X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/91db6ba6a3466dc9834d1fca0bea98efa2ff8584..f8e7b4cec7149a685b0a4a58ba56088613124377:/src/kernel/EngineImpl.cpp diff --git a/src/kernel/EngineImpl.cpp b/src/kernel/EngineImpl.cpp index b105c175be..a5cd1612c5 100644 --- a/src/kernel/EngineImpl.cpp +++ b/src/kernel/EngineImpl.cpp @@ -1,4 +1,4 @@ -/* Copyright (c) 2016-2022. The SimGrid Team. All rights reserved. */ +/* Copyright (c) 2016-2023. The SimGrid Team. All rights reserved. */ /* This program is free software; you can redistribute it and/or modify it * under the terms of the license (GNU LGPL) which comes with this package. */ @@ -10,9 +10,6 @@ #include #include -#define SIMIX_H_NO_DEPRECATED_WARNING // avoid deprecation warning on include (remove with XBT_ATTRIB_DEPRECATED_v332) -#include - #include "mc/mc.h" #include "src/kernel/EngineImpl.hpp" #include "src/kernel/resource/StandardLinkImpl.hpp" @@ -21,30 +18,27 @@ #include "src/mc/mc_replay.hpp" #include "src/smpi/include/smpi_actor.hpp" #include "src/surf/xml/platf.hpp" +#include "xbt/module.h" #include "xbt/xbt_modinter.h" /* whether initialization was already done */ #include -#ifndef _WIN32 #include -#endif /* _WIN32 */ #if SIMGRID_HAVE_MC #include "src/mc/remote/AppSide.hpp" #endif -double NOW = 0; - XBT_LOG_NEW_DEFAULT_CATEGORY(ker_engine, "Logging specific to Engine (kernel)"); -namespace simgrid { -namespace kernel { +namespace simgrid::kernel { +double EngineImpl::now_ = 0.0; EngineImpl* EngineImpl::instance_ = nullptr; /* That singleton is awful too. */ config::Flag cfg_breakpoint{"debug/breakpoint", "When non-negative, raise a SIGTRAP after given (simulated) time", -1.0}; config::Flag cfg_verbose_exit{"debug/verbose-exit", "Display the actor status at exit", true}; -constexpr std::initializer_list> context_factories = { +constexpr std::initializer_list> context_factories = { #if HAVE_RAW_CONTEXTS {"raw", &context::raw_factory}, #endif @@ -64,19 +58,18 @@ static inline std::string contexts_list() { std::string res; std::string sep = ""; - for (auto const& factory : context_factories) { - res += sep + factory.first; + for (auto const& [factory_name, _] : context_factories) { + res += sep + factory_name; sep = ", "; } return res; } static config::Flag context_factory_name("contexts/factory", - (std::string("Possible values: ") + contexts_list()).c_str(), + ("Possible values: " + contexts_list()).c_str(), context_factories.begin()->first); -} // namespace kernel -} // namespace simgrid +} // namespace simgrid::kernel XBT_ATTRIB_NORETURN static void inthandler(int) { @@ -90,7 +83,6 @@ XBT_ATTRIB_NORETURN static void inthandler(int) exit(1); } -#ifndef _WIN32 static void segvhandler(int signum, siginfo_t* siginfo, void* /*context*/) { if ((siginfo->si_signo == SIGSEGV && siginfo->si_code == SEGV_ACCERR) || siginfo->si_signo == SIGBUS) { @@ -111,7 +103,7 @@ static void segvhandler(int signum, siginfo_t* siginfo, void* /*context*/) } else if (siginfo->si_signo == SIGSEGV) { fprintf(stderr, "Segmentation fault.\n"); #if HAVE_SMPI - if (smpi_enabled() && smpi_cfg_privatization() == SmpiPrivStrategies::NONE) { + if (SMPI_is_inited() && smpi_cfg_privatization() == SmpiPrivStrategies::NONE) { #if HAVE_PRIVATIZATION fprintf(stderr, "Try to enable SMPI variable privatization with --cfg=smpi/privatization:yes.