X-Git-Url: http://bilbo.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/998dcee3c1fd2e8b274506571e381296101d72e6..13f8c10bd72f90cb6f70e9a03ce37dcbf6e01226:/src/kernel/activity/ExecImpl.cpp diff --git a/src/kernel/activity/ExecImpl.cpp b/src/kernel/activity/ExecImpl.cpp index 20fdf8ddb2..88432f686d 100644 --- a/src/kernel/activity/ExecImpl.cpp +++ b/src/kernel/activity/ExecImpl.cpp @@ -1,161 +1,255 @@ -/* Copyright (c) 2007-2018. The SimGrid Team. All rights reserved. */ +/* Copyright (c) 2007-2023. 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. */ -#include "simgrid/modelchecker.h" -#include "src/mc/mc_replay.hpp" +#include +#include +#include +#include #include "src/kernel/activity/ExecImpl.hpp" -#include "src/simix/smx_host_private.hpp" -#include "src/surf/surf_interface.hpp" -#include "src/surf/cpu_interface.hpp" +#include "src/kernel/actor/ActorImpl.hpp" +#include "src/kernel/actor/SimcallObserver.hpp" +#include "src/kernel/resource/CpuImpl.hpp" +#include "src/kernel/resource/HostImpl.hpp" +#include "src/mc/mc_replay.hpp" -#include "simgrid/s4u/Host.hpp" +XBT_LOG_NEW_DEFAULT_SUBCATEGORY(ker_cpu, kernel, "Kernel cpu-related synchronization"); -XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(simix_process); +namespace simgrid::kernel::activity { -simgrid::kernel::activity::ExecImpl::ExecImpl(std::string name, resource::Action* surf_action, - resource::Action* timeout_detector, s4u::Host* host) - : ActivityImpl(name), host_(host), surf_action_(surf_action), timeout_detector_(timeout_detector) +ExecImpl::ExecImpl() { - this->state_ = SIMIX_RUNNING; + piface_ = new s4u::Exec(this); + actor::ActorImpl* self = actor::ActorImpl::self(); + if (self) { + set_actor(self); + self->activities_.insert(this); + } +} - surf_action_->set_data(this); - if (timeout_detector != nullptr) - timeout_detector->set_data(this); +ExecImpl& ExecImpl::set_host(s4u::Host* host) +{ + ActivityImpl::set_hosts({host}); + return *this; +} - XBT_DEBUG("Create exec %p", this); +ExecImpl& ExecImpl::set_hosts(const std::vector& hosts) +{ + ActivityImpl::set_hosts(hosts); + return *this; } -simgrid::kernel::activity::ExecImpl::~ExecImpl() +ExecImpl& ExecImpl::set_timeout(double timeout) { - if (surf_action_) - surf_action_->unref(); - if (timeout_detector_) - timeout_detector_->unref(); - XBT_DEBUG("Destroy exec %p", this); + if (timeout >= 0 && not MC_is_active() && not MC_record_replay_is_active()) { + timeout_detector_.reset(get_host()->get_cpu()->sleep(timeout)); + timeout_detector_->set_activity(this); + } + return *this; } -void simgrid::kernel::activity::ExecImpl::suspend() +ExecImpl& ExecImpl::set_flops_amount(double flops_amount) { - XBT_VERB("This exec is suspended (remain: %f)", surf_action_->get_remains()); - if (surf_action_ != nullptr) - surf_action_->suspend(); - on_suspended(this); + flops_amounts_.assign(1, flops_amount); + return *this; } -void simgrid::kernel::activity::ExecImpl::resume() +ExecImpl& ExecImpl::set_flops_amounts(const std::vector& flops_amounts) { - XBT_VERB("This exec is resumed (remain: %f)", surf_action_->get_remains()); - if (surf_action_ != nullptr) - surf_action_->resume(); - on_resumed(this); + flops_amounts_ = flops_amounts; + return *this; } -void simgrid::kernel::activity::ExecImpl::cancel() + +ExecImpl& ExecImpl::set_bytes_amounts(const std::vector& bytes_amounts) { - XBT_VERB("This exec %p is canceled", this); - if (surf_action_ != nullptr) - surf_action_->cancel(); + bytes_amounts_ = bytes_amounts; + + return *this; +} +ExecImpl& ExecImpl::set_thread_count(int thread_count) +{ + thread_count_ = thread_count; + + return *this; } -double simgrid::kernel::activity::ExecImpl::get_remaining() +ExecImpl* ExecImpl::start() { - xbt_assert(host_ != nullptr, "Calling remains() on a parallel execution is not allowed. " - "We would need to return a vector instead of a scalar. " - "Did you mean remainingRatio() instead?"); + set_state(State::RUNNING); + if (not MC_is_active() && not MC_record_replay_is_active()) { + if (get_hosts().size() == 1) { + if (thread_count_ == 1) { + model_action_ = get_host()->get_cpu()->execution_start(flops_amounts_.front(), bound_); + model_action_->set_sharing_penalty(sharing_penalty_); + } else { + auto host_model = get_host()->get_netpoint()->get_englobing_zone()->get_host_model(); + model_action_ = host_model->execute_thread(get_host(), flops_amounts_.front(), thread_count_); + } + model_action_->set_category(get_tracing_category()); + } else { + // get the model from first host since we have only 1 by now + auto host_model = get_host()->get_netpoint()->get_englobing_zone()->get_host_model(); + model_action_ = host_model->execute_parallel(get_hosts(), flops_amounts_.data(), bytes_amounts_.data(), -1); + } + model_action_->set_activity(this); + set_start_time(model_action_->get_start_time()); + } - return surf_action_ ? surf_action_->get_remains() : 0; + XBT_DEBUG("Create execute synchro %p: %s", this, get_cname()); + return this; } -double simgrid::kernel::activity::ExecImpl::get_remaining_ratio() +double ExecImpl::get_remaining() const { - if (host_ == - nullptr) // parallel task: their remain is already between 0 and 1 (see comment in ExecImpl::get_remaining()) - return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains(); - else // Actually compute the ratio for sequential tasks - return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains() / surf_action_->get_cost(); + if (get_state() == State::WAITING || get_state() == State::FAILED) + return flops_amounts_.front(); + return ActivityImpl::get_remaining(); } -void simgrid::kernel::activity::ExecImpl::set_bound(double bound) +double ExecImpl::get_seq_remaining_ratio() { - if (surf_action_) - surf_action_->set_bound(bound); + if (get_state() == State::WAITING) + return 1; + return (model_action_ == nullptr) ? 0 : model_action_->get_remains() / model_action_->get_cost(); } -void simgrid::kernel::activity::ExecImpl::set_priority(double priority) + +double ExecImpl::get_par_remaining_ratio() { - if (surf_action_) - surf_action_->set_priority(priority); + // parallel task: their remain is already between 0 and 1 + if (get_state() == State::WAITING) + return 1; + return (model_action_ == nullptr) ? 0 : model_action_->get_remains(); } -void simgrid::kernel::activity::ExecImpl::set_category(std::string category) +ExecImpl& ExecImpl::set_bound(double bound) { - if (surf_action_) - surf_action_->set_category(category); + bound_ = bound; + return *this; } -void simgrid::kernel::activity::ExecImpl::post() +ExecImpl& ExecImpl::set_sharing_penalty(double sharing_penalty) { - if (host_ && host_->is_off()) { /* FIXME: handle resource failure for parallel tasks too */ - /* If the host running the synchro failed, notice it. This way, the asking + sharing_penalty_ = sharing_penalty; + return *this; +} + +ExecImpl& ExecImpl::update_sharing_penalty(double sharing_penalty) +{ + sharing_penalty_ = sharing_penalty; + model_action_->set_sharing_penalty(sharing_penalty); + return *this; +} + +void ExecImpl::post() +{ + xbt_assert(model_action_ != nullptr); + if (auto const& hosts = get_hosts(); + std::any_of(hosts.begin(), hosts.end(), [](const s4u::Host* host) { return not host->is_on(); })) { + /* If one of the hosts running the synchro failed, notice it. This way, the asking * process can be killed if it runs on that host itself */ - state_ = SIMIX_FAILED; - } else if (surf_action_ && surf_action_->get_state() == simgrid::kernel::resource::Action::State::FAILED) { - /* If the host running the synchro didn't fail, then the synchro was canceled */ - state_ = SIMIX_CANCELED; - } else if (timeout_detector_ && - timeout_detector_->get_state() == simgrid::kernel::resource::Action::State::FINISHED) { - state_ = SIMIX_TIMEOUT; + set_state(State::FAILED); + } else if (model_action_->get_state() == resource::Action::State::FAILED) { + /* If all the hosts are running the synchro didn't fail, then the synchro was canceled */ + set_state(State::CANCELED); + } else if (timeout_detector_ && timeout_detector_->get_state() == resource::Action::State::FINISHED && + model_action_->get_remains() > 0.0) { + model_action_->set_state(resource::Action::State::FAILED); + set_state(State::TIMEOUT); } else { - state_ = SIMIX_DONE; + set_state(State::DONE); } - on_completion(this); - - if (surf_action_) { - surf_action_->unref(); - surf_action_ = nullptr; + clean_action(); + timeout_detector_.reset(); + if (get_actor() != nullptr) { + get_actor()->activities_.erase(this); } - if (timeout_detector_) { - timeout_detector_->unref(); - timeout_detector_ = nullptr; + if (get_state() != State::FAILED && cb_id_ >= 0) + s4u::Host::on_state_change.disconnect(cb_id_); + /* Answer all simcalls associated with the synchro */ + finish(); +} + +void ExecImpl::set_exception(actor::ActorImpl* issuer) +{ + switch (get_state()) { + case State::FAILED: + static_cast(get_iface())->complete(s4u::Activity::State::FAILED); + if (issuer->get_host()->is_on()) + issuer->exception_ = std::make_exception_ptr(HostFailureException(XBT_THROW_POINT, "Host failed")); + else /* else, the actor will be killed with no possibility to survive */ + issuer->set_wannadie(); + break; + + case State::CANCELED: + issuer->exception_ = std::make_exception_ptr(CancelException(XBT_THROW_POINT, "Execution Canceled")); + break; + + case State::TIMEOUT: + issuer->exception_ = std::make_exception_ptr(TimeoutException(XBT_THROW_POINT, "Timeouted")); + break; + + default: + xbt_assert(get_state() == State::DONE, "Internal error in ExecImpl::finish(): unexpected synchro state %s", + get_state_str()); } +} +void ExecImpl::finish() +{ + XBT_DEBUG("ExecImpl::finish() in state %s", get_state_str()); + while (not simcalls_.empty()) { + actor::Simcall* simcall = simcalls_.front(); + simcalls_.pop_front(); + + if (simcall->call_ == actor::Simcall::Type::NONE) // FIXME: maybe a better way to handle this case + continue; // if process handling comm is killed + + handle_activity_waitany(simcall); - /* If there are simcalls associated with the synchro, then answer them */ - if (not simcalls_.empty()) - SIMIX_execution_finish(this); + set_exception(simcall->issuer_); + + simcall->issuer_->waiting_synchro_ = nullptr; + /* Fail the process if the host is down */ + if (simcall->issuer_->get_host()->is_on()) + simcall->issuer_->simcall_answer(); + else + simcall->issuer_->set_wannadie(); + } } -simgrid::kernel::activity::ActivityImpl* -simgrid::kernel::activity::ExecImpl::migrate(simgrid::s4u::Host* to) +void ExecImpl::reset() { + clear_hosts(); + bytes_amounts_.clear(); + flops_amounts_.clear(); + set_start_time(-1.0); +} +ActivityImpl* ExecImpl::migrate(s4u::Host* to) +{ if (not MC_is_active() && not MC_record_replay_is_active()) { - simgrid::kernel::resource::Action* old_action = this->surf_action_; - simgrid::kernel::resource::Action* new_action = to->pimpl_cpu->execution_start(old_action->get_cost()); + resource::Action* old_action = this->model_action_; + resource::Action* new_action = to->get_cpu()->execution_start(old_action->get_cost(), old_action->get_user_bound()); new_action->set_remains(old_action->get_remains()); - new_action->set_data(this); - new_action->set_priority(old_action->get_priority()); + new_action->set_activity(this); + new_action->set_sharing_penalty(old_action->get_sharing_penalty()); + new_action->set_user_bound(old_action->get_user_bound()); - // FIXME: the user-defined bound seem to not be kept by LMM, that seem to overwrite it for the multi-core modeling. - // I hope that the user did not provide any. - - old_action->set_data(nullptr); + old_action->set_activity(nullptr); old_action->cancel(); old_action->unref(); - this->surf_action_ = new_action; + this->model_action_ = new_action; } - on_migration(this, to); + on_migration(*this, to); return this; } /************* * Callbacks * *************/ -simgrid::xbt::signal simgrid::kernel::activity::ExecImpl::on_creation; -simgrid::xbt::signal simgrid::kernel::activity::ExecImpl::on_completion; -simgrid::xbt::signal simgrid::kernel::activity::ExecImpl::on_resumed; -simgrid::xbt::signal simgrid::kernel::activity::ExecImpl::on_suspended; -simgrid::xbt::signal - simgrid::kernel::activity::ExecImpl::on_migration; +xbt::signal ExecImpl::on_migration; + +} // namespace simgrid::kernel::activity