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;
44 static unsigned long maxpid = 0;
50 ActorImpl::ActorImpl(simgrid::xbt::string name, s4u::Host* host) : host_(host), name_(std::move(name)), piface_(this)
53 simcall.issuer_ = this;
56 ActorImpl::~ActorImpl()
58 if (simix_global != nullptr && this != simix_global->maestro_process) {
59 if (context_.get() != nullptr) /* the actor was not start()ed yet. This happens if its host was initially off */
60 context_->iwannadie = false; // don't let the simcall's yield() do a Context::stop(), to avoid infinite loops
61 simgrid::kernel::actor::simcall([this] { simgrid::s4u::Actor::on_destruction(*ciface()); });
62 if (context_.get() != nullptr)
63 context_->iwannadie = true;
67 /* Become an actor in the simulation
69 * Currently this can only be called by the main thread (once) and only work with some thread factories
70 * (currently ThreadContextFactory).
72 * In the future, it might be extended in order to attach other threads created by a third party library.
75 ActorImplPtr ActorImpl::attach(const std::string& name, void* data, s4u::Host* host,
76 const std::unordered_map<std::string, std::string>* properties)
78 // This is mostly a copy/paste from create(), it'd be nice to share some code between those two functions.
80 XBT_DEBUG("Attach process %s on host '%s'", name.c_str(), host->get_cname());
82 if (not host->is_on()) {
83 XBT_WARN("Cannot launch process '%s' on failed host '%s'", name.c_str(), host->get_cname());
84 throw simgrid::HostFailureException(XBT_THROW_POINT, "Cannot attach actor on failed host.");
87 ActorImpl* actor = new ActorImpl(xbt::string(name), host);
89 actor->set_user_data(data);
90 actor->code_ = nullptr;
92 XBT_VERB("Create context %s", actor->get_cname());
93 xbt_assert(simix_global != nullptr, "simix is not initialized, please call MSG_init first");
94 actor->context_.reset(simix_global->context_factory->attach(actor));
97 if (properties != nullptr)
98 actor->set_properties(*properties);
100 /* Add the process to it's host process list */
101 host->pimpl_->add_actor(actor);
103 /* Now insert it in the global process list and in the process to run list */
104 simix_global->process_list[actor->get_pid()] = actor;
105 XBT_DEBUG("Inserting [%p] %s(%s) in the to_run list", actor, actor->get_cname(), host->get_cname());
106 simix_global->actors_to_run.push_back(actor);
107 intrusive_ptr_add_ref(actor);
109 auto* context = dynamic_cast<simgrid::kernel::context::AttachContext*>(actor->context_.get());
110 xbt_assert(nullptr != context, "Not a suitable context");
111 context->attach_start();
113 /* The on_creation() signal must be delayed until there, where the pid and everything is set */
114 simgrid::s4u::Actor::on_creation(*actor->ciface());
116 return ActorImplPtr(actor);
118 /** @brief Detach an actor attached with `attach()`
120 * This is called when the current actor has finished its job.
121 * Used in the main thread, it waits for the simulation to finish before returning. When it returns, the other
122 * simulated actors and the maestro are destroyed.
124 void ActorImpl::detach()
126 auto* context = dynamic_cast<context::AttachContext*>(context::Context::self());
127 if (context == nullptr)
128 xbt_die("Not a suitable context");
130 context->get_actor()->cleanup();
131 context->attach_stop();
134 void ActorImpl::cleanup()
138 if (has_to_auto_restart() && not get_host()->is_on()) {
139 XBT_DEBUG("Insert host %s to watched_hosts because it's off and %s needs to restart", get_host()->get_cname(),
141 watched_hosts.insert(get_host()->get_name());
145 // Execute the termination callbacks
146 bool failed = context_->iwannadie;
147 for (auto exit_fun = on_exit->crbegin(); exit_fun != on_exit->crend(); ++exit_fun)
153 /* cancel non-blocking activities */
154 for (auto activity : comms)
155 boost::static_pointer_cast<activity::CommImpl>(activity)->cancel();
158 XBT_DEBUG("%s@%s(%ld) should not run anymore", get_cname(), get_host()->get_cname(), get_pid());
160 if (this == simix_global->maestro_process) /* Do not cleanup maestro */
163 XBT_DEBUG("Cleanup actor %s (%p), waiting synchro %p", get_cname(), this, waiting_synchro.get());