\n"); #else @@ -161,34 +153,26 @@ static void install_segvhandler() } } -#endif /* _WIN32 */ - -namespace simgrid { -namespace kernel { +namespace simgrid::kernel { EngineImpl::~EngineImpl() { - /* Since hosts_ is a std::map, the hosts are destroyed in the lexicographic order, which ensures that the output is - * reproducible. - */ - while (not hosts_.empty()) - hosts_.begin()->second->destroy(); - /* Also delete the other data */ delete netzone_root_; - for (auto const& kv : netpoints_) - delete kv.second; + for (auto const& [_, netpoint] : netpoints_) + delete netpoint; - while (not links_.empty()) - links_.begin()->second->destroy(); + for (auto const& [_, mailbox] : mailboxes_) + delete mailbox; - for (auto const& kv : mailboxes_) - delete kv.second; + /* Kill all actors (but maestro) */ + maestro_->kill_all(); + run_all_actors(); + empty_trash(); + + delete maestro_; + delete context_factory_; - /* Free the remaining data structures */ -#if SIMGRID_HAVE_MC - xbt_dynar_free(&actors_vector_); -#endif /* clear models before freeing handle, network models can use external callback defined in the handle */ models_prio_.clear(); } @@ -202,7 +186,7 @@ void EngineImpl::initialize(int* argc, char** argv) // The communication initialization is done ASAP, as we need to get some init parameters from the MC for different // layers. But instance_ needs to be created, as we send the address of some of its fields to the MC that wants to // read them directly. - simgrid::mc::AppSide::initialize(actors_vector_); + simgrid::mc::AppSide::initialize(); #endif if (xbt_initialized == 0) { @@ -215,10 +199,7 @@ void EngineImpl::initialize(int* argc, char** argv) /* Prepare to display some more info when dying on Ctrl-C pressing */ std::signal(SIGINT, inthandler); - -#ifndef _WIN32 install_segvhandler(); -#endif /* register a function to be called by SURF after the environment creation */ sg_platf_init(); @@ -250,14 +231,14 @@ void EngineImpl::context_mod_init() const #endif /* select the context factory to use to create the contexts */ - if (context::factory_initializer != nullptr) { // Give Java a chance to hijack the factory mechanism - instance_->set_context_factory(context::factory_initializer()); + if (context::ContextFactory::initializer) { // Give Java a chance to hijack the factory mechanism + instance_->set_context_factory(context::ContextFactory::initializer()); return; } /* use the factory specified by --cfg=contexts/factory:value */ - for (auto const& factory : context_factories) - if (context_factory_name == factory.first) { - instance_->set_context_factory(factory.second()); + for (auto const& [factory_name, factory] : context_factories) + if (context_factory_name == factory_name) { + instance_->set_context_factory(factory()); break; } @@ -279,7 +260,7 @@ void EngineImpl::context_mod_init() const XBT_ERROR(" (boost was disabled at compilation time on this machine -- check configure logs for details. Did you " "install the libboost-context-dev package?)"); #endif - XBT_ERROR(" thread: slow portability layer using pthreads as provided by gcc"); + XBT_ERROR(" thread: slow portability layer using standard threads as provided by libstdc"); xbt_die("Please use a valid factory."); } } @@ -291,7 +272,7 @@ void EngineImpl::shutdown() XBT_DEBUG("EngineImpl::shutdown() called. Simulation's over."); #if HAVE_SMPI if (not instance_->actor_list_.empty()) { - if (smpi_process()->initialized()) { + if (smpi_process() && smpi_process()->initialized()) { xbt_die("Process exited without calling MPI_Finalize - Killing simulation"); } else { XBT_WARN("Process called exit when leaving - Skipping cleanups"); @@ -307,18 +288,6 @@ void EngineImpl::shutdown() xbt_die("Bailing out to avoid that stop-before-start madness. Please fix your code."); } - /* Kill all actors (but maestro) */ - instance_->maestro_->kill_all(); - instance_->run_all_actors(); - instance_->empty_trash(); - - /* Let's free maestro now */ - delete instance_->maestro_; - instance_->maestro_ = nullptr; - - /* Finish context module and SURF */ - instance_->destroy_context_factory(); - while (not timer::kernel_timers().empty()) { delete