-/* Copyright (c) 2016-2021. The SimGrid Team. All rights reserved. */
+/* Copyright (c) 2016-2022. 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. */
#include "mc/mc.h"
#include "src/kernel/EngineImpl.hpp"
+#include "src/kernel/resource/StandardLinkImpl.hpp"
#include "src/kernel/resource/profile/Profile.hpp"
#include "src/mc/mc_record.hpp"
#include "src/mc/mc_replay.hpp"
#include "src/smpi/include/smpi_actor.hpp"
-#include "src/surf/network_interface.hpp"
#include "src/surf/xml/platf.hpp"
-#include "surf/surf.hpp" //surf_presolve() and surf_solve()
-#include "xbt/xbt_modinter.h" /* whether initialization was already done */
+#include "xbt/xbt_modinter.h" /* whether initialization was already done */
#include <boost/algorithm/string/predicate.hpp>
#ifndef _WIN32
#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 {
EngineImpl* EngineImpl::instance_ = nullptr; /* That singleton is awful too. */
/* register a function to be called by SURF after the environment creation */
sg_platf_init();
- s4u::Engine::on_platform_created.connect(surf_presolve);
+ s4u::Engine::on_platform_created_cb([this]() { this->presolve(); });
if (config::get_value<bool>("debug/clean-atexit"))
atexit(shutdown);
{
xbt_assert(not instance_->has_context_factory());
-#if HAVE_SMPI && (defined(__APPLE__) || defined(__NetBSD__))
+#if HAVE_SMPI && defined(__NetBSD__)
smpi_init_options_internal(false);
std::string priv = config::get_value<std::string>("smpi/privatization");
if (context_factory_name == "thread" && (priv == "dlopen" || priv == "yes" || priv == "default" || priv == "1")) {
instance_->empty_trash();
/* Let's free maestro now */
- instance_->destroy_maestro();
+ delete instance_->maestro_;
+ instance_->maestro_ = nullptr;
/* Finish context module and SURF */
instance_->destroy_context_factory();
instance_ = nullptr;
}
+void EngineImpl::seal_platform() const
+{
+ /* 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();
+}
+
void EngineImpl::load_platform(const std::string& platf)
{
double start = xbt_os_time();
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 (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()->get_actor() == maestro_)
+ action->get_activity()->get_iface()->complete(s4u::Activity::State::FAILED);
+
activity::ActivityImplPtr(action->get_activity())->post();
+ }
}
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
+ 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!
+ action->get_activity()->set_finish_time(action->get_finish_time());
+
+ if (action->get_activity()->get_actor() == maestro_)
+ action->get_activity()->get_iface()->complete(s4u::Activity::State::FINISHED);
+
activity::ActivityImplPtr(action->get_activity())->post();
+ }
}
}
}
}
}
-void EngineImpl::run()
+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)
+ break;
+
+ double value = -1.0;
+ resource::Resource* resource = nullptr;
+ while (auto* event = profile::future_evt_set.pop_leq(next_event_date, &value, &resource)) {
+ if (value >= 0)
+ resource->apply_event(event, value);
+ }
+ }
+
+ 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);
+}
+
+double EngineImpl::solve(double max_date) const
+{
+ double time_delta = -1.0; /* duration */
+ double value = -1.0;
+ 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);
+
+ time_delta = max_date - NOW;
+ }
+
+ XBT_DEBUG("Looking for next event in all models");
+ for (auto model : models_) {
+ if (not model->next_occurring_event_is_idempotent()) {
+ continue;
+ }
+ 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;
+ }
+ }
+
+ XBT_DEBUG("Min for resources (remember that NS3 don't update that value): %f", time_delta);
+
+ XBT_DEBUG("Looking for next trace event");
+
+ while (true) { // Handle next occurring events until none remains
+ double next_event_date = profile::future_evt_set.next_date();
+ XBT_DEBUG("Next TRACE event: %f", next_event_date);
+
+ for (auto model : models_) {
+ /* Skip all idempotent models, they were already treated above
+ * NS3 is the one to handled here */
+ if (model->next_occurring_event_is_idempotent())
+ continue;
+
+ if (next_event_date != -1.0) {
+ 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
+ }
+
+ XBT_DEBUG("Run the NS3 network at most %fs", time_delta);
+ // run until min or next flow
+ double model_next_action_end = model->next_occurring_event(time_delta);
+
+ XBT_DEBUG("Min for network : %f", model_next_action_end);
+ if (model_next_action_end >= 0.0)
+ time_delta = model_next_action_end;
+ }
+
+ 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);
+
+ while (auto* event = profile::future_evt_set.pop_leq(next_event_date, &value, &resource)) {
+ if (resource->is_used() || (watched_hosts().find(resource->get_cname()) != watched_hosts().end())) {
+ 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
+ // will work
+ 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;
+ }
+ }
+
+ /* FIXME: Moved this test to here to avoid stopping simulation if there are actions running on cpus and all cpus are
+ * with availability = 0. This may cause an infinite loop if one cpu has a trace with periodicity = 0 and the other a
+ * trace with periodicity > 0.