165 /* Unregister from the kill timer if any */
166 if (kill_timer != nullptr) {
167 kill_timer->remove();
168 kill_timer = nullptr;
171 simix_global->mutex.lock();
173 simix_global->process_list.erase(pid_);
174 if (host_ && host_actor_list_hook.is_linked())
175 host_->pimpl_->remove_actor(this);
176 if (not smx_destroy_list_hook.is_linked()) {
178 xbt_dynar_push_as(simix_global->dead_actors_vector, ActorImpl*, this);
180 simix_global->actors_to_destroy.push_back(*this);
183 simix_global->mutex.unlock();
185 context_->iwannadie = false; // don't let the simcall's yield() do a Context::stop(), to avoid infinite loops
186 simgrid::kernel::actor::simcall([this] { simgrid::s4u::Actor::on_termination(*ciface()); });
187 context_->iwannadie = true;
190 void ActorImpl::exit()
192 context_->iwannadie = true;
194 exception_ = nullptr;
196 /* destroy the blocking synchro if any */
197 if (waiting_synchro != nullptr) {
198 waiting_synchro->cancel();
199 waiting_synchro->state_ = SIMIX_FAILED;
201 activity::ExecImplPtr exec = boost::dynamic_pointer_cast<activity::ExecImpl>(waiting_synchro);
202 activity::CommImplPtr comm = boost::dynamic_pointer_cast<activity::CommImpl>(waiting_synchro);
204 if (exec != nullptr) {
205 exec->clean_action();
206 } else if (comm != nullptr) {
207 comms.remove(waiting_synchro);
208 // Remove first occurrence of &actor->simcall:
209 auto i = boost::range::find(waiting_synchro->simcalls_, &simcall);
210 if (i != waiting_synchro->simcalls_.end())
211 waiting_synchro->simcalls_.remove(&simcall);
213 activity::ActivityImplPtr(waiting_synchro)->finish();
216 waiting_synchro = nullptr;
219 // Forcefully kill the actor if its host is turned off. Not a HostFailureException because you should not survive that
220 this->throw_exception(std::make_exception_ptr(ForcefulKillException(host_->is_on() ? "exited" : "host failed")));
223 void ActorImpl::kill(ActorImpl* actor)
225 xbt_assert(actor != simix_global->maestro_process, "Killing maestro is a rather bad idea");
226 if (actor->finished_) {
227 XBT_DEBUG("Ignoring request to kill actor %s@%s that is already dead", actor->get_cname(),
228 actor->host_->get_cname());
232 XBT_DEBUG("Actor '%s'@%s is killing actor '%s'@%s", get_cname(), host_ ? host_->get_cname() : "", actor->get_cname(),
233 actor->host_ ? actor->host_->get_cname() : "");
237 if (std::find(begin(simix_global->actors_to_run), end(simix_global->actors_to_run), actor) ==
238 end(simix_global->actors_to_run) &&
240 XBT_DEBUG("Inserting %s in the to_run list", actor->get_cname());
241 simix_global->actors_to_run.push_back(actor);
245 void ActorImpl::kill_all()
247 for (auto const& kv : simix_global->process_list)
248 if (kv.second != this)
249 this->kill(kv.second);
252 void ActorImpl::set_kill_time(double kill_time)
254 if (kill_time <= SIMIX_get_clock())
256 XBT_DEBUG("Set kill time %f for actor %s@%s", kill_time, get_cname(), host_->get_cname());
257 kill_timer = simix::Timer::set(kill_time, [this] {
259 kill_timer = nullptr;
263 double ActorImpl::get_kill_time()
265 return kill_timer ? kill_timer->get_date() : 0;
268 void ActorImpl::yield()
270 XBT_DEBUG("Yield actor '%s'", get_cname());
272 /* Go into sleep and return control to maestro */
275 /* Ok, maestro returned control to us */
276 XBT_DEBUG("Control returned to me: '%s'", get_cname());
278 if (context_->iwannadie) {
279 XBT_DEBUG("Actor %s@%s is dead", get_cname(), host_->get_cname());
280 // throw simgrid::kernel::context::ForcefulKillException(); Does not seem to properly kill the actor
286 XBT_DEBUG("Hey! I'm suspended.");
288 xbt_assert(exception_ == nullptr, "Gasp! This exception may be lost by subsequent calls.");
293 if (exception_ != nullptr) {
294 XBT_DEBUG("Wait, maestro left me an exception");
295 std::exception_ptr exception = std::move(exception_);
296 exception_ = nullptr;
297 std::rethrow_exception(std::move(exception));
300 if (SMPI_switch_data_segment && not finished_) {
301 SMPI_switch_data_segment(iface());
305 /** This actor will be terminated automatically when the last non-daemon actor finishes */
306 void ActorImpl::daemonize()
310 simix_global->daemons.push_back(this);
314 void ActorImpl::undaemonize()
317 auto& vect = simix_global->daemons;
