1 /* Copyright (c) 2016-2022. 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. */
6 #include <simgrid/Exception.hpp>
7 #include <simgrid/kernel/Timer.hpp>
8 #include <simgrid/kernel/routing/NetPoint.hpp>
9 #include <simgrid/kernel/routing/NetZoneImpl.hpp>
10 #include <simgrid/s4u/Host.hpp>
11 #include <simgrid/sg_config.hpp>
13 #define SIMIX_H_NO_DEPRECATED_WARNING // avoid deprecation warning on include (remove with XBT_ATTRIB_DEPRECATED_v332)
14 #include <simgrid/simix.h>
17 #include "src/kernel/EngineImpl.hpp"
18 #include "src/kernel/resource/StandardLinkImpl.hpp"
19 #include "src/kernel/resource/profile/Profile.hpp"
20 #include "src/mc/mc_record.hpp"
21 #include "src/mc/mc_replay.hpp"
22 #include "src/smpi/include/smpi_actor.hpp"
23 #include "src/surf/xml/platf.hpp"
24 #include "xbt/xbt_modinter.h" /* whether initialization was already done */
26 #include <boost/algorithm/string/predicate.hpp>
32 #include "src/mc/remote/AppSide.hpp"
37 XBT_LOG_NEW_DEFAULT_CATEGORY(ker_engine, "Logging specific to Engine (kernel)");
41 EngineImpl* EngineImpl::instance_ = nullptr; /* That singleton is awful too. */
43 config::Flag<double> cfg_breakpoint{"debug/breakpoint",
44 "When non-negative, raise a SIGTRAP after given (simulated) time", -1.0};
45 config::Flag<bool> cfg_verbose_exit{"debug/verbose-exit", "Display the actor status at exit", true};
47 xbt_dynar_t get_actors_addr()
50 return EngineImpl::get_instance()->get_actors_vector();
52 xbt_die("This function is intended to be used when compiling with MC");
56 constexpr std::initializer_list<std::pair<const char*, context::ContextFactoryInitializer>> context_factories = {
58 {"raw", &context::raw_factory},
60 #if HAVE_UCONTEXT_CONTEXTS
61 {"ucontext", &context::sysv_factory},
63 #if HAVE_BOOST_CONTEXTS
64 {"boost", &context::boost_factory},
66 {"thread", &context::thread_factory},
69 static_assert(context_factories.size() > 0, "No context factories are enabled for this build");
71 // Create the list of possible contexts:
72 static inline std::string contexts_list()
76 for (auto const& factory : context_factories) {
77 res += sep + factory.first;
83 static config::Flag<std::string> context_factory_name("contexts/factory",
84 (std::string("Possible values: ") + contexts_list()).c_str(),
85 context_factories.begin()->first);
88 } // namespace simgrid
90 XBT_ATTRIB_NORETURN static void inthandler(int)
92 if (simgrid::kernel::cfg_verbose_exit) {
93 XBT_INFO("CTRL-C pressed. The current status will be displayed before exit (disable that behavior with option "
94 "'debug/verbose-exit').");
95 simgrid::kernel::EngineImpl::get_instance()->display_all_actor_status();
97 XBT_INFO("CTRL-C pressed, exiting. Hiding the current process status since 'debug/verbose-exit' is set to false.");
103 static void segvhandler(int signum, siginfo_t* siginfo, void* /*context*/)
105 if ((siginfo->si_signo == SIGSEGV && siginfo->si_code == SEGV_ACCERR) || siginfo->si_signo == SIGBUS) {
107 "Access violation or Bus error detected.\n"
108 "This probably comes from a programming error in your code, or from a stack\n"
109 "overflow. If you are certain of your code, try increasing the stack size\n"
110 " --cfg=contexts/stack-size:XXX (current size is %u KiB).\n"
112 "If it does not help, this may have one of the following causes:\n"
113 "a bug in SimGrid, a bug in the OS or a bug in a third-party libraries.\n"
114 "Failing hardware can sometimes generate such errors too.\n"
116 "If you think you've found a bug in SimGrid, please report it along with a\n"
117 "Minimal Working Example (MWE) reproducing your problem and a full backtrace\n"
118 "of the fault captured with gdb or valgrind.\n",
119 simgrid::kernel::context::stack_size / 1024);
120 } else if (siginfo->si_signo == SIGSEGV) {
121 fprintf(stderr, "Segmentation fault.\n");
123 if (smpi_enabled() && smpi_cfg_privatization() == SmpiPrivStrategies::NONE) {
124 #if HAVE_PRIVATIZATION
125 fprintf(stderr, "Try to enable SMPI variable privatization with --cfg=smpi/privatization:yes.\n");
127 fprintf(stderr, "Sadly, your system does not support --cfg=smpi/privatization:yes (yet).\n");
128 #endif /* HAVE_PRIVATIZATION */
130 #endif /* HAVE_SMPI */
136 * Install signal handler for SIGSEGV. Check that nobody has already installed
137 * its own handler. For example, the Java VM does this.