timer::kernel_timers().top().second; timer::kernel_timers().pop(); @@ -339,9 +308,6 @@ void EngineImpl::seal_platform() const return; sealed = true; - /* sealing resources before run: links */ - for (auto const& kv : links_) - kv.second->get_iface()->seal(); /* seal netzone root, recursively seal children netzones, hosts and disks */ netzone_root_->seal(); } @@ -349,10 +315,7 @@ void EngineImpl::seal_platform() const void EngineImpl::load_platform(const std::string& platf) { double start = xbt_os_time(); - if (boost::algorithm::ends_with(platf, ".so") or boost::algorithm::ends_with(platf, ".dylib")) { -#ifdef _WIN32 - xbt_die("loading platform through shared library isn't supported on windows"); -#else + if (boost::algorithm::ends_with(platf, ".so") || boost::algorithm::ends_with(platf, ".dylib")) { void* handle = dlopen(platf.c_str(), RTLD_LAZY); xbt_assert(handle, "Impossible to open platform file: %s", platf.c_str()); platf_handle_ = std::unique_ptr>(handle, dlclose); @@ -361,7 +324,6 @@ void EngineImpl::load_platform(const std::string& platf) const char* dlsym_error = dlerror(); xbt_assert(not dlsym_error, "Error: %s", dlsym_error); callable(*simgrid::s4u::Engine::get_instance()); -#endif /* _WIN32 */ } else { parse_platform_file(platf); } @@ -395,7 +357,7 @@ void EngineImpl::add_model(std::shared_ptr model, const std::ve xbt_assert(models_prio_.find(model_name) == models_prio_.end(), "Model %s already exists, use model.set_name() to change its name", model_name.c_str()); - for (const auto dep : dependencies) { + for (const auto* dep : dependencies) { xbt_assert(models_prio_.find(dep->get_name()) != models_prio_.end(), "Model %s doesn't exists. Impossible to use it as dependency.", dep->get_name().c_str()); } @@ -403,21 +365,15 @@ void EngineImpl::add_model(std::shared_ptr model, const std::ve models_prio_[model_name] = std::move(model); } -void EngineImpl::add_split_duplex_link(const std::string& name, std::unique_ptr link) -{ - split_duplex_links_[name] = std::move(link); -} - /** Wake up all actors waiting for a Surf action to finish */ -void EngineImpl::wake_all_waiting_actors() const +void EngineImpl::handle_ended_actions() const { for (auto const& model : models_) { XBT_DEBUG("Handling the failed actions (if any)"); while (auto* action = model->extract_failed_action()) { XBT_DEBUG(" Handling Action %p", action); - if (action->get_activity() != nullptr) { - // If nobody told the interface that the activity has failed, that's because no actor waits on it (maestro - // started it). SimDAG I see you! + if (action->get_activity() != nullptr) { // Skip vcpu actions + // Action failures are not automatically reported when the action is started by maestro (as in SimDAG) if (action->get_activity()->get_actor() == maestro_) action->get_activity()->get_iface()->complete(s4u::Activity::State::FAILED); @@ -427,11 +383,8 @@ void EngineImpl::wake_all_waiting_actors() const XBT_DEBUG("Handling the terminated actions (if any)"); while (auto* action = model->extract_done_action()) { XBT_DEBUG(" Handling Action %p", action); - if (action->get_activity() == nullptr) - XBT_DEBUG("probably vcpu's action %p, skip", action); - else { - // If nobody told the interface that the activity is finished, that's because no actor waits on it (maestro - // started it). SimDAG I see you! + if (action->get_activity() != nullptr) { + // Action termination are not automatically reported when the action is started by maestro (as in SimDAG) action->get_activity()->set_finish_time(action->get_finish_time()); if (action->get_activity()->get_actor() == maestro_) @@ -451,10 +404,10 @@ void EngineImpl::wake_all_waiting_actors() const */ void EngineImpl::run_all_actors() { - instance_->get_context_factory()->run_all(); + instance_->get_context_factory()->run_all(actors_to_run_); for (auto const& actor : actors_to_run_) - if (actor->context_->to_be_freed()) + if (actor->to_be_freed()) actor->cleanup_from_kernel(); actors_to_run_.swap(actors_that_ran_); @@ -464,14 +417,7 @@ void EngineImpl::run_all_actors() actor::ActorImpl* EngineImpl::get_actor_by_pid(aid_t pid) { auto item = actor_list_.find(pid); - if (item != actor_list_.end()) - return item->second; - - // Search the trash - for (auto& a : actors_to_destroy_) - if (a.get_pid() == pid) - return &a; - return nullptr; // Not found, even in the trash + return item == actor_list_.end() ? nullptr : item->second; } void EngineImpl::remove_daemon(actor::ActorImpl* actor) @@ -511,9 +457,7 @@ void EngineImpl::display_all_actor_status() const XBT_INFO("%zu actors are still running, waiting for something.", actor_list_.size()); /* List the actors and their state */ XBT_INFO("Legend of the following listing: \"Actor (@): \""); - for (auto const& kv : actor_list_) { - const actor::ActorImpl* actor = kv.second; - + for (auto const& [_, actor] : actor_list_) { if (actor->waiting_synchro_) { const char* synchro_description = "unknown"; @@ -532,13 +476,17 @@ void EngineImpl::display_all_actor_status() const if (boost::dynamic_pointer_cast(actor->waiting_synchro_) != nullptr) synchro_description = "I/O"; - XBT_INFO("Actor %ld (%s@%s): waiting for %s activity %#zx (%s) in state %s to finish", actor->get_pid(), + XBT_INFO("Actor %ld (%s@%s): waiting for %s activity %#zx (%s) in state %s to finish %s", actor->get_pid(), actor->get_cname(), actor->get_host()->get_cname(), synchro_description, (xbt_log_no_loc ? (size_t)0xDEADBEEF : (size_t)actor->waiting_synchro_.get()), - actor->waiting_synchro_->get_cname(), actor->waiting_synchro_->get_state_str()); + actor->waiting_synchro_->get_cname(), actor->waiting_synchro_->get_state_str(), + (actor->simcall_.observer_ != nullptr && not xbt_log_no_loc + ? actor->simcall_.observer_->to_string().c_str() + : "")); } else { XBT_INFO("Actor %ld (%s@%s) simcall %s", actor->get_pid(), actor->get_cname(), actor->get_host()->get_cname(), - actor->simcall_.get_cname()); + (actor->simcall_.observer_ != nullptr ? actor->simcall_.observer_->to_string().c_str() + : actor->simcall_.get_cname())); } } } @@ -548,7 +496,7 @@ void EngineImpl::presolve() const XBT_DEBUG("Consume all trace events occurring before the starting time."); double next_event_date; while ((next_event_date = profile::future_evt_set.next_date()) != -1.0) { - if (next_event_date > NOW) + if (next_event_date > now_) break; double value = -1.0; @@ -561,7 +509,7 @@ void EngineImpl::presolve() const XBT_DEBUG("Set every models in the right state by updating them to 0."); for (auto const& model : models_) - model->update_actions_state(NOW, 0.0); + model->update_actions_state(now_, 0.0); } double EngineImpl::solve(double max_date) const @@ -571,9 +519,9 @@ double EngineImpl::solve(double max_date) const resource::Resource* resource = nullptr; if (max_date != -1.0) { - xbt_assert(max_date >= NOW, "You asked to simulate up to %f, but that's in the past already", max_date); + xbt_assert(max_date >= now_, "You asked to simulate up to %f, but that's in the past already", max_date); - time_delta = max_date - NOW; + time_delta = max_date - now_; } XBT_DEBUG("Looking for next event in all models"); @@ -581,7 +529,7 @@ double EngineImpl::solve(double max_date) const if (not model->next_occurring_event_is_idempotent()) { continue; } - double next_event = model->next_occurring_event(NOW); + double next_event = model->next_occurring_event(now_); if ((time_delta < 0.0 || next_event < time_delta) && next_event >= 0.0) { time_delta = next_event; } @@ -602,9 +550,9 @@ double EngineImpl::solve(double max_date) const