+ * The options are: all traces with same periodicity(0 or >0) or we need to change the way how the events are managed
+ */
+ if (time_delta < 0) {
+ XBT_DEBUG("No next event at all. Bail out now.");
+ return -1.0;
+ }
+
+ XBT_DEBUG("Duration set to %f", time_delta);
+
+ // Bump the time: jump into the future
+ NOW = NOW + time_delta;
+
+ // Inform the models of the date change
+ for (auto const& model : models_)
+ model->update_actions_state(NOW, time_delta);
+
+ s4u::Engine::on_time_advance(time_delta);
+
+ return time_delta;
+}
+
+void EngineImpl::run(double max_date)
+{
+ seal_platform();
+
if (MC_record_replay_is_active()) {
mc::replay(MC_record_path());
empty_trash();
return;
}
- double time = 0;
+ double elapsed_time = -1;
+ const std::set<s4u::Activity*>* vetoed_activities = s4u::Activity::get_vetoed_activities();
do {
XBT_DEBUG("New Schedule Round; size(queue)=%zu", actors_to_run_.size());
* - You are added from ActorImpl::simcall_answer() only. When this function is called depends on the
* resource kind (network, cpu, disk, whatever), but the same arguments hold. Let's take communications
* as an example.
- * - For communications, this function is called from SIMIX_comm_finish().
+ * - For communications, this function is called from CommImpl::finish().
* This function itself don't mess with the order since simcalls are handled in FIFO order.
* The function is called:
* - before the comm starts (invalid parameters, or resource already dead or whatever).
}
}
- time = timer::Timer::next();
- if (time > -1.0 || not actor_list_.empty()) {
- XBT_DEBUG("Calling surf_solve");
- time = surf_solve(time);
- XBT_DEBUG("Moving time ahead : %g", time);
+ // Compute the max_date of the next solve.
+ // It's either when a timer occurs, or when user-specified deadline is reached, or -1 if none is given
+ double next_time = timer::Timer::next();
+ if (next_time < 0 && max_date > -1) {
+ next_time = max_date;
+ } else if (next_time > -1 && max_date > -1) { // either both <0, or both >0
+ next_time = std::min(next_time, max_date);
}
+ 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 */
/* Clean actors to destroy */
empty_trash();
- XBT_DEBUG("### time %f, #actors %zu, #to_run %zu", time, actor_list_.size(), actors_to_run_.size());
+ XBT_DEBUG("### elapsed time %f, #actors %zu, #to_run %zu, #vetoed %d", elapsed_time, actor_list_.size(),
+ actors_to_run_.size(), (vetoed_activities == nullptr ? -1 : static_cast<int>(vetoed_activities->size())));
- if (time < 0. && actors_to_run_.empty() && not actor_list_.empty()) {
+ if (elapsed_time < 0. && actors_to_run_.empty() && not actor_list_.empty()) {
if (actor_list_.size() <= daemons_.size()) {
XBT_CRITICAL("Oops! Daemon actors cannot do any blocking activity (communications, synchronization, etc) "
"once the simulation is over. Please fix your on_exit() functions.");
maestro_->kill(kv.second);
}
}
- } while (time > -1.0 || has_actors_to_run());
- if (not actor_list_.empty())
+ } while ((vetoed_activities == nullptr || vetoed_activities->empty()) &&
+ ((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;
simgrid::s4u::Engine::on_simulation_end();
}
+
+double EngineImpl::get_clock()
+{
+ return NOW;
+}
} // namespace kernel
} // namespace simgrid
void SIMIX_run() // XBT_ATTRIB_DEPRECATED_v332
{
- simgrid::kernel::EngineImpl::get_instance()->run();
+ simgrid::kernel::EngineImpl::get_instance()->run(-1);
}