139 static void install_segvhandler()
143 if (simgrid::kernel::context::Context::install_sigsegv_stack(&old_stack, true) == -1) {
144 XBT_WARN("Failed to register alternate signal stack: %s", strerror(errno));
147 if (not(old_stack.ss_flags & SS_DISABLE)) {
148 XBT_DEBUG("An alternate stack was already installed (sp=%p, size=%zu, flags=%x). Restore it.", old_stack.ss_sp,
149 old_stack.ss_size, (unsigned)old_stack.ss_flags);
150 sigaltstack(&old_stack, nullptr);
153 struct sigaction action;
154 struct sigaction old_action;
155 action.sa_sigaction = &segvhandler;
156 action.sa_flags = SA_ONSTACK | SA_RESETHAND | SA_SIGINFO;
157 sigemptyset(&action.sa_mask);
159 /* Linux tend to raise only SIGSEGV where other systems also raise SIGBUS on severe error */
160 for (int sig : {SIGSEGV, SIGBUS}) {
161 if (sigaction(sig, &action, &old_action) == -1) {
162 XBT_WARN("Failed to register signal handler for signal %d: %s", sig, strerror(errno));
165 if ((old_action.sa_flags & SA_SIGINFO) || old_action.sa_handler != SIG_DFL) {
166 XBT_DEBUG("A signal handler was already installed for signal %d (%p). Restore it.", sig,
167 (old_action.sa_flags & SA_SIGINFO) ? (void*)old_action.sa_sigaction : (void*)old_action.sa_handler);
168 sigaction(sig, &old_action, nullptr);
178 EngineImpl::~EngineImpl()
180 /* Since hosts_ is a std::map, the hosts are destroyed in the lexicographic order, which ensures that the output is
183 while (not hosts_.empty())
184 hosts_.begin()->second->destroy();
186 /* Also delete the other data */
187 delete netzone_root_;
188 for (auto const& kv : netpoints_)
191 while (not links_.empty())
192 links_.begin()->second->destroy();
194 for (auto const& kv : mailboxes_)
197 /* Free the remaining data structures */
199 xbt_dynar_free(&actors_vector_);
201 /* clear models before freeing handle, network models can use external callback defined in the handle */
202 models_prio_.clear();
205 void EngineImpl::initialize(int* argc, char** argv)
207 xbt_assert(EngineImpl::instance_ == nullptr,
208 "It is currently forbidden to create more than one instance of kernel::EngineImpl");
209 EngineImpl::instance_ = this;
211 // The communication initialization is done ASAP, as we need to get some init parameters from the MC for different
212 // layers. But simix_global needs to be created, as we send the address of some of its fields to the MC that wants to
213 // read them directly.
