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>
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 simgrid::kernel::context::Context* self_context = simgrid::kernel::context::Context::self();
37 return (self_context != nullptr) ? self_context->get_actor() : nullptr;
41 * @brief Returns whether a process has pending asynchronous communications.
42 * @return true if there are asynchronous communications in this process
45 int SIMIX_process_has_pending_comms(smx_actor_t process)
48 return process->comms.size() > 0;
55 static unsigned long maxpid = 0;
61 ActorImpl::ActorImpl(simgrid::xbt::string name, s4u::Host* host) : host_(host), name_(std::move(name)), piface_(this)
64 simcall.issuer_ = this;
67 ActorImpl::~ActorImpl()
69 if (simix_global != nullptr && this != simix_global->maestro_process) {
70 if (context_.get() != nullptr) /* the actor was not start()ed yet. This happens if its host was initially off */
71 context_->iwannadie = false; // don't let the simcall's yield() do a Context::stop(), to avoid infinite loops
72 simgrid::kernel::actor::simcall([this] { simgrid::s4u::Actor::on_destruction(*ciface()); });
73 if (context_.get() != nullptr)
74 context_->iwannadie = true;
78 /* Become an actor in the simulation
80 * Currently this can only be called by the main thread (once) and only work with some thread factories
81 * (currently ThreadContextFactory).
83 * In the future, it might be extended in order to attach other threads created by a third party library.
86 ActorImplPtr ActorImpl::attach(const std::string& name, void* data, s4u::Host* host,
87 const std::unordered_map<std::string, std::string>* properties)
89 // This is mostly a copy/paste from create(), it'd be nice to share some code between those two functions.
91 XBT_DEBUG("Attach process %s on host '%s'", name.c_str(), host->get_cname());
93 if (not host->is_on()) {
94 XBT_WARN("Cannot launch process '%s' on failed host '%s'", name.c_str(), host->get_cname());
95 throw simgrid::HostFailureException(XBT_THROW_POINT, "Cannot attach actor on failed host.");
98 ActorImpl* actor = new ActorImpl(xbt::string(name), host);
100 actor->set_user_data(data);
101 actor->code_ = nullptr;
103 XBT_VERB("Create context %s", actor->get_cname());
104 xbt_assert(simix_global != nullptr, "simix is not initialized, please call MSG_init first");
105 actor->context_.reset(simix_global->context_factory->attach(actor));
108 if (properties != nullptr)
109 actor->set_properties(*properties);
111 /* Add the process to it's host process list */
112 host->pimpl_->process_list_.push_back(*actor);
114 /* Now insert it in the global process list and in the process to run list */
115 simix_global->process_list[actor->get_pid()] = actor;
116 XBT_DEBUG("Inserting [%p] %s(%s) in the to_run list", actor, actor->get_cname(), host->get_cname());
117 simix_global->actors_to_run.push_back(actor);
118 intrusive_ptr_add_ref(actor);
120 auto* context = dynamic_cast<simgrid::kernel::context::AttachContext*>(actor->context_.get());
121 xbt_assert(nullptr != context, "Not a suitable context");
122 context->attach_start();
124 /* The on_creation() signal must be delayed until there, where the pid and everything is set */
125 simgrid::s4u::Actor::on_creation(*actor->ciface());
127 return ActorImplPtr(actor);
129 /** @brief Detach an actor attached with `attach()`
131 * This is called when the current actor has finished its job.
132 * Used in the main thread, it waits for the simulation to finish before returning. When it returns, the other
133 * simulated actors and the maestro are destroyed.
