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>
22 #include <boost/range/algorithm.hpp>
25 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_process, simix, "Logging specific to SIMIX (process)");
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 return simgrid::kernel::actor::ActorImpl::self();
43 static unsigned long maxpid = 0;
44 unsigned long get_maxpid()
48 ActorImpl* ActorImpl::by_PID(aid_t PID)
50 auto item = simix_global->process_list.find(PID);
51 if (item != simix_global->process_list.end())
55 for (auto& a : simix_global->actors_to_destroy)
56 if (a.get_pid() == PID)
58 return nullptr; // Not found, even in the trash
61 ActorImpl* ActorImpl::self()
63 const context::Context* self_context = context::Context::self();
65 return (self_context != nullptr) ? self_context->get_actor() : nullptr;
68 ActorImpl::ActorImpl(xbt::string name, s4u::Host* host) : host_(host), name_(std::move(name)), piface_(this)
71 simcall_.issuer_ = this;
72 stacksize_ = smx_context_stack_size;
75 ActorImpl::~ActorImpl()
77 if (simix_global != nullptr && this != simix_global->maestro_)
78 s4u::Actor::on_destruction(*get_ciface());
81 /* Become an actor in the simulation
83 * Currently this can only be called by the main thread (once) and only work with some thread factories
84 * (currently ThreadContextFactory).
86 * In the future, it might be extended in order to attach other threads created by a third party library.
89 ActorImplPtr ActorImpl::attach(const std::string& name, void* data, s4u::Host* host,
90 const std::unordered_map<std::string, std::string>* properties)
92 // This is mostly a copy/paste from create(), it'd be nice to share some code between those two functions.
94 XBT_DEBUG("Attach actor %s on host '%s'", name.c_str(), host->get_cname());
96 if (not host->is_on()) {
97 XBT_WARN("Cannot attach actor '%s' on failed host '%s'", name.c_str(), host->get_cname());
98 throw HostFailureException(XBT_THROW_POINT, "Cannot attach actor on failed host.");
101 auto* actor = new ActorImpl(xbt::string(name), host);
103 actor->set_user_data(data);
104 actor->code_ = nullptr;
106 XBT_VERB("Create context %s", actor->get_cname());
107 xbt_assert(simix_global != nullptr, "simix is not initialized, please call MSG_init first");
108 actor->context_.reset(simix_global->context_factory->attach(actor));
111 if (properties != nullptr)
112 actor->set_properties(*properties);
114 /* Add the actor to it's host actor list */
115 host->pimpl_->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 XBT_DEBUG("Inserting [%p] %s(%s) in the to_run list", actor, actor->get_cname(), host->get_cname());
120 simix_global->actors_to_run.push_back(actor);
121 intrusive_ptr_add_ref(actor);
123 auto* context = dynamic_cast<context::AttachContext*>(actor->context_.get());
124 xbt_assert(nullptr != context, "Not a suitable context");
125 context->attach_start();
127 /* The on_creation() signal must be delayed until there, where the pid and everything is set */
128 s4u::Actor::on_creation(*actor->get_ciface());
130 return ActorImplPtr(actor);
132 /** @brief Detach an actor attached with `attach()`
134 * This is called when the current actor has finished its job.
135 * Used in the main thread, it waits for the simulation to finish before returning. When it returns, the other
136 * simulated actors and the maestro are destroyed.