214 simgrid::mc::AppSide::initialize();
217 if (xbt_initialized == 0) {
218 xbt_init(argc, argv);
220 sg_config_init(argc, argv);
223 instance_->context_mod_init();
225 /* Prepare to display some more info when dying on Ctrl-C pressing */
226 std::signal(SIGINT, inthandler);
229 install_segvhandler();
232 /* register a function to be called by SURF after the environment creation */
234 s4u::Engine::on_platform_created_cb([this]() { this->presolve(); });
236 if (config::get_value<bool>("debug/clean-atexit"))
240 void EngineImpl::context_mod_init() const
242 xbt_assert(not instance_->has_context_factory());
244 #if HAVE_SMPI && defined(__NetBSD__)
245 smpi_init_options_internal(false);
246 std::string priv = config::get_value<std::string>("smpi/privatization");
247 if (context_factory_name == "thread" && (priv == "dlopen" || priv == "yes" || priv == "default" || priv == "1")) {
248 XBT_WARN("dlopen+thread broken on Apple and BSD. Switching to raw contexts.");
249 context_factory_name = "raw";
253 #if HAVE_SMPI && defined(__FreeBSD__)
254 smpi_init_options_internal(false);
255 if (context_factory_name == "thread" && config::get_value<std::string>("smpi/privatization") != "no") {
256 XBT_WARN("mmap broken on FreeBSD, but dlopen+thread broken too. Switching to dlopen+raw contexts.");
257 context_factory_name = "raw";
261 /* select the context factory to use to create the contexts */
262 if (context::factory_initializer != nullptr) { // Give Java a chance to hijack the factory mechanism
263 instance_->set_context_factory(context::factory_initializer());
266 /* use the factory specified by --cfg=contexts/factory:value */
267 for (auto const& factory : context_factories)
268 if (context_factory_name == factory.first) {
269 instance_->set_context_factory(factory.second());
273 if (not instance_->has_context_factory()) {
274 XBT_ERROR("Invalid context factory specified. Valid factories on this machine:");
275 #if HAVE_RAW_CONTEXTS
276 XBT_ERROR(" raw: high performance context factory implemented specifically for SimGrid");
278 XBT_ERROR(" (raw contexts were disabled at compilation time on this machine -- check configure logs for details)");
280 #if HAVE_UCONTEXT_CONTEXTS
281 XBT_ERROR(" ucontext: classical system V contexts (implemented with makecontext, swapcontext and friends)");
283 XBT_ERROR(" (ucontext was disabled at compilation time on this machine -- check configure logs for details)");
285 #if HAVE_BOOST_CONTEXTS
286 XBT_ERROR(" boost: this uses the boost libraries context implementation");
288 XBT_ERROR(" (boost was disabled at compilation time on this machine -- check configure logs for details. Did you "
289 "install the libboost-context-dev package?)");
291 XBT_ERROR(" thread: slow portability layer using pthreads as provided by gcc");
292 xbt_die("Please use a valid factory.");
296 void EngineImpl::shutdown()
298 if (EngineImpl::instance_ == nullptr)
300 XBT_DEBUG("EngineImpl::shutdown() called. Simulation's over.");
302 if (not instance_->actor_list_.empty()) {
303 if (smpi_process()->initialized()) {
304 xbt_die("Process exited without calling MPI_Finalize - Killing simulation");
306 XBT_WARN("Process called exit when leaving - Skipping cleanups");
312 if (instance_->has_actors_to_run() && simgrid_get_clock() <= 0.0) {
314 XBT_CRITICAL("The time is still 0, and you still have processes ready to run.");
315 XBT_CRITICAL("It seems that you forgot to run the simulation that you setup.");
316 xbt_die("Bailing out to avoid that stop-before-start madness. Please fix your code.");
319 /* Kill all actors (but maestro) */
320 instance_->maestro_->kill_all();
321 instance_->run_all_actors();
322 instance_->empty_trash();
324 /* Let's free maestro now */
325 delete instance_->maestro_;
326 instance_->maestro_ = nullptr;
328 /* Finish context module and SURF */
329 instance_->destroy_context_factory();
331 while (not timer::kernel_timers().empty()) {
332 delete timer::kernel_timers().top().second;
333 timer::kernel_timers().pop();
