X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/b468c99043663686aeeb0912a43b6a8c4a1d684b..39c935d6d5ee86d153f6f7e6a10d723ae7c57f6f:/src/kernel/activity/ExecImpl.cpp diff --git a/src/kernel/activity/ExecImpl.cpp b/src/kernel/activity/ExecImpl.cpp index dc7dad509e..3e27a03b91 100644 --- a/src/kernel/activity/ExecImpl.cpp +++ b/src/kernel/activity/ExecImpl.cpp @@ -1,151 +1,280 @@ -/* Copyright (c) 2007-2018. The SimGrid Team. All rights reserved. */ +/* Copyright (c) 2007-2021. 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 "src/kernel/activity/ExecImpl.hpp" +#include "simgrid/Exception.hpp" #include "simgrid/modelchecker.h" #include "src/mc/mc_replay.hpp" - -#include "src/kernel/activity/ExecImpl.hpp" -#include "src/simix/smx_host_private.hpp" -#include "src/surf/surf_interface.hpp" +#include "src/surf/HostImpl.hpp" #include "src/surf/cpu_interface.hpp" +#include "src/surf/surf_interface.hpp" #include "simgrid/s4u/Host.hpp" +#include + XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(simix_process); -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) +void simcall_HANDLER_execution_waitany_for(smx_simcall_t simcall, simgrid::kernel::activity::ExecImpl* execs[], + size_t count, double timeout) { - this->state_ = SIMIX_RUNNING; + if (timeout < 0.0) { + simcall->timeout_cb_ = nullptr; + } else { + simcall->timeout_cb_ = simgrid::simix::Timer::set(SIMIX_get_clock() + timeout, [simcall, execs, count]() { + simcall->timeout_cb_ = nullptr; + for (size_t i = 0; i < count; i++) { + // Remove the first occurrence of simcall: + auto* exec = execs[i]; + auto j = boost::range::find(exec->simcalls_, simcall); + if (j != exec->simcalls_.end()) + exec->simcalls_.erase(j); + } + simcall_execution_waitany_for__set__result(simcall, -1); + simcall->issuer_->simcall_answer(); + }); + } - surf_action_->set_data(this); - if (timeout_detector != nullptr) - timeout_detector->set_data(this); + for (size_t i = 0; i < count; i++) { + /* associate this simcall to the the synchro */ + auto* exec = execs[i]; + exec->simcalls_.push_back(simcall); - XBT_DEBUG("Create exec %p", this); + /* see if the synchro is already finished */ + if (exec->state_ != simgrid::kernel::activity::State::WAITING && + exec->state_ != simgrid::kernel::activity::State::RUNNING) { + exec->finish(); + break; + } + } } -simgrid::kernel::activity::ExecImpl::~ExecImpl() +namespace simgrid { +namespace kernel { +namespace activity { + +ExecImpl::ExecImpl() { - if (surf_action_) - surf_action_->unref(); - if (timeout_detector_) - timeout_detector_->unref(); - XBT_DEBUG("Destroy exec %p", this); + actor::ActorImpl* self = actor::ActorImpl::self(); + if (self) { + actor_ = self; + self->activities_.emplace_back(this); + } } -void simgrid::kernel::activity::ExecImpl::suspend() +ExecImpl& ExecImpl::set_host(s4u::Host* host) { - XBT_VERB("This exec is suspended (remain: %f)", surf_action_->get_remains()); - if (surf_action_ != nullptr) - surf_action_->suspend(); + hosts_.assign(1, host); + return *this; } -void simgrid::kernel::activity::ExecImpl::resume() +ExecImpl& ExecImpl::set_hosts(const std::vector& hosts) { - XBT_VERB("This exec is resumed (remain: %f)", surf_action_->get_remains()); - if (surf_action_ != nullptr) - surf_action_->resume(); + hosts_ = hosts; + return *this; } -void simgrid::kernel::activity::ExecImpl::cancel() + +ExecImpl& ExecImpl::set_timeout(double timeout) { - XBT_VERB("This exec %p is canceled", this); - if (surf_action_ != nullptr) - surf_action_->cancel(); + if (timeout > 0 && not MC_is_active() && not MC_record_replay_is_active()) { + timeout_detector_.reset(hosts_.front()->pimpl_cpu->sleep(timeout)); + timeout_detector_->set_activity(this); + } + return *this; } -double simgrid::kernel::activity::ExecImpl::get_remaining() +ExecImpl& ExecImpl::set_flops_amount(double flops_amount) { - 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?"); + flops_amounts_.assign(1, flops_amount); + return *this; +} + +ExecImpl& ExecImpl::set_flops_amounts(const std::vector& flops_amounts) +{ + flops_amounts_ = flops_amounts; + return *this; +} + +ExecImpl& ExecImpl::set_bytes_amounts(const std::vector& bytes_amounts) +{ + bytes_amounts_ = bytes_amounts; - return surf_action_ ? surf_action_->get_remains() : 0; + return *this; } -double simgrid::kernel::activity::ExecImpl::get_remaining_ratio() +ExecImpl* ExecImpl::start() { - 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(); + state_ = State::RUNNING; + if (not MC_is_active() && not MC_record_replay_is_active()) { + if (hosts_.size() == 1) { + surf_action_ = hosts_.front()->pimpl_cpu->execution_start(flops_amounts_.front()); + surf_action_->set_sharing_penalty(sharing_penalty_); + surf_action_->set_category(get_tracing_category()); + + if (bound_ > 0) + surf_action_->set_bound(bound_); + } else { + surf_action_ = surf_host_model->execute_parallel(hosts_, flops_amounts_.data(), bytes_amounts_.data(), -1); + } + surf_action_->set_activity(this); + } + + XBT_DEBUG("Create execute synchro %p: %s", this, get_cname()); + return this; } -void simgrid::kernel::activity::ExecImpl::set_bound(double bound) +double ExecImpl::get_seq_remaining_ratio() { - if (surf_action_) - surf_action_->set_bound(bound); + return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains() / surf_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 + return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains(); } -void simgrid::kernel::activity::ExecImpl::post() +ExecImpl& ExecImpl::set_bound(double bound) { - if (host_ && host_->is_off()) { /* FIXME: handle resource failure for parallel tasks too */ + bound_ = bound; + return *this; +} + +ExecImpl& ExecImpl::set_sharing_penalty(double sharing_penalty) +{ + sharing_penalty_ = sharing_penalty; + return *this; +} + +void ExecImpl::post() +{ + if (hosts_.size() == 1 && not hosts_.front()->is_on()) { /* FIXME: handle resource failure for parallel tasks too */ /* If the host 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) { + state_ = State::FAILED; + } else if (surf_action_ && surf_action_->get_state() == 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; + state_ = State::CANCELED; + } else if (timeout_detector_ && timeout_detector_->get_state() == resource::Action::State::FINISHED) { + state_ = State::TIMEOUT; } else { - state_ = SIMIX_DONE; + state_ = State::DONE; } - on_completion(this); - - if (surf_action_) { - surf_action_->unref(); - surf_action_ = nullptr; - } - if (timeout_detector_) { - timeout_detector_->unref(); - timeout_detector_ = nullptr; + clean_action(); + timeout_detector_.reset(); + if (actor_) { + actor_->activities_.remove(this); + actor_ = nullptr; } - - /* If there are simcalls associated with the synchro, then answer them */ - if (not simcalls_.empty()) - SIMIX_execution_finish(this); + /* Answer all simcalls associated with the synchro */ + finish(); } -simgrid::kernel::activity::ActivityImpl* -simgrid::kernel::activity::ExecImpl::migrate(simgrid::s4u::