135 void ActorImpl::detach()
137 auto* context = dynamic_cast<context::AttachContext*>(context::Context::self());
138 if (context == nullptr)
139 xbt_die("Not a suitable context");
141 context->get_actor()->cleanup();
142 context->attach_stop();
145 void ActorImpl::cleanup()
149 if (has_to_auto_restart() && not get_host()->is_on()) {
150 XBT_DEBUG("Insert host %s to watched_hosts because it's off and %s needs to restart", get_host()->get_cname(),
152 watched_hosts.insert(get_host()->get_name());
156 // Execute the termination callbacks
157 bool failed = context_->iwannadie;
158 for (auto exit_fun = on_exit->crbegin(); exit_fun != on_exit->crend(); ++exit_fun)
164 /* cancel non-blocking activities */
165 for (auto activity : comms)
166 boost::static_pointer_cast<activity::CommImpl>(activity)->cancel();
169 XBT_DEBUG("%s@%s(%ld) should not run anymore", get_cname(), get_host()->get_cname(), get_pid());
171 if (this == simix_global->maestro_process) /* Do not cleanup maestro */
174 XBT_DEBUG("Cleanup actor %s (%p), waiting synchro %p", get_cname(), this, waiting_synchro.get());
176 /* Unregister from the kill timer if any */
177 if (kill_timer != nullptr) {
178 kill_timer->remove();
179 kill_timer = nullptr;
182 simix_global->mutex.lock();
184 simix_global->process_list.erase(pid_);
185 if (host_ && host_process_list_hook.is_linked())
186 simgrid::xbt::intrusive_erase(host_->pimpl_->process_list_, *this);
187 if (not smx_destroy_list_hook.is_linked()) {
189 xbt_dynar_push_as(simix_global->dead_actors_vector, ActorImpl*, this);
191 simix_global->actors_to_destroy.push_back(*this);
194 simix_global->mutex.unlock();
196 context_->iwannadie = false; // don't let the simcall's yield() do a Context::stop(), to avoid infinite loops
197 simgrid::kernel::actor::simcall([this] { simgrid::s4u::Actor::on_termination(*ciface()); });
198 context_->iwannadie = true;
201 void ActorImpl::exit()
203 context_->iwannadie = true;
205 exception_ = nullptr;
207 /* destroy the blocking synchro if any */
208 if (waiting_synchro != nullptr) {
209 waiting_synchro->cancel();
210 waiting_synchro->state_ = SIMIX_FAILED;
212 activity::ExecImplPtr exec = boost::dynamic_pointer_cast<activity::ExecImpl>(waiting_synchro);
213 activity::CommImplPtr comm = boost::dynamic_pointer_cast<activity::CommImpl>(waiting_synchro);
215 if (exec != nullptr) {
216 exec->clean_action();
217 } else if (comm != nullptr) {
218 comms.remove(waiting_synchro);
219 // Remove first occurrence of &actor->simcall:
220 auto i = boost::range::find(waiting_synchro->simcalls_, &simcall);
221 if (i != waiting_synchro->simcalls_.end())
222 waiting_synchro->simcalls_.remove(&simcall);
224 activity::ActivityImplPtr(waiting_synchro)->finish();
227 waiting_synchro = nullptr;
230 // Forcefully kill the actor if its host is turned off. Not a HostFailureException because you should not survive that
231 this->throw_exception(std::make_exception_ptr(ForcefulKillException(host_->is_on() ? "exited" : "host failed")));
234 void ActorImpl::kill(ActorImpl* actor)
236 if (actor->finished_) {
237 XBT_DEBUG("Ignoring request to kill actor %s@%s that is already dead", actor->get_cname(),
238 actor->host_->get_cname());
242 XBT_DEBUG("Actor '%s'@%s is killing actor '%s'@%s", get_cname(), host_ ? host_->get_cname() : "", actor->get_cname(),
243 actor->host_ ? actor->host_->get_cname() : "");
247 if (std::find(begin(simix_global->actors_to_run), end(simix_global->actors_to_run), actor) ==
248 end(simix_global->actors_to_run) &&
250 XBT_DEBUG("Inserting %s in the to_run list", actor->get_cname());
251 simix_global->actors_to_run.push_back(actor);
255 void ActorImpl::kill_all()
257 for (auto const& kv : simix_global->process_list)