343 void EngineImpl::seal_platform() const
345 /* sealing resources before run: links */
346 for (auto const& kv : links_)
347 kv.second->get_iface()->seal();
348 /* seal netzone root, recursively seal children netzones, hosts and disks */
349 netzone_root_->seal();
352 void EngineImpl::load_platform(const std::string& platf)
354 double start = xbt_os_time();
355 if (boost::algorithm::ends_with(platf, ".so") or boost::algorithm::ends_with(platf, ".dylib")) {
357 xbt_die("loading platform through shared library isn't supported on windows");
359 void* handle = dlopen(platf.c_str(), RTLD_LAZY);
360 xbt_assert(handle, "Impossible to open platform file: %s", platf.c_str());
361 platf_handle_ = std::unique_ptr<void, std::function<int(void*)>>(handle, dlclose);
362 using load_fct_t = void (*)(const simgrid::s4u::Engine&);
363 auto callable = (load_fct_t)dlsym(platf_handle_.get(), "load_platform");
364 const char* dlsym_error = dlerror();
365 xbt_assert(not dlsym_error, "Error: %s", dlsym_error);
366 callable(*simgrid::s4u::Engine::get_instance());
369 parse_platform_file(platf);
372 double end = xbt_os_time();
373 XBT_DEBUG("PARSE TIME: %g", (end - start));
376 void EngineImpl::load_deployment(const std::string& file) const
381 surf_parse_open(file);
386 void EngineImpl::register_function(const std::string& name, const actor::ActorCodeFactory& code)
388 registered_functions[name] = code;
390 void EngineImpl::register_default(const actor::ActorCodeFactory& code)
392 default_function = code;
395 void EngineImpl::add_model(std::shared_ptr<resource::Model> model, const std::vector<resource::Model*>& dependencies)
397 auto model_name = model->get_name();
398 xbt_assert(models_prio_.find(model_name) == models_prio_.end(),
399 "Model %s already exists, use model.set_name() to change its name", model_name.c_str());
401 for (const auto dep : dependencies) {
402 xbt_assert(models_prio_.find(dep->get_name()) != models_prio_.end(),
403 "Model %s doesn't exists. Impossible to use it as dependency.", dep->get_name().c_str());
405 models_.push_back(model.get());
406 models_prio_[model_name] = std::move(model);
409 void EngineImpl::add_split_duplex_link(const std::string& name, std::unique_ptr<resource::SplitDuplexLinkImpl> link)
411 split_duplex_links_[name] = std::move(link);
414 /** Wake up all actors waiting for a Surf action to finish */
415 void EngineImpl::wake_all_waiting_actors() const
417 for (auto const& model : models_) {
418 XBT_DEBUG("Handling the failed actions (if any)");
419 while (auto* action = model->extract_failed_action()) {
420 XBT_DEBUG(" Handling Action %p", action);
421 if (action->get_activity() != nullptr) {
422 // If nobody told the interface that the activity has failed, that's because no actor waits on it (maestro
423 // started it). SimDAG I see you!
424 if (action->get_activity()->get_actor() == maestro_)
425 action->get_activity()->get_iface()->complete(s4u::Activity::State::FAILED);
427 activity::ActivityImplPtr(action->get_activity())->post();
430 XBT_DEBUG("Handling the terminated actions (if any)");
431 while (auto* action = model->extract_done_action()) {
432 XBT_DEBUG(" Handling Action %p", action);
433 if (action->get_activity() == nullptr)
434 XBT_DEBUG("probably vcpu's action %p, skip", action);
436 // If nobody told the interface that the activity is finished, that's because no actor waits on it (maestro
437 // started it). SimDAG I see you!
438 action->get_activity()->set_finish_time(action->get_finish_time());
440 if (action->get_activity()->get_actor() == maestro_)
441 action->get_activity()->get_iface()->complete(s4u::Activity::State::FINISHED);
443 activity::ActivityImplPtr(action->get_activity())->post();
449 * @brief Executes the actors in actors_to_run.
451 * The actors in actors_to_run are run (in parallel if possible). On exit, actors_to_run is empty, and actors_that_ran
452 * contains the list of actors that just ran. The two lists are swapped so, be careful when using them before and after
453 * a call to this function.