continue; if (next_event_date != -1.0) { - time_delta = std::min(next_event_date - NOW, time_delta); + time_delta = std::min(next_event_date - now_, time_delta); } else { - time_delta = std::max(next_event_date - NOW, time_delta); // Get the positive component + time_delta = std::max(next_event_date - now_, time_delta); // Get the positive component } XBT_DEBUG("Run the NS3 network at most %fs", time_delta); @@ -616,29 +564,31 @@ double EngineImpl::solve(double max_date) const time_delta = model_next_action_end; } - if (next_event_date < 0.0 || (next_event_date > NOW + time_delta)) { + if (next_event_date < 0.0 || (next_event_date > now_ + time_delta)) { // next event may have already occurred or will after the next resource change, then bail out XBT_DEBUG("no next usable TRACE event. Stop searching for it"); break; } - XBT_DEBUG("Updating models (min = %g, NOW = %g, next_event_date = %g)", time_delta, NOW, next_event_date); + XBT_DEBUG("Updating models (min = %g, NOW = %g, next_event_date = %g)", time_delta, now_, next_event_date); while (auto* event = profile::future_evt_set.pop_leq(next_event_date, &value, &resource)) { + if(value<0) + continue; if (resource->is_used() || (watched_hosts().find(resource->get_cname()) != watched_hosts().end())) { - time_delta = next_event_date - NOW; + time_delta = next_event_date - now_; XBT_DEBUG("This event invalidates the next_occurring_event() computation of models. Next event set to %f", time_delta); } - // FIXME: I'm too lame to update NOW live, so I change it and restore it so that the real update with surf_min + // FIXME: I'm too lame to update now_ live, so I change it and restore it so that the real update with surf_min // will work - double round_start = NOW; - NOW = next_event_date; + double round_start = now_; + now_ = next_event_date; /* update state of the corresponding resource to the new value. Does not touch lmm. It will be modified if needed when updating actions */ XBT_DEBUG("Calling update_resource_state for resource %s", resource->get_cname()); resource->apply_event(event, value); - NOW = round_start; + now_ = round_start; } } @@ -655,11 +605,11 @@ double EngineImpl::solve(double max_date) const XBT_DEBUG("Duration set to %f", time_delta); // Bump the time: jump into the future - NOW = NOW + time_delta; + now_ += time_delta; // Inform the models of the date change for (auto const& model : models_) - model->update_actions_state(NOW, time_delta); + model->update_actions_state(now_, time_delta); s4u::Engine::on_time_advance(time_delta); @@ -693,7 +643,7 @@ void EngineImpl::run(double max_date) if (cfg_breakpoint >= 0.0 && simgrid_get_clock() >= cfg_breakpoint) { XBT_DEBUG("Breakpoint reached (%g)", cfg_breakpoint.get()); - cfg_breakpoint = -1.0; + cfg_breakpoint = -1.0; // Let the simulation continue without hiting the breakpoint again and again #ifdef SIGTRAP std::raise(SIGTRAP); #else @@ -707,34 +657,16 @@ void EngineImpl::run(double max_date) /* Run all actors that are ready to run, possibly in parallel */ run_all_actors(); - /* answer sequentially and in a fixed arbitrary order all the simcalls that were issued during that sub-round */ - - /* WARNING, the order *must* be fixed or you'll jeopardize the simulation reproducibility (see RR-7653) */ - - /* Here, the order is ok because: - * - * Only maestro adds stuff to the actors_to_run array, so the execution order of user contexts do not impact its order. - * - * In addition, actors remain sorted through an arbitrary but fixed order in all cases: - * - * - If there is no killing during the simulation, actors remain sorted according by their PID. - * - Killer actors are moved to the end of the scheduling round (to let victims finish their simcall before dying), but - * (1) this