258 if (kv.second != this)
259 this->kill(kv.second);
262 void ActorImpl::set_kill_time(double kill_time)
264 if (kill_time <= SIMIX_get_clock())
266 XBT_DEBUG("Set kill time %f for actor %s@%s", kill_time, get_cname(), host_->get_cname());
267 kill_timer = simix::Timer::set(kill_time, [this] {
269 kill_timer = nullptr;
273 double ActorImpl::get_kill_time()
275 return kill_timer ? kill_timer->get_date() : 0;
278 void ActorImpl::yield()
280 XBT_DEBUG("Yield actor '%s'", get_cname());
282 /* Go into sleep and return control to maestro */
285 /* Ok, maestro returned control to us */
286 XBT_DEBUG("Control returned to me: '%s'", get_cname());
288 if (context_->iwannadie) {
289 XBT_DEBUG("Actor %s@%s is dead", get_cname(), host_->get_cname());
290 // throw simgrid::kernel::context::ForcefulKillException(); Does not seem to properly kill the actor
296 XBT_DEBUG("Hey! I'm suspended.");
298 xbt_assert(exception_ == nullptr, "Gasp! This exception may be lost by subsequent calls.");
303 if (exception_ != nullptr) {
304 XBT_DEBUG("Wait, maestro left me an exception");
305 std::exception_ptr exception = std::move(exception_);
306 exception_ = nullptr;
307 std::rethrow_exception(std::move(exception));
310 if (SMPI_switch_data_segment && not finished_) {
311 SMPI_switch_data_segment(iface());
315 /** This actor will be terminated automatically when the last non-daemon actor finishes */
316 void ActorImpl::daemonize()
320 simix_global->daemons.push_back(this);
324 void ActorImpl::undaemonize()
327 auto& vect = simix_global->daemons;
328 auto it = std::find(vect.begin(), vect.end(), this);
329 xbt_assert(it != vect.end(), "The dying daemon is not a daemon after all. Please report that bug.");
330 /* Don't move the whole content since we don't really care about the order */
332 std::swap(*it, vect.back());
338 s4u::Actor* ActorImpl::restart()
340 xbt_assert(this != simix_global->maestro_process, "Restarting maestro is not supported");
342 XBT_DEBUG("Restarting actor %s on %s", get_cname(), host_->get_cname());
344 // retrieve the arguments of the old actor
345 ProcessArg arg = ProcessArg(host_, this);
347 // kill the old actor
348 context::Context::self()->get_actor()->kill(this);
350 // start the new actor
352 ActorImpl::create(arg.name, std::move(arg.code), arg.data, arg.host, arg.properties.get(), nullptr);
353 *actor->on_exit = std::move(*arg.on_exit);
354 actor->set_kill_time(arg.kill_time);
355 actor->set_auto_restart(arg.auto_restart);
357 return actor->ciface();
360 void ActorImpl::suspend()
363 XBT_DEBUG("Actor '%s' is already suspended", get_cname());
369 /* If the suspended actor is waiting on a sync, suspend its synchronization. */
370 if (waiting_synchro == nullptr) {
371 auto exec = new activity::ExecImpl();
372 exec->set_name("suspend").set_host(host_).set_flops_amount(0.0).start();
373 waiting_synchro = activity::ExecImplPtr(exec);
375 waiting_synchro->simcalls_.push_back(&simcall);
377 waiting_synchro->suspend();
380 void ActorImpl::resume()
382 XBT_IN("actor = %p", this);
384 if (context_->iwannadie) {
385 XBT_VERB("Ignoring request to suspend an actor that is currently dying.");
393 /* resume the synchronization that was blocking the resumed actor. */
395 waiting_synchro->resume();
400 activity::ActivityImplPtr ActorImpl::join(ActorImpl* actor, double timeout)
402 activity::ActivityImplPtr sleep = this->sleep(timeout);
403 SIMIX_process_on_exit(actor, [sleep](bool) {
404 if (sleep->surf_action_)
405 sleep->surf_action_->finish(resource::Action::State::FINISHED);
410 activity::ActivityImplPtr ActorImpl::sleep(double duration)
412 if (not host_->is_on())
413 throw_exception(std::make_exception_ptr(simgrid::HostFailureException(