455 void EngineImpl::run_all_actors()
457 instance_->get_context_factory()->run_all();
459 actors_to_run_.swap(actors_that_ran_);
460 actors_to_run_.clear();
463 actor::ActorImpl* EngineImpl::get_actor_by_pid(aid_t pid)
465 auto item = actor_list_.find(pid);
466 if (item != actor_list_.end())
470 for (auto& a : actors_to_destroy_)
471 if (a.get_pid() == pid)
473 return nullptr; // Not found, even in the trash
476 /** Execute all the tasks that are queued, e.g. `.then()` callbacks of futures. */
477 bool EngineImpl::execute_tasks()
482 std::vector<xbt::Task<void()>> tasksTemp;
484 // We don't want the callbacks to modify the vector we are iterating over:
485 tasks.swap(tasksTemp);
487 // Execute all the queued tasks:
488 for (auto& task : tasksTemp)
492 } while (not tasks.empty());
497 void EngineImpl::remove_daemon(actor::ActorImpl* actor)
499 auto it = daemons_.find(actor);
500 xbt_assert(it != daemons_.end(), "The dying daemon is not a daemon after all. Please report that bug.");
504 void EngineImpl::add_actor_to_run_list_no_check(actor::ActorImpl* actor)
506 XBT_DEBUG("Inserting [%p] %s(%s) in the to_run list", actor, actor->get_cname(), actor->get_host()->get_cname());
507 actors_to_run_.push_back(actor);
510 void EngineImpl::add_actor_to_run_list(actor::ActorImpl* actor)
512 if (std::find(begin(actors_to_run_), end(actors_to_run_), actor) != end(actors_to_run_)) {
513 XBT_DEBUG("Actor %s is already in the to_run list", actor->get_cname());
515 XBT_DEBUG("Inserting [%p] %s(%s) in the to_run list", actor, actor->get_cname(), actor->get_host()->get_cname());
516 actors_to_run_.push_back(actor);
519 void EngineImpl::empty_trash()
521 while (not actors_to_destroy_.empty()) {
522 actor::ActorImpl* actor = &actors_to_destroy_.front();
523 actors_to_destroy_.pop_front();
524 XBT_DEBUG("Getting rid of %s (refcount: %d)", actor->get_cname(), actor->get_refcount());
525 intrusive_ptr_release(actor);
529 void EngineImpl::display_all_actor_status() const
531 XBT_INFO("%zu actors are still running, waiting for something.", actor_list_.size());
532 /* List the actors and their state */
533 XBT_INFO("Legend of the following listing: \"Actor <pid> (<name>@<host>): <status>\"");
534 for (auto const& kv : actor_list_) {
535 const actor::ActorImpl* actor = kv.second;
537 if (actor->waiting_synchro_) {
538 const char* synchro_description = "unknown";
540 if (boost::dynamic_pointer_cast<kernel::activity::ExecImpl>(actor->waiting_synchro_) != nullptr)
541 synchro_description = "execution";
543 if (boost::dynamic_pointer_cast<kernel::activity::CommImpl>(actor->waiting_synchro_) != nullptr)
544 synchro_description = "communication";
546 if (boost::dynamic_pointer_cast<kernel::activity::SleepImpl>(actor->waiting_synchro_) != nullptr)
547 synchro_description = "sleeping";
549 if (boost::dynamic_pointer_cast<kernel::activity::SynchroImpl>(actor->waiting_synchro_) != nullptr)
550 synchro_description = "synchronization";
552 if (boost::dynamic_pointer_cast<kernel::activity::IoImpl>(actor->waiting_synchro_) != nullptr)
553 synchro_description = "I/O";
555 XBT_INFO("Actor %ld (%s@%s): waiting for %s activity %#zx (%s) in state %s to finish", actor->get_pid(),
556 actor->get_cname(), actor->get_host()->get_cname(), synchro_description,
557 (xbt_log_no_loc ? (size_t)0xDEADBEEF : (size_t)actor->waiting_synchro_.get()),
558 actor->waiting_synchro_->get_cname(), actor->waiting_synchro_->get_state_str());
560 XBT_INFO("Actor %ld (%s@%s) simcall %s", actor->get_pid(), actor->get_cname(), actor->get_host()->get_cname(),
561 actor->simcall_.get_cname());
566 void EngineImpl::presolve() const
568 XBT_DEBUG("Consume all trace events occurring before the starting time.");
569 double next_event_date;
570 while ((next_event_date = profile::future_evt_set.next_date()) != -1.0) {
571 if (next_event_date > NOW)
575 resource::Resource* resource = nullptr;
576 while (auto* event = profile::future_evt_set.pop_leq(next_event_date, &value, &resource)) {
578 resource->apply_event(event, value);
582 XBT_DEBUG("Set every models in the right state by updating them to 0.");
583 for (auto const& model : models_)
584 model->update_actions_state(NOW, 0.0);
587 double EngineImpl::solve(double max_date) const
589 double time_delta = -1.0; /* duration */
591 resource::Resource* resource = nullptr;
593 if (max_date != -1.0) {
594 xbt_assert(max_date >= NOW, "You asked to simulate up to %f, but that's in the past already", max_date);