318 auto it = std::find(vect.begin(), vect.end(), this);
319 xbt_assert(it != vect.end(), "The dying daemon is not a daemon after all. Please report that bug.");
320 /* Don't move the whole content since we don't really care about the order */
322 std::swap(*it, vect.back());
328 s4u::Actor* ActorImpl::restart()
330 xbt_assert(this != simix_global->maestro_process, "Restarting maestro is not supported");
332 XBT_DEBUG("Restarting actor %s on %s", get_cname(), host_->get_cname());
334 // retrieve the arguments of the old actor
335 ProcessArg arg = ProcessArg(host_, this);
337 // kill the old actor
338 context::Context::self()->get_actor()->kill(this);
340 // start the new actor
342 ActorImpl::create(arg.name, std::move(arg.code), arg.data, arg.host, arg.properties.get(), nullptr);
343 *actor->on_exit = std::move(*arg.on_exit);
344 actor->set_kill_time(arg.kill_time);
345 actor->set_auto_restart(arg.auto_restart);
347 return actor->ciface();
350 void ActorImpl::suspend()
353 XBT_DEBUG("Actor '%s' is already suspended", get_cname());
359 /* If the suspended actor is waiting on a sync, suspend its synchronization. */
360 if (waiting_synchro == nullptr) {
361 auto exec = new activity::ExecImpl();
362 exec->set_name("suspend").set_host(host_).set_flops_amount(0.0).start();
363 waiting_synchro = activity::ExecImplPtr(exec);
365 waiting_synchro->simcalls_.push_back(&simcall);
367 waiting_synchro->suspend();
370 void ActorImpl::resume()
372 XBT_IN("actor = %p", this);
374 if (context_->iwannadie) {
375 XBT_VERB("Ignoring request to suspend an actor that is currently dying.");
383 /* resume the synchronization that was blocking the resumed actor. */
385 waiting_synchro->resume();
390 activity::ActivityImplPtr ActorImpl::join(ActorImpl* actor, double timeout)
392 activity::ActivityImplPtr sleep = this->sleep(timeout);
393 SIMIX_process_on_exit(actor, [sleep](bool) {
394 if (sleep->surf_action_)
395 sleep->surf_action_->finish(resource::Action::State::FINISHED);
400 activity::ActivityImplPtr ActorImpl::sleep(double duration)
402 if (not host_->is_on())
403 throw_exception(std::make_exception_ptr(simgrid::HostFailureException(
404 XBT_THROW_POINT, std::string("Host ") + host_->get_cname() + " failed, you cannot sleep there.")));
406 auto sleep = new activity::SleepImpl();
407 sleep->set_name("sleep").set_host(host_).set_duration(duration).start();
408 return activity::SleepImplPtr(sleep);
411 void ActorImpl::throw_exception(std::exception_ptr e)
418 /* cancel the blocking synchro if any */
419 if (waiting_synchro) {
420 waiting_synchro->cancel();
422 activity::CommImplPtr comm = boost::dynamic_pointer_cast<activity::CommImpl>(waiting_synchro);
427 waiting_synchro = nullptr;
431 void ActorImpl::simcall_answer()
433 if (this != simix_global->maestro_process){
434 XBT_DEBUG("Answer simcall %s (%d) issued by %s (%p)", SIMIX_simcall_name(simcall.call_), (int)simcall.call_,
436 simcall.call_ = SIMCALL_NONE;
437 xbt_assert(not XBT_LOG_ISENABLED(simix_process, xbt_log_priority_debug) ||
438 std::find(begin(simix_global->actors_to_run), end(simix_global->actors_to_run), this) ==
439 end(simix_global->actors_to_run),
440 "Actor %p should not exist in actors_to_run!", this);
441 simix_global->actors_to_run.push_back(this);
445 void ActorImpl::set_host(s4u::Host* dest)
447 host_->pimpl_->remove_actor(this);
449 dest->pimpl_->add_actor(this);
452 ActorImplPtr ActorImpl::init(const std::string& name, s4u::Host* host)
454 ActorImpl* actor = new ActorImpl(xbt::string(name), host);
455 actor->set_ppid(this->pid_);
457 intrusive_ptr_add_ref(actor);
458 /* The on_creation() signal must be delayed until there, where the pid and everything is set */
459 s4u::Actor::on_creation(*actor->ciface());
461 return ActorImplPtr(actor);
464 ActorImpl* ActorImpl::start(const simix::ActorCode& code)
466 xbt_assert(code && host_ != nullptr, "Invalid parameters");
468 if (not host_->is_on()) {
469 XBT_WARN("Cannot launch actor '%s' on failed host '%s'", name_.c_str(), host_->get_cname());
470 intrusive_ptr_release(this);
471 throw simgrid::HostFailureException(XBT_THROW_POINT, "Cannot start actor on failed host.");
475 XBT_VERB("Create context %s", get_cname());