138 void ActorImpl::detach()
140 auto* context = dynamic_cast<context::AttachContext*>(context::Context::self());
141 xbt_assert(context != nullptr, "Not a suitable context");
143 context->get_actor()->cleanup();
144 context->attach_stop();
147 void ActorImpl::cleanup_from_simix()
149 const std::lock_guard<std::mutex> lock(simix_global->mutex);
150 simix_global->process_list.erase(pid_);
151 if (host_ && host_actor_list_hook.is_linked())
152 host_->pimpl_->remove_actor(this);
153 if (not smx_destroy_list_hook.is_linked()) {
155 xbt_dynar_push_as(simix_global->dead_actors_vector, ActorImpl*, this);
157 simix_global->actors_to_destroy.push_back(*this);
161 void ActorImpl::cleanup()
165 if (has_to_auto_restart() && not get_host()->is_on()) {
166 XBT_DEBUG("Insert host %s to watched_hosts because it's off and %s needs to restart", get_host()->get_cname(),
168 watched_hosts().insert(get_host()->get_name());
172 // Execute the termination callbacks
173 bool failed = context_->wannadie();
174 for (auto exit_fun = on_exit->crbegin(); exit_fun != on_exit->crend(); ++exit_fun)
180 /* cancel non-blocking activities */
181 for (auto activity : activities_)
185 XBT_DEBUG("%s@%s(%ld) should not run anymore", get_cname(), get_host()->get_cname(), get_pid());
187 if (this == simix_global->maestro_) /* Do not cleanup maestro */
190 XBT_DEBUG("Cleanup actor %s (%p), waiting synchro %p", get_cname(), this, waiting_synchro_.get());
192 /* Unregister associated timers if any */
193 if (kill_timer_ != nullptr) {
194 kill_timer_->remove();
195 kill_timer_ = nullptr;
197 if (simcall_.timeout_cb_) {
198 simcall_.timeout_cb_->remove();
199 simcall_.timeout_cb_ = nullptr;
202 cleanup_from_simix();
204 context_->set_wannadie(false); // don't let the simcall's yield() do a Context::stop(), to avoid infinite loops
205 actor::simcall([this] { s4u::Actor::on_termination(*get_ciface()); });
206 context_->set_wannadie();
209 void ActorImpl::exit()
211 context_->set_wannadie();
213 exception_ = nullptr;
215 /* destroy the blocking synchro if any */
216 if (waiting_synchro_ != nullptr) {
217 waiting_synchro_->cancel();
218 waiting_synchro_->state_ = activity::State::FAILED;
220 activity::ExecImplPtr exec = boost::dynamic_pointer_cast<activity::ExecImpl>(waiting_synchro_);
221 activity::CommImplPtr comm = boost::dynamic_pointer_cast<activity::CommImpl>(waiting_synchro_);
223 if (exec != nullptr) {
224 exec->clean_action();
225 } else if (comm != nullptr) {
226 // Remove first occurrence of &actor->simcall:
227 auto i = boost::range::find(waiting_synchro_->simcalls_, &simcall_);
228 if (i != waiting_synchro_->simcalls_.end())
229 waiting_synchro_->simcalls_.remove(&simcall_);
231 activity::ActivityImplPtr(waiting_synchro_)->finish();
234 activities_.remove(waiting_synchro_);
235 waiting_synchro_ = nullptr;
238 // Forcefully kill the actor if its host is turned off. Not a HostFailureException because you should not survive that
239 this->throw_exception(std::make_exception_ptr(ForcefulKillException(host_->is_on() ? "exited" : "host failed")));
242 void ActorImpl::kill(ActorImpl* actor) const
244 xbt_assert(actor != simix_global->maestro_, "Killing maestro is a rather bad idea");
245 if (actor->finished_) {
246 XBT_DEBUG("Ignoring request to kill actor %s@%s that is already dead", actor->get_cname(),
247 actor->host_->get_cname());
251 XBT_DEBUG("Actor '%s'@%s is killing actor '%s'@%s", get_cname(), host_ ? host_->get_cname() : "", actor->get_cname(),
252 actor->host_ ? actor->host_->get_cname() : "");
257 XBT_DEBUG("Go on, this is a suicide,");
258 } else if (std::find(begin(simix_global->actors_to_run), end(simix_global->actors_to_run), actor) !=
259 end(simix_global->actors_to_run)) {
260 XBT_DEBUG("Actor %s is already in the to_run list", actor->get_cname());
262 XBT_DEBUG("Inserting %s in the to_run list", actor->get_cname());
263 simix_global->actors_to_run.push_back(actor);
267 void ActorImpl::kill_all() const
269 for (auto const& kv : simix_global->process_list)
270 if (kv.second != this)
271 this->kill(kv.second);
274 void ActorImpl::set_kill_time(double kill_time)
276 if (kill_time <= SIMIX_get_clock())
278 XBT_DEBUG("Set kill time %f for actor %s@%s", kill_time, get_cname(), host_->get_cname());
279 kill_timer_ = simix::Timer::set(kill_time, [this] {