596 time_delta = max_date - NOW;
599 XBT_DEBUG("Looking for next event in all models");
600 for (auto model : models_) {
601 if (not model->next_occurring_event_is_idempotent()) {
604 double next_event = model->next_occurring_event(NOW);
605 if ((time_delta < 0.0 || next_event < time_delta) && next_event >= 0.0) {
606 time_delta = next_event;
610 XBT_DEBUG("Min for resources (remember that NS3 don't update that value): %f", time_delta);
612 XBT_DEBUG("Looking for next trace event");
614 while (true) { // Handle next occurring events until none remains
615 double next_event_date = profile::future_evt_set.next_date();
616 XBT_DEBUG("Next TRACE event: %f", next_event_date);
618 for (auto model : models_) {
619 /* Skip all idempotent models, they were already treated above
620 * NS3 is the one to handled here */
621 if (model->next_occurring_event_is_idempotent())
624 if (next_event_date != -1.0) {
625 time_delta = std::min(next_event_date - NOW, time_delta);
627 time_delta = std::max(next_event_date - NOW, time_delta); // Get the positive component
630 XBT_DEBUG("Run the NS3 network at most %fs", time_delta);
631 // run until min or next flow
632 double model_next_action_end = model->next_occurring_event(time_delta);
634 XBT_DEBUG("Min for network : %f", model_next_action_end);
635 if (model_next_action_end >= 0.0)
636 time_delta = model_next_action_end;
639 if (next_event_date < 0.0 || (next_event_date > NOW + time_delta)) {
640 // next event may have already occurred or will after the next resource change, then bail out
641 XBT_DEBUG("no next usable TRACE event. Stop searching for it");
645 XBT_DEBUG("Updating models (min = %g, NOW = %g, next_event_date = %g)", time_delta, NOW, next_event_date);
647 while (auto* event = profile::future_evt_set.pop_leq(next_event_date, &value, &resource)) {
648 if (resource->is_used() || (watched_hosts().find(resource->get_cname()) != watched_hosts().end())) {
649 time_delta = next_event_date - NOW;
650 XBT_DEBUG("This event invalidates the next_occurring_event() computation of models. Next event set to %f",
653 // FIXME: I'm too lame to update NOW live, so I change it and restore it so that the real update with surf_min
655 double round_start = NOW;
656 NOW = next_event_date;
657 /* update state of the corresponding resource to the new value. Does not touch lmm.
658 It will be modified if needed when updating actions */
659 XBT_DEBUG("Calling update_resource_state for resource %s", resource->get_cname());
660 resource->apply_event(event, value);
665 /* FIXME: Moved this test to here to avoid stopping simulation if there are actions running on cpus and all cpus are
666 * with availability = 0. This may cause an infinite loop if one cpu has a trace with periodicity = 0 and the other a
667 * trace with periodicity > 0.
668 * The options are: all traces with same periodicity(0 or >0) or we need to change the way how the events are managed
670 if (time_delta < 0) {
671 XBT_DEBUG("No next event at all. Bail out now.");
675 XBT_DEBUG("Duration set to %f", time_delta);
677 // Bump the time: jump into the future
678 NOW = NOW + time_delta;
680 // Inform the models of the date change
681 for (auto const& model : models_)
682 model->update_actions_state(NOW, time_delta);
684 s4u::Engine::on_time_advance(time_delta);
689 void EngineImpl::run(double max_date)
693 if (MC_record_replay_is_active()) {
694 mc::RecordTrace::replay(MC_record_path());
699 double elapsed_time = -1;
700 const std::set<s4u::Activity*>* vetoed_activities = s4u::Activity::get_vetoed_activities();
703 XBT_DEBUG("New Schedule Round; size(queue)=%zu", actors_to_run_.size());
705 if (cfg_breakpoint >= 0.0 && simgrid_get_clock() >= cfg_breakpoint) {
706 XBT_DEBUG("Breakpoint reached (%g)", cfg_breakpoint.get());
707 cfg_breakpoint = -1.0;
717 while (not actors_to_run_.empty()) {
718 XBT_DEBUG("New Sub-Schedule Round; size(queue)=%zu", actors_to_run_.size());
720 /* Run all actors that are ready to run, possibly in parallel */
723 /* answer sequentially and in a fixed arbitrary order all the simcalls that were issued during that sub-round */
725 /* WARNING, the order *must* be fixed or you'll jeopardize the simulation reproducibility (see RR-7653) */
727 /* Here, the order is ok because:
729 * Only maestro adds stuff to the actors_to_run array, so the execution order of user contexts do not impact its order.