476 context_.reset(simix_global->context_factory->create_context(simix::ActorCode(code), this));
478 XBT_DEBUG("Start context '%s'", get_cname());
480 /* Add the actor to its host's actor list */
481 host_->pimpl_->add_actor(this);
482 simix_global->process_list[pid_] = this;
484 /* Now insert it in the global actor list and in the actor to run list */
485 XBT_DEBUG("Inserting [%p] %s(%s) in the to_run list", this, get_cname(), host_->get_cname());
486 simix_global->actors_to_run.push_back(this);
491 ActorImplPtr ActorImpl::create(const std::string& name, const simix::ActorCode& code, void* data, s4u::Host* host,
492 const std::unordered_map<std::string, std::string>* properties, ActorImpl* parent_actor)
494 XBT_DEBUG("Start actor %s@'%s'", name.c_str(), host->get_cname());
497 if (parent_actor != nullptr)
498 actor = parent_actor->init(xbt::string(name), host);
500 actor = SIMIX_process_self()->init(xbt::string(name), host);
503 actor->set_user_data(data);
506 if (properties != nullptr)
507 actor->set_properties(*properties);
514 void create_maestro(const std::function<void()>& code)
516 /* Create maestro actor and initialize it */
517 ActorImpl* maestro = new ActorImpl(xbt::string(""), /*host*/ nullptr);
520 maestro->context_.reset(simix_global->context_factory->create_context(simix::ActorCode(), maestro));
522 maestro->context_.reset(simix_global->context_factory->create_maestro(simix::ActorCode(code), maestro));
525 maestro->simcall.issuer_ = maestro;
526 simix_global->maestro_process = maestro;
530 } // namespace kernel
531 } // namespace simgrid
533 int SIMIX_process_count()
535 return simix_global->process_list.size();
538 void* SIMIX_process_self_get_data()
540 smx_actor_t self = SIMIX_process_self();
542 if (self == nullptr) {
545 return self->get_user_data();
548 void SIMIX_process_self_set_data(void* data)
550 SIMIX_process_self()->set_user_data(data);
553 /* needs to be public and without simcall because it is called
554 by exceptions and logging events */
555 const char* SIMIX_process_self_get_name()
557 return SIMIX_is_maestro() ? "maestro" : SIMIX_process_self()->get_cname();
561 * @brief Calling this function makes the process to yield.
563 * Only the current process can call this function, giving back the control to maestro.
565 * @param self the current process
568 /** @brief Returns the process from PID. */
569 smx_actor_t SIMIX_process_from_PID(aid_t PID)
571 auto item = simix_global->process_list.find(PID);
572 if (item == simix_global->process_list.end()) {
573 for (auto& a : simix_global->actors_to_destroy)
574 if (a.get_pid() == PID)
576 return nullptr; // Not found, even in the trash
581 void SIMIX_process_on_exit(smx_actor_t actor, const std::function<void(bool /*failed*/)>& fun)
583 xbt_assert(actor, "current process not found: are you in maestro context ?");
584 actor->on_exit->emplace_back(fun);
587 /** @brief Restart a process, starting it again from the beginning. */
589 * @ingroup simix_process_management
590 * @brief Creates and runs a new SIMIX process.
592 * The structure and the corresponding thread are created and put in the list of ready processes.
594 * @param name a name for the process. It is for user-level information and can be nullptr.
595 * @param code the main function of the process
596 * @param data a pointer to any data one may want to attach to the new object. It is for user-level information and can
598 * It can be retrieved with the method ActorImpl::getUserData().
599 * @param host where the new agent is executed.
600 * @param properties the properties of the process
602 smx_actor_t simcall_process_create(const std::string& name, const simgrid::simix::ActorCode& code, void* data,
603 sg_host_t host, std::unordered_map<std::string, std::string>* properties)
605 smx_actor_t self = SIMIX_process_self();
606 return simgrid::kernel::actor::simcall([&name, &code, data, host, properties, self] {
607 return simgrid::kernel::actor::ActorImpl::create(name, code, data, host, properties, self).get();
611 void simcall_process_set_data(smx_actor_t process, void* data)
613 simgrid::kernel::actor::simcall([process, data] { process->set_user_data(data); });