414 XBT_THROW_POINT, std::string("Host ") + host_->get_cname() + " failed, you cannot sleep there.")));
416 auto sleep = new activity::SleepImpl();
417 sleep->set_name("sleep").set_host(host_).set_duration(duration).start();
418 return activity::SleepImplPtr(sleep);
421 void ActorImpl::throw_exception(std::exception_ptr e)
428 /* cancel the blocking synchro if any */
429 if (waiting_synchro) {
430 waiting_synchro->cancel();
432 activity::CommImplPtr comm = boost::dynamic_pointer_cast<activity::CommImpl>(waiting_synchro);
437 waiting_synchro = nullptr;
441 void ActorImpl::simcall_answer()
443 if (this != simix_global->maestro_process){
444 XBT_DEBUG("Answer simcall %s (%d) issued by %s (%p)", SIMIX_simcall_name(simcall.call_), (int)simcall.call_,
446 simcall.call_ = SIMCALL_NONE;
447 xbt_assert(not XBT_LOG_ISENABLED(simix_process, xbt_log_priority_debug) ||
448 std::find(begin(simix_global->actors_to_run), end(simix_global->actors_to_run), this) ==
449 end(simix_global->actors_to_run),
450 "Actor %p should not exist in actors_to_run!", this);
451 simix_global->actors_to_run.push_back(this);
455 void ActorImpl::set_host(s4u::Host* dest)
457 xbt::intrusive_erase(host_->pimpl_->process_list_, *this);
459 dest->pimpl_->process_list_.push_back(*this);
462 ActorImplPtr ActorImpl::init(const std::string& name, s4u::Host* host)
464 ActorImpl* actor = new ActorImpl(xbt::string(name), host);
465 actor->set_ppid(this->pid_);
467 intrusive_ptr_add_ref(actor);
468 /* The on_creation() signal must be delayed until there, where the pid and everything is set */
469 s4u::Actor::on_creation(*actor->ciface());
471 return ActorImplPtr(actor);
474 ActorImpl* ActorImpl::start(const simix::ActorCode& code)
476 xbt_assert(code && host_ != nullptr, "Invalid parameters");
478 if (not host_->is_on()) {
479 XBT_WARN("Cannot launch actor '%s' on failed host '%s'", name_.c_str(), host_->get_cname());
480 intrusive_ptr_release(this);
481 throw simgrid::HostFailureException(XBT_THROW_POINT, "Cannot start actor on failed host.");
485 XBT_VERB("Create context %s", get_cname());
486 context_.reset(simix_global->context_factory->create_context(simix::ActorCode(code), this));
488 XBT_DEBUG("Start context '%s'", get_cname());
490 /* Add the actor to its host's actor list */
491 host_->pimpl_->process_list_.push_back(*this);
492 simix_global->process_list[pid_] = this;
494 /* Now insert it in the global actor list and in the actor to run list */
495 XBT_DEBUG("Inserting [%p] %s(%s) in the to_run list", this, get_cname(), host_->get_cname());
496 simix_global->actors_to_run.push_back(this);
501 ActorImplPtr ActorImpl::create(const std::string& name, const simix::ActorCode& code, void* data, s4u::Host* host,
502 const std::unordered_map<std::string, std::string>* properties, ActorImpl* parent_actor)
504 XBT_DEBUG("Start actor %s@'%s'", name.c_str(), host->get_cname());
507 if (parent_actor != nullptr)
508 actor = parent_actor->init(xbt::string(name), host);
510 actor = SIMIX_process_self()->init(xbt::string(name), host);
513 actor->set_user_data(data);
516 if (properties != nullptr)
517 actor->set_properties(*properties);
524 void create_maestro(const std::function<void()>& code)
526 /* Create maestro actor and initialize it */
527 ActorImpl* maestro = new ActorImpl(xbt::string(""), /*host*/ nullptr);
530 maestro->context_.reset(simix_global->context_factory->create_context(simix::ActorCode(), maestro));
532 maestro->context_.reset(simix_global->context_factory->create_maestro(simix::ActorCode(code), maestro));
535 maestro->simcall.issuer_ = maestro;
536 simix_global->maestro_process = maestro;
540 } // namespace kernel
541 } // namespace simgrid
543 void SIMIX_process_detach() // deprecated v3.25