731 * In addition, actors remain sorted through an arbitrary but fixed order in all cases:
733 * - If there is no killing during the simulation, actors remain sorted according by their PID.
734 * - Killer actors are moved to the end of the scheduling round (to let victims finish their simcall before dying), but
735 * (1) this decision of killing is reproducible because the simulation was reproducible until then
736 * (2) this reordering introduces no reproducibility hazard in the subsequent simulation.
737 * Even the order change induced by the actor killing is perfectly reproducible.
739 * So the array order is implicit and somewhat complex, but fixed and reproducible (science works, http://xkcd.com/54/).
741 * We could manually sort the actors_that_ran array so that simcalls are handled in an easy to predict order
742 * (e.g. "according to the PID of issuer"), but it's not mandatory for the simulation soundness and reproducibility,
743 * and would thus be a pure waste of time.
746 for (auto const& actor : actors_that_ran_) {
747 if (actor->simcall_.call_ != simix::Simcall::Type::NONE) {
748 actor->simcall_handle(0);
754 wake_all_waiting_actors();
755 } while (execute_tasks());
757 /* If only daemon actors remain, cancel their actions, mark them to die and reschedule them */
758 if (actor_list_.size() == daemons_.size())
759 for (auto const& dmon : daemons_) {
760 XBT_DEBUG("Kill %s", dmon->get_cname());
761 maestro_->kill(dmon);
765 // Compute the max_date of the next solve.
766 // It's either when a timer occurs, or when user-specified deadline is reached, or -1 if none is given
767 double next_time = timer::Timer::next();
768 if (next_time < 0 && max_date > -1) {
769 next_time = max_date;
770 } else if (next_time > -1 && max_date > -1) { // either both <0, or both >0
771 next_time = std::min(next_time, max_date);
774 XBT_DEBUG("Calling solve(%g) %g", next_time, NOW);
775 elapsed_time = solve(next_time);
776 XBT_DEBUG("Moving time ahead. NOW=%g; elapsed: %g", NOW, elapsed_time);
778 /* Notify all the hosts that have failed */
779 /* FIXME: iterate through the list of failed host and mark each of them */
780 /* as failed. On each host, signal all the running actors with host_fail */
782 // Execute timers and tasks until there isn't anything to be done:
785 again = timer::Timer::execute_all();
788 wake_all_waiting_actors();
791 /* Clean actors to destroy */
794 XBT_DEBUG("### elapsed time %f, #actors %zu, #to_run %zu, #vetoed %d", elapsed_time, actor_list_.size(),
795 actors_to_run_.size(), (vetoed_activities == nullptr ? -1 : static_cast<int>(vetoed_activities->size())));
797 if (elapsed_time < 0. && actors_to_run_.empty() && not actor_list_.empty()) {
798 if (actor_list_.size() <= daemons_.size()) {
799 XBT_CRITICAL("Oops! Daemon actors cannot do any blocking activity (communications, synchronization, etc) "
800 "once the simulation is over. Please fix your on_exit() functions.");
802 XBT_CRITICAL("Oops! Deadlock or code not perfectly clean.");
804 display_all_actor_status();
805 simgrid::s4u::Engine::on_deadlock();
806 for (auto const& kv : actor_list_) {
807 XBT_DEBUG("Kill %s", kv.second->get_cname());
808 maestro_->kill(kv.second);
811 } while ((vetoed_activities == nullptr || vetoed_activities->empty()) &&
812 ((elapsed_time > -1.0 && not double_equals(max_date, NOW, 0.00001)) || has_actors_to_run()));
814 if (not actor_list_.empty() && max_date < 0 && not(vetoed_activities == nullptr || vetoed_activities->empty()))
817 simgrid::s4u::Engine::on_simulation_end();
820 double EngineImpl::get_clock()
824 } // namespace kernel
825 } // namespace simgrid
827 void SIMIX_run() // XBT_ATTRIB_DEPRECATED_v332
829 simgrid::kernel::EngineImpl::get_instance()->run(-1);