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/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/AppSide.hpp"
17 #include "src/simix/smx_private.hpp"
18 #include "src/surf/HostImpl.hpp"
19 #include "src/surf/cpu_interface.hpp"
21 #include <boost/core/demangle.hpp>
24 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_process, simix, "Logging specific to SIMIX (process)");
27 * @brief Returns the current agent.
29 * This functions returns the currently running SIMIX process.
31 * @return The SIMIX process
33 smx_actor_t SIMIX_process_self()
35 return simgrid::kernel::actor::ActorImpl::self();
42 static unsigned long maxpid = 0;
43 unsigned long get_maxpid()
47 unsigned long* get_maxpid_addr()
51 ActorImpl* ActorImpl::by_pid(aid_t pid)
53 auto item = simix_global->process_list.find(pid);
54 if (item != simix_global->process_list.end())
58 for (auto& a : simix_global->actors_to_destroy)
59 if (a.get_pid() == pid)
61 return nullptr; // Not found, even in the trash
64 ActorImpl* ActorImpl::self()
66 const context::Context* self_context = context::Context::self();
68 return (self_context != nullptr) ? self_context->get_actor() : nullptr;
71 ActorImpl::ActorImpl(xbt::string name, s4u::Host* host) : host_(host), name_(std::move(name)), piface_(this)
74 simcall_.issuer_ = this;
75 stacksize_ = smx_context_stack_size;
78 ActorImpl::~ActorImpl()
80 if (simix_global != nullptr && this != simix_global->maestro_)
81 s4u::Actor::on_destruction(*get_ciface());
84 /* Become an actor in the simulation
86 * Currently this can only be called by the main thread (once) and only work with some thread factories
87 * (currently ThreadContextFactory).
89 * In the future, it might be extended in order to attach other threads created by a third party library.
92 ActorImplPtr ActorImpl::attach(const std::string& name, void* data, s4u::Host* host)
94 // This is mostly a copy/paste from create(), it'd be nice to share some code between those two functions.
96 XBT_DEBUG("Attach actor %s on host '%s'", name.c_str(), host->get_cname());
98 if (not host->is_on()) {
99 XBT_WARN("Cannot attach actor '%s' on failed host '%s'", name.c_str(), host->get_cname());
100 throw HostFailureException(XBT_THROW_POINT, "Cannot attach actor on failed host.");
103 auto* actor = new ActorImpl(xbt::string(name), host);
105 actor->set_user_data(data);
106 actor->code_ = nullptr;
108 XBT_VERB("Create context %s", actor->get_cname());
109 xbt_assert(simix_global != nullptr, "simix is not initialized, please call MSG_init first");
110 actor->context_.reset(simix_global->context_factory->attach(actor));
112 /* Add the actor to it's host actor list */
113 host->get_impl()->add_actor(actor);
115 /* Now insert it in the global actor list and in the actors to run list */
116 simix_global->process_list[actor->get_pid()] = actor;
117 XBT_DEBUG("Inserting [%p] %s(%s) in the to_run list", actor, actor->get_cname(), host->get_cname());
118 simix_global->actors_to_run.push_back(actor);
119 intrusive_ptr_add_ref(actor);
121 auto* context = dynamic_cast<context::AttachContext*>(actor->context_.get());
122 xbt_assert(nullptr != context, "Not a suitable context");
123 context->attach_start();
125 /* The on_creation() signal must be delayed until there, where the pid and everything is set */
126 s4u::Actor::on_creation(*actor->get_ciface());
128 return ActorImplPtr(actor);
130 /** @brief Detach an actor attached with `attach()`
132 * This is called when the current actor has finished its job.
133 * Used in the main thread, it waits for the simulation to finish before returning. When it returns, the other
134 * simulated actors and the maestro are destroyed.