545 simgrid::kernel::actor::ActorImpl::detach();
548 smx_actor_t SIMIX_process_attach(const char* name, void* data, const char* hostname,
549 std::unordered_map<std::string, std::string>* properties,
550 smx_actor_t /*parent_process*/) // deprecated 3.25
552 return simgrid::kernel::actor::ActorImpl::attach(name, data, sg_host_by_name(hostname), properties).get();
555 int SIMIX_process_count()
557 return simix_global->process_list.size();
560 void* SIMIX_process_self_get_data() // deprecated
562 smx_actor_t self = SIMIX_process_self();
564 if (self == nullptr) {
567 return self->get_user_data();
570 void SIMIX_process_self_set_data(void* data) // deprecated
572 SIMIX_process_self()->set_user_data(data);
575 /* needs to be public and without simcall because it is called
576 by exceptions and logging events */
577 const char* SIMIX_process_self_get_name()
580 smx_actor_t process = SIMIX_process_self();
581 if (process == nullptr || process == simix_global->maestro_process)
584 return process->get_cname();
588 * @brief Calling this function makes the process to yield.
590 * Only the current process can call this function, giving back the control to maestro.
592 * @param self the current process
595 /** @brief Returns the list of processes to run.
598 const std::vector<smx_actor_t>& simgrid::simix::process_get_runnable()
600 return simix_global->actors_to_run;
603 /** @brief Returns the process from PID. */
604 smx_actor_t SIMIX_process_from_PID(aid_t PID)
606 auto item = simix_global->process_list.find(PID);
607 if (item == simix_global->process_list.end()) {
608 for (auto& a : simix_global->actors_to_destroy)
609 if (a.get_pid() == PID)
611 return nullptr; // Not found, even in the trash
616 void SIMIX_process_on_exit(smx_actor_t actor, int_f_pvoid_pvoid_t fun, void* data)
618 SIMIX_process_on_exit(actor, [fun, data](bool failed) {
619 intptr_t status = failed ? SMX_EXIT_FAILURE : SMX_EXIT_SUCCESS;
620 fun(reinterpret_cast<void*>(status), data);
624 void SIMIX_process_on_exit(smx_actor_t actor, const std::function<void(int, void*)>& fun, void* data)
626 SIMIX_process_on_exit(actor, [fun, data](bool failed) { fun(failed ? SMX_EXIT_FAILURE : SMX_EXIT_SUCCESS, data); });
629 void SIMIX_process_on_exit(smx_actor_t actor, const std::function<void(bool /*failed*/)>& fun)
631 xbt_assert(actor, "current process not found: are you in maestro context ?");
632 actor->on_exit->emplace_back(fun);
635 /** @brief Restart a process, starting it again from the beginning. */
637 * @ingroup simix_process_management
638 * @brief Creates and runs a new SIMIX process.
640 * The structure and the corresponding thread are created and put in the list of ready processes.
642 * @param name a name for the process. It is for user-level information and can be nullptr.
643 * @param code the main function of the process
644 * @param data a pointer to any data one may want to attach to the new object. It is for user-level information and can
646 * It can be retrieved with the method ActorImpl::getUserData().
647 * @param host where the new agent is executed.
648 * @param properties the properties of the process
650 smx_actor_t simcall_process_create(const std::string& name, const simgrid::simix::ActorCode& code, void* data,
651 sg_host_t host, std::unordered_map<std::string, std::string>* properties)
653 smx_actor_t self = SIMIX_process_self();
654 return simgrid::kernel::actor::simcall([&name, &code, data, host, properties, self] {
655 return simgrid::kernel::actor::ActorImpl::create(name, code, data, host, properties, self).get();
659 void simcall_process_set_data(smx_actor_t process, void* data)
661 simgrid::kernel::actor::simcall([process, data] { process->set_user_data(data); });