->is_on())
410 throw_exception(std::make_exception_ptr(HostFailureException(
411 XBT_THROW_POINT, std::string("Host ") + host_->get_cname() + " failed, you cannot sleep there.")));
413 auto sleep = new activity::SleepImpl();
414 sleep->set_name("sleep").set_host(host_).set_duration(duration).start();
415 return activity::SleepImplPtr(sleep);
418 void ActorImpl::throw_exception(std::exception_ptr e)
425 /* cancel the blocking synchro if any */
426 if (waiting_synchro_) {
427 waiting_synchro_->cancel();
428 activities_.remove(waiting_synchro_);
429 waiting_synchro_ = nullptr;
433 void ActorImpl::simcall_answer()
435 if (this != simix_global->maestro_) {
436 XBT_DEBUG("Answer simcall %s issued by %s (%p)", SIMIX_simcall_name(simcall_), get_cname(), this);
437 xbt_assert(simcall_.call_ != simix::Simcall::NONE);
438 simcall_.call_ = simix::Simcall::NONE;
439 auto* engine = EngineImpl::get_instance();
440 const auto& actors_to_run = engine->get_actors_to_run();
441 xbt_assert(not XBT_LOG_ISENABLED(simix_process, xbt_log_priority_debug) ||
442 std::find(begin(actors_to_run), end(actors_to_run), this) == end(actors_to_run),
443 "Actor %p should not exist in actors_to_run!", this);
444 engine->add_actor_to_run_list_no_check(this);
448 void ActorImpl::set_host(s4u::Host* dest)
450 host_->get_impl()->remove_actor(this);
452 dest->get_impl()->add_actor(this);
455 ActorImplPtr ActorImpl::init(const std::string& name, s4u::Host* host) const
457 auto* actor = new ActorImpl(xbt::string(name), host);
458 actor->set_ppid(this->pid_);
460 intrusive_ptr_add_ref(actor);
461 /* The on_creation() signal must be delayed until there, where the pid and everything is set */
462 s4u::Actor::on_creation(*actor->get_ciface());
464 return ActorImplPtr(actor);
467 ActorImpl* ActorImpl::start(const ActorCode& code)
469 xbt_assert(code && host_ != nullptr, "Invalid parameters");
471 if (not host_->is_on()) {
472 XBT_WARN("Cannot launch actor '%s' on failed host '%s'", name_.c_str(), host_->get_cname());
473 intrusive_ptr_release(this);
474 throw HostFailureException(XBT_THROW_POINT, "Cannot start actor on failed host.");
478 XBT_VERB("Create context %s", get_cname());
479 context_.reset(simix_global->context_factory->create_context(ActorCode(code), this));
481 XBT_DEBUG("Start context '%s'", get_cname());
483 /* Add the actor to its host's actor list */
484 host_->get_impl()->add_actor(this);
485 simix_global->process_list[pid_] = this;
487 /* Now insert it in the global actor list and in the actor to run list */
488 EngineImpl::get_instance()->add_actor_to_run_list_no_check(this);
493 ActorImplPtr ActorImpl::create(const std::string& name, const ActorCode& code, void* data, s4u::Host* host,
494 const ActorImpl* parent_actor)
496 XBT_DEBUG("Start actor %s@'%s'", name.c_str(), host->get_cname());
499 if (parent_actor != nullptr)
500 actor = parent_actor->init(xbt::string(name), host);
502 actor = self()->init(xbt::string(name), host);
505 actor->set_user_data(data);
512 void create_maestro(const std::function<void()>& code)
514 /* Create maestro actor and initialize it */
515 auto* maestro = new ActorImpl(xbt::string(""), /*host*/ nullptr);
518 maestro->context_.reset(simix_global->context_factory->create_context(ActorCode(), maestro));
520 maestro->context_.reset(simix_global->context_factory->create_maestro(ActorCode(code), maestro));
523 maestro->simcall_.issuer_ = maestro;
524 simix_global->maestro_ = maestro;
528 } // namespace kernel
529 } // namespace simgrid
531 int SIMIX_process_count() // XBT_ATTRIB_DEPRECATED_v329
533 return simix_global->process_list.size();
536 void* SIMIX_process_self_get_data() // XBT_ATTRIB_DEPRECATED_v329
538 smx_actor_t self = simgrid::kernel::actor::ActorImpl::self();
540 if (self == nullptr) {
543 return self->get_user_data();
546 void SIMIX_process_self_set_data(void* data) // XBT_ATTRIB_DEPRECATED_v329
548 simgrid::kernel::actor::ActorImpl::self()->set_user_data(data);
551 /* needs to be public and without simcall because it is called
552 by exceptions and logging events */
553 const char* SIMIX_process_self_get_name()
555 return SIMIX_is_maestro() ? "maestro" : simgrid::kernel::actor::ActorImpl::self()->get_cname();
558 /** @brief Returns the process from PID. */
559 smx_actor_t SIMIX_process_from_PID(aid_t pid)
561 return simgrid::kernel::actor::ActorImpl::by_pid(pid);
564 void SIMIX_process_on_exit(smx_actor_t actor,
565 const std::function<void(bool /*failed*/)>& fun) // XBT_ATTRIB_DEPRECATED_v329
567 xbt_assert(actor, "current process not found: are you in maestro context ?");
568 actor->on_exit->emplace_back(fun);
571 void simcall_process_set_data(smx_actor_t process, void* data) // XBT_ATTRIB_DEPRECATED_v329
573 simgrid::kernel::actor::simcall([process, data] { process->set_user_data(data); });