136 void ActorImpl::detach()
138 auto* context = dynamic_cast<context::AttachContext*>(context::Context::self());
139 xbt_assert(context != nullptr, "Not a suitable context");
141 context->get_actor()->cleanup();
142 context->attach_stop();
145 void ActorImpl::cleanup_from_simix()
147 const std::lock_guard<std::mutex> lock(simix_global->mutex);
148 simix_global->process_list.erase(pid_);
149 if (host_ && host_actor_list_hook.is_linked())
150 host_->get_impl()->remove_actor(this);
151 if (not smx_destroy_list_hook.is_linked()) {
153 xbt_dynar_push_as(simix_global->dead_actors_vector, ActorImpl*, this);
155 simix_global->actors_to_destroy.push_back(*this);
159 void ActorImpl::cleanup()
163 if (has_to_auto_restart() && not get_host()->is_on()) {
164 XBT_DEBUG("Insert host %s to watched_hosts because it's off and %s needs to restart", get_host()->get_cname(),
166 watched_hosts().insert(get_host()->get_name());
170 // Execute the termination callbacks
171 bool failed = context_->wannadie();
172 for (auto exit_fun = on_exit->crbegin(); exit_fun != on_exit->crend(); ++exit_fun)
178 /* cancel non-blocking activities */
179 for (auto activity : activities_)
183 XBT_DEBUG("%s@%s(%ld) should not run anymore", get_cname(), get_host()->get_cname(), get_pid());
185 if (this == simix_global->maestro_) /* Do not cleanup maestro */
188 XBT_DEBUG("Cleanup actor %s (%p), waiting synchro %p", get_cname(), this, waiting_synchro_.get());
190 /* Unregister associated timers if any */
191 if (kill_timer_ != nullptr) {
192 kill_timer_->remove();
193 kill_timer_ = nullptr;
195 if (simcall_.timeout_cb_) {
196 simcall_.timeout_cb_->remove();
197 simcall_.timeout_cb_ = nullptr;
200 cleanup_from_simix();
202 context_->set_wannadie(false); // don't let the simcall's yield() do a Context::stop(), to avoid infinite loops
203 actor::simcall([this] { s4u::Actor::on_termination(*get_ciface()); });
204 context_->set_wannadie();
207 void ActorImpl::exit()
209 context_->set_wannadie();
211 exception_ = nullptr;
213 /* destroy the blocking synchro if any */
214 if (waiting_synchro_ != nullptr) {
215 waiting_synchro_->cancel();
216 waiting_synchro_->state_ = activity::State::FAILED;
218 activity::ExecImplPtr exec = boost::dynamic_pointer_cast<activity::ExecImpl>(waiting_synchro_);
219 activity::CommImplPtr comm = boost::dynamic_pointer_cast<activity::CommImpl>(waiting_synchro_);
221 if (exec != nullptr) {
222 exec->clean_action();
223 } else if (comm != nullptr) {
224 comm->unregister_simcall(&simcall_);
226 activity::ActivityImplPtr(waiting_synchro_)->finish();
229 activities_.remove(waiting_synchro_);
230 waiting_synchro_ = nullptr;
233 // Forcefully kill the actor if its host is turned off. Not a HostFailureException because you should not survive that
234 this->throw_exception(std::make_exception_ptr(ForcefulKillException(host_->is_on() ? "exited" : "host failed")));
237 void ActorImpl::kill(ActorImpl* actor) const
239 xbt_assert(actor != simix_global->maestro_, "Killing maestro is a rather bad idea");
240 if (actor->finished_) {
241 XBT_DEBUG("Ignoring request to kill actor %s@%s that is already dead", actor->get_cname(),
242 actor->host_->get_cname());
246 XBT_DEBUG("Actor '%s'@%s is killing actor '%s'@%s", get_cname(), host_ ? host_->get_cname() : "", actor->get_cname(),
247 actor->host_ ? actor->host_->get_cname() : "");
252 XBT_DEBUG("Go on, this is a suicide,");
253 } else if (std::find(begin(simix_global->actors_to_run), end(simix_global->actors_to_run), actor) !=
254 end(simix_global->actors_to_run)) {
255 XBT_DEBUG("Actor %s is already in the to_run list", actor->get_cname());
257 XBT_DEBUG("Inserting %s in the to_run list", actor->get_cname());
258 simix_global->actors_to_run.push_back(actor);
262 void ActorImpl::kill_all() const
264 for (auto const& kv : simix_global->process_list)
265 if (kv.second != this)
266 this->kill(kv.second);
269 void ActorImpl::set_kill_time(double kill_time)
271 if (kill_time <= SIMIX_get_clock())
273 XBT_DEBUG("Set kill time %f for actor %s@%s", kill_time, get_cname(), host_->get_cname());
274 kill_timer_ = simix::Timer::set(kill_time, [this] {
276 kill_timer_ = nullptr;
280 double ActorImpl::get_kill_time() const