281 kill_timer_ = nullptr;
285 double ActorImpl::get_kill_time() const
287 return kill_timer_ ? kill_timer_->get_date() : 0;
290 void ActorImpl::yield()
292 XBT_DEBUG("Yield actor '%s'", get_cname());
294 /* Go into sleep and return control to maestro */
297 /* Ok, maestro returned control to us */
298 XBT_DEBUG("Control returned to me: '%s'", get_cname());
300 if (context_->wannadie()) {
301 XBT_DEBUG("Actor %s@%s is dead", get_cname(), host_->get_cname());
307 XBT_DEBUG("Hey! I'm suspended.");
309 xbt_assert(exception_ == nullptr, "Gasp! This exception may be lost by subsequent calls.");
311 if (waiting_synchro_ != nullptr) // Not sure of when this will happen. Maybe when suspending early in the SR when a
312 waiting_synchro_->suspend(); // waiting_synchro was terminated
314 yield(); // Yield back to maestro without proceeding with my execution. I'll get rescheduled by resume()
317 if (exception_ != nullptr) {
318 XBT_DEBUG("Wait, maestro left me an exception");
319 std::exception_ptr exception = std::move(exception_);
320 exception_ = nullptr;
322 std::rethrow_exception(std::move(exception));
323 } catch (const simgrid::Exception& e) {
324 e.rethrow_nested(XBT_THROW_POINT, boost::core::demangle(typeid(e).name()) + " raised in kernel mode.");
328 if (SMPI_switch_data_segment && not finished_) {
329 SMPI_switch_data_segment(get_iface());
333 /** This actor will be terminated automatically when the last non-daemon actor finishes */
334 void ActorImpl::daemonize()
338 simix_global->daemons.push_back(this);
342 void ActorImpl::undaemonize()
345 auto& vect = simix_global->daemons;
346 auto it = std::find(vect.begin(), vect.end(), this);
347 xbt_assert(it != vect.end(), "The dying daemon is not a daemon after all. Please report that bug.");
348 /* Don't move the whole content since we don't really care about the order */
350 std::swap(*it, vect.back());
356 s4u::Actor* ActorImpl::restart()
358 xbt_assert(this != simix_global->maestro_, "Restarting maestro is not supported");
360 XBT_DEBUG("Restarting actor %s on %s", get_cname(), host_->get_cname());
362 // retrieve the arguments of the old actor
363 ProcessArg arg = ProcessArg(host_, this);
365 // kill the old actor
366 context::Context::self()->get_actor()->kill(this);
368 // start the new actor
369 ActorImplPtr actor = ActorImpl::create(arg.name, arg.code, arg.data, arg.host, arg.properties.get(), nullptr);
370 *actor->on_exit = std::move(*arg.on_exit);
371 actor->set_kill_time(arg.kill_time);
372 actor->set_auto_restart(arg.auto_restart);
374 return actor->get_ciface();
377 void ActorImpl::suspend()
380 XBT_DEBUG("Actor '%s' is already suspended", get_cname());
386 /* If the suspended actor is waiting on a sync, suspend its synchronization.
387 * Otherwise, it will suspend itself when scheduled, ie, very soon. */
388 if (waiting_synchro_ != nullptr)
389 waiting_synchro_->suspend();
392 void ActorImpl::resume()
394 XBT_IN("actor = %p", this);
396 if (context_->wannadie()) {
397 XBT_VERB("Ignoring request to suspend an actor that is currently dying.");
405 /* resume the activity that was blocking the resumed actor. */
406 if (waiting_synchro_)
407 waiting_synchro_->resume();
408 else // Reschedule the actor if it was forcefully unscheduled in yield()
409 simix_global->actors_to_run.push_back(this);
414 activity::ActivityImplPtr ActorImpl::join(const ActorImpl* actor, double timeout)
416 activity::ActivityImplPtr sleep = this->sleep(timeout);
417 actor->on_exit->emplace_back([sleep](bool) {
418 if (sleep->surf_action_)
419 sleep->surf_action_->finish(resource::Action::State::FINISHED);
424 activity::ActivityImplPtr ActorImpl::sleep(double duration)
426 if (not host_->is_on())
427 throw_exception(std::make_exception_ptr(HostFailureException(
428 XBT_THROW_POINT, std::string("Host ") + host_->get_cname() + " failed, you cannot sleep there.")));
430 auto sleep = new activity::SleepImpl();
431 sleep->set_name("sleep").set_host(host_).set_duration(duration).start();
432 return activity::SleepImplPtr(sleep);
435 void ActorImpl::throw_exception(std::exception_ptr e)
442 /* cancel the blocking synchro if any */
443 if (waiting_synchro_) {
444 waiting_synchro_->cancel();
445 activities_.remove(waiting_synchro_);
446 waiting_synchro_ = nullptr;
450 void ActorImpl::simcall_answer()