decision of killing is reproducible because the simulation was reproducible until then - * (2) this reordering introduces no reproducibility hazard in the subsequent simulation. - * Even the order change induced by the actor killing is perfectly reproducible. - * - * So the array order is implicit and somewhat complex, but fixed and reproducible (science works, http://xkcd.com/54/). - * - * We could manually sort the actors_that_ran array so that simcalls are handled in an easy to predict order - * (e.g. "according to the PID of issuer"), but it's not mandatory for the simulation soundness and reproducibility, - * and would thus be a pure waste of time. + /* answer sequentially and in a fixed arbitrary order all the simcalls that were issued during that sub-round. + * The order must be fixed for the simulation to be reproducible (see RR-7653). It's OK here because only maestro + * changes the list. Killer actors are moved to the end to let victims finish their simcall before dying, but + * the order remains reproducible (even if arbitrarily). No need to sort the vector for sake of reproducibility. */ - for (auto const& actor : actors_that_ran_) if (actor->simcall_.call_ != actor::Simcall::Type::NONE) actor->simcall_handle(0); - wake_all_waiting_actors(); + handle_ended_actions(); /* If only daemon actors remain, cancel their actions, mark them to die and reschedule them */ if (actor_list_.size() == daemons_.size()) @@ -753,19 +685,15 @@ void EngineImpl::run(double max_date) next_time = std::min(next_time, max_date); } - XBT_DEBUG("Calling solve(%g) %g", next_time, NOW); + XBT_DEBUG("Calling solve(%g) %g", next_time, now_); elapsed_time = solve(next_time); - XBT_DEBUG("Moving time ahead. NOW=%g; elapsed: %g", NOW, elapsed_time); - - /* Notify all the hosts that have failed */ - /* FIXME: iterate through the list of failed host and mark each of them */ - /* as failed. On each host, signal all the running actors with host_fail */ + XBT_DEBUG("Moving time ahead. NOW=%g; elapsed: %g", now_, elapsed_time); // Execute timers until there isn't anything to be done: bool again = false; do { again = timer::Timer::execute_all(); - wake_all_waiting_actors(); + handle_ended_actions(); } while (again); /* Clean actors to destroy */ @@ -779,17 +707,18 @@ void EngineImpl::run(double max_date) XBT_CRITICAL("Oops! Daemon actors cannot do any blocking activity (communications, synchronization, etc) " "once the simulation is over. Please fix your on_exit() functions."); } else { - XBT_CRITICAL("Oops! Deadlock or code not perfectly clean."); + XBT_CRITICAL("Oops! Deadlock detected, some activities are still around but will never complete. " + "This usually happens when the user code is not perfectly clean."); } display_all_actor_status(); simgrid::s4u::Engine::on_deadlock(); - for (auto const& kv : actor_list_) { - XBT_DEBUG("Kill %s", kv.second->get_cname()); - maestro_->kill(kv.second); + for (auto const& [_, actor] : actor_list_) { + XBT_DEBUG("Kill %s", actor->get_cname()); + maestro_->kill(actor); } } } while ((vetoed_activities == nullptr || vetoed_activities->empty()) && - ((elapsed_time > -1.0 && not double_equals(max_date, NOW, 0.00001)) || has_actors_to_run())); + ((elapsed_time > -1.0 && not double_equals(max_date, now_, 0.00001)) || has_actors_to_run())); if (not actor_list_.empty() && max_date < 0 && not(vetoed_activities == nullptr || vetoed_activities->empty())) THROW_IMPOSSIBLE; @@ -799,12 +728,6 @@ void EngineImpl::run(double max_date) double EngineImpl::get_clock() { - return NOW; -} -} // namespace kernel -} // namespace simgrid - -void SIMIX_run() // XBT_ATTRIB_DEPRECATED_v332 -{ - simgrid::kernel::EngineImpl::get_instance()->run(-1); + return now_; } +} // namespace simgrid::kernel