282 return kill_timer_ ? kill_timer_->date : 0.0;
285 void ActorImpl::yield()
287 XBT_DEBUG("Yield actor '%s'", get_cname());
289 /* Go into sleep and return control to maestro */
292 /* Ok, maestro returned control to us */
293 XBT_DEBUG("Control returned to me: '%s'", get_cname());
295 if (context_->wannadie()) {
296 XBT_DEBUG("Actor %s@%s is dead", get_cname(), host_->get_cname());
302 XBT_DEBUG("Hey! I'm suspended.");
304 xbt_assert(exception_ == nullptr, "Gasp! This exception may be lost by subsequent calls.");
306 if (waiting_synchro_ != nullptr) // Not sure of when this will happen. Maybe when suspending early in the SR when a
307 waiting_synchro_->suspend(); // waiting_synchro was terminated
309 yield(); // Yield back to maestro without proceeding with my execution. I'll get rescheduled by resume()
312 if (exception_ != nullptr) {
313 XBT_DEBUG("Wait, maestro left me an exception");
314 std::exception_ptr exception = std::move(exception_);
315 exception_ = nullptr;
317 std::rethrow_exception(std::move(exception));
318 } catch (const simgrid::Exception& e) {
319 e.rethrow_nested(XBT_THROW_POINT, boost::core::demangle(typeid(e).name()) + " raised in kernel mode.");
323 if (SMPI_switch_data_segment && not finished_) {
324 SMPI_switch_data_segment(get_iface());
328 /** This actor will be terminated automatically when the last non-daemon actor finishes */
329 void ActorImpl::daemonize()
333 simix_global->daemons.push_back(this);
337 void ActorImpl::undaemonize()
340 auto& vect = simix_global->daemons;
341 auto it = std::find(vect.begin(), vect.end(), this);
342 xbt_assert(it != vect.end(), "The dying daemon is not a daemon after all. Please report that bug.");
343 /* Don't move the whole content since we don't really care about the order */
345 std::swap(*it, vect.back());
351 s4u::Actor* ActorImpl::restart()
353 xbt_assert(this != simix_global->maestro_, "Restarting maestro is not supported");
355 XBT_DEBUG("Restarting actor %s on %s", get_cname(), host_->get_cname());
357 // retrieve the arguments of the old actor
358 ProcessArg arg(host_, this);
360 // kill the old actor
361 context::Context::self()->get_actor()->kill(this);
363 // start the new actor
364 ActorImplPtr actor = ActorImpl::create(arg.name, arg.code, arg.data, arg.host, nullptr);
365 actor->set_properties(arg.properties);
366 *actor->on_exit = std::move(*arg.on_exit);
367 actor->set_kill_time(arg.kill_time);
368 actor->set_auto_restart(arg.auto_restart);
370 return actor->get_ciface();
373 void ActorImpl::suspend()
376 XBT_DEBUG("Actor '%s' is already suspended", get_cname());
382 /* If the suspended actor is waiting on a sync, suspend its synchronization.
383 * Otherwise, it will suspend itself when scheduled, ie, very soon. */
384 if (waiting_synchro_ != nullptr)
385 waiting_synchro_->suspend();
388 void ActorImpl::resume()
390 XBT_IN("actor = %p", this);
392 if (context_->wannadie()) {
393 XBT_VERB("Ignoring request to suspend an actor that is currently dying.");
401 /* resume the activity that was blocking the resumed actor. */
402 if (waiting_synchro_)
403 waiting_synchro_->resume();
404 else // Reschedule the actor if it was forcefully unscheduled in yield()
405 simix_global->actors_to_run.push_back(this);
410 activity::ActivityImplPtr ActorImpl::join(const ActorImpl* actor, double timeout)
412 activity::ActivityImplPtr sleep = this->sleep(timeout);
413 actor->on_exit->emplace_back([sleep](bool) {
414 if (sleep->surf_action_)
415 sleep->surf_action_->finish(resource::Action::State::FINISHED);
420 activity::ActivityImplPtr ActorImpl::sleep(double duration)
422 if (not host_->is_on())
423 throw_exception(std::make_exception_ptr(HostFailureException(
424 XBT_THROW_POINT, std::string("Host ") + host_->get_cname() + " failed, you cannot sleep there.")));
426 auto sleep = new activity::SleepImpl();
427 sleep->set_name("sleep").set_host(host_).set_duration(duration).start();
428 return activity::SleepImplPtr(sleep);
431 void ActorImpl::throw_exception(std::exception_ptr e)
438 /* cancel the blocking synchro if any */
439 if (waiting_synchro_) {
440 waiting_synchro_->cancel();
441 activities_.remove(waiting_synchro_);
442 waiting_synchro_ = nullptr;
446 void ActorImpl::simcall_answer()