452 if (this != simix_global->maestro_) {
453 XBT_DEBUG("Answer simcall %s (%d) issued by %s (%p)", SIMIX_simcall_name(simcall_.call_), (int)simcall_.call_,
455 xbt_assert(simcall_.call_ != simix::Simcall::NONE);
456 simcall_.call_ = simix::Simcall::NONE;
457 xbt_assert(not XBT_LOG_ISENABLED(simix_process, xbt_log_priority_debug) ||
458 std::find(begin(simix_global->actors_to_run), end(simix_global->actors_to_run), this) ==
459 end(simix_global->actors_to_run),
460 "Actor %p should not exist in actors_to_run!", this);
461 simix_global->actors_to_run.push_back(this);
465 void ActorImpl::set_host(s4u::Host* dest)
467 host_->pimpl_->remove_actor(this);
469 dest->pimpl_->add_actor(this);
472 ActorImplPtr ActorImpl::init(const std::string& name, s4u::Host* host) const
474 auto* actor = new ActorImpl(xbt::string(name), host);
475 actor->set_ppid(this->pid_);
477 intrusive_ptr_add_ref(actor);
478 /* The on_creation() signal must be delayed until there, where the pid and everything is set */
479 s4u::Actor::on_creation(*actor->get_ciface());
481 return ActorImplPtr(actor);
484 ActorImpl* ActorImpl::start(const ActorCode& code)
486 xbt_assert(code && host_ != nullptr, "Invalid parameters");
488 if (not host_->is_on()) {
489 XBT_WARN("Cannot launch actor '%s' on failed host '%s'", name_.c_str(), host_->get_cname());
490 intrusive_ptr_release(this);
491 throw HostFailureException(XBT_THROW_POINT, "Cannot start actor on failed host.");
495 XBT_VERB("Create context %s", get_cname());
496 context_.reset(simix_global->context_factory->create_context(ActorCode(code), this));
498 XBT_DEBUG("Start context '%s'", get_cname());
500 /* Add the actor to its host's actor list */
501 host_->pimpl_->add_actor(this);
502 simix_global->process_list[pid_] = this;
504 /* Now insert it in the global actor list and in the actor to run list */
505 XBT_DEBUG("Inserting [%p] %s(%s) in the to_run list", this, get_cname(), host_->get_cname());
506 simix_global->actors_to_run.push_back(this);
511 ActorImplPtr ActorImpl::create(const std::string& name, const ActorCode& code, void* data, s4u::Host* host,
512 const std::unordered_map<std::string, std::string>* properties,
513 const ActorImpl* parent_actor)
515 XBT_DEBUG("Start actor %s@'%s'", name.c_str(), host->get_cname());
518 if (parent_actor != nullptr)
519 actor = parent_actor->init(xbt::string(name), host);
521 actor = self()->init(xbt::string(name), host);
524 actor->set_user_data(data);
527 if (properties != nullptr)
528 actor->set_properties(*properties);
535 void create_maestro(const std::function<void()>& code)
537 /* Create maestro actor and initialize it */
538 auto* maestro = new ActorImpl(xbt::string(""), /*host*/ nullptr);
541 maestro->context_.reset(simix_global->context_factory->create_context(ActorCode(), maestro));
543 maestro->context_.reset(simix_global->context_factory->create_maestro(ActorCode(code), maestro));
546 maestro->simcall_.issuer_ = maestro;
547 simix_global->maestro_ = maestro;
551 } // namespace kernel
552 } // namespace simgrid
554 int SIMIX_process_count() // XBT_ATTRIB_DEPRECATED_v329
556 return simix_global->process_list.size();
559 void* SIMIX_process_self_get_data() // XBT_ATTRIB_DEPRECATED_v329
561 smx_actor_t self = simgrid::kernel::actor::ActorImpl::self();
563 if (self == nullptr) {
566 return self->get_user_data();
569 void SIMIX_process_self_set_data(void* data) // XBT_ATTRIB_DEPRECATED_v329
571 simgrid::kernel::actor::ActorImpl::self()->set_user_data(data);
574 /* needs to be public and without simcall because it is called
575 by exceptions and logging events */
576 const char* SIMIX_process_self_get_name()
578 return SIMIX_is_maestro() ? "maestro" : simgrid::kernel::actor::ActorImpl::self()->get_cname();
581 /** @brief Returns the process from PID. */
582 smx_actor_t SIMIX_process_from_PID(aid_t PID)
584 return simgrid::kernel::actor::ActorImpl::by_PID(PID);
587 void SIMIX_process_on_exit(smx_actor_t actor,
588 const std::function<void(bool /*failed*/)>& fun) // XBT_ATTRIB_DEPRECATED_v329
590 xbt_assert(actor, "current process not found: are you in maestro context ?");
591 actor->on_exit->emplace_back(fun);
594 void simcall_process_set_data(smx_actor_t process, void* data) // XBT_ATTRIB_DEPRECATED_v329
596 simgrid::kernel::actor::simcall([process, data] { process->set_user_data(data); });