448 if (this != simix_global->maestro_) {
449 XBT_DEBUG("Answer simcall %s (%d) issued by %s (%p)", SIMIX_simcall_name(simcall_.call_), (int)simcall_.call_,
451 xbt_assert(simcall_.call_ != simix::Simcall::NONE);
452 simcall_.call_ = simix::Simcall::NONE;
453 xbt_assert(not XBT_LOG_ISENABLED(simix_process, xbt_log_priority_debug) ||
454 std::find(begin(simix_global->actors_to_run), end(simix_global->actors_to_run), this) ==
455 end(simix_global->actors_to_run),
456 "Actor %p should not exist in actors_to_run!", this);
457 simix_global->actors_to_run.push_back(this);
461 void ActorImpl::set_host(s4u::Host* dest)
463 host_->get_impl()->remove_actor(this);
465 dest->get_impl()->add_actor(this);
468 ActorImplPtr ActorImpl::init(const std::string& name, s4u::Host* host) const
470 auto* actor = new ActorImpl(xbt::string(name), host);
471 actor->set_ppid(this->pid_);
473 intrusive_ptr_add_ref(actor);
474 /* The on_creation() signal must be delayed until there, where the pid and everything is set */
475 s4u::Actor::on_creation(*actor->get_ciface());
477 return ActorImplPtr(actor);
480 ActorImpl* ActorImpl::start(const ActorCode& code)
482 xbt_assert(code && host_ != nullptr, "Invalid parameters");
484 if (not host_->is_on()) {
485 XBT_WARN("Cannot launch actor '%s' on failed host '%s'", name_.c_str(), host_->get_cname());
486 intrusive_ptr_release(this);
487 throw HostFailureException(XBT_THROW_POINT, "Cannot start actor on failed host.");
491 XBT_VERB("Create context %s", get_cname());
492 context_.reset(simix_global->context_factory->create_context(ActorCode(code), this));
494 XBT_DEBUG("Start context '%s'", get_cname());
496 /* Add the actor to its host's actor list */
497 host_->get_impl()->add_actor(this);
498 simix_global->process_list[pid_] = this;
500 /* Now insert it in the global actor list and in the actor to run list */
501 XBT_DEBUG("Inserting [%p] %s(%s) in the to_run list", this, get_cname(), host_->get_cname());
502 simix_global->actors_to_run.push_back(this);
507 ActorImplPtr ActorImpl::create(const std::string& name, const ActorCode& code, void* data, s4u::Host* host,
508 const ActorImpl* parent_actor)
510 XBT_DEBUG("Start actor %s@'%s'", name.c_str(), host->get_cname());
513 if (parent_actor != nullptr)
514 actor = parent_actor->init(xbt::string(name), host);
516 actor = self()->init(xbt::string(name), host);
519 actor->set_user_data(data);
526 void create_maestro(const std::function<void()>& code)
528 /* Create maestro actor and initialize it */
529 auto* maestro = new ActorImpl(xbt::string(""), /*host*/ nullptr);
532 maestro->context_.reset(simix_global->context_factory->create_context(ActorCode(), maestro));
534 maestro->context_.reset(simix_global->context_factory->create_maestro(ActorCode(code), maestro));
537 maestro->simcall_.issuer_ = maestro;
538 simix_global->maestro_ = maestro;
542 } // namespace kernel
543 } // namespace simgrid
545 int SIMIX_process_count() // XBT_ATTRIB_DEPRECATED_v329
547 return simix_global->process_list.size();
550 void* SIMIX_process_self_get_data() // XBT_ATTRIB_DEPRECATED_v329
552 smx_actor_t self = simgrid::kernel::actor::ActorImpl::self();
554 if (self == nullptr) {
557 return self->get_user_data();
560 void SIMIX_process_self_set_data(void* data) // XBT_ATTRIB_DEPRECATED_v329
562 simgrid::kernel::actor::ActorImpl::self()->set_user_data(data);
565 /* needs to be public and without simcall because it is called
566 by exceptions and logging events */
567 const char* SIMIX_process_self_get_name()
569 return SIMIX_is_maestro() ? "maestro" : simgrid::kernel::actor::ActorImpl::self()->get_cname();
572 /** @brief Returns the process from PID. */
573 smx_actor_t SIMIX_process_from_PID(aid_t pid)
575 return simgrid::kernel::actor::ActorImpl::by_pid(pid);
578 void SIMIX_process_on_exit(smx_actor_t actor,
579 const std::function<void(bool /*failed*/)>& fun) // XBT_ATTRIB_DEPRECATED_v329
581 xbt_assert(actor, "current process not found: are you in maestro context ?");
582 actor->on_exit->emplace_back(fun);
585 void simcall_process_set_data(smx_actor_t process, void* data) // XBT_ATTRIB_DEPRECATED_v329
587 simgrid::kernel::actor::simcall([process, data] { process->set_user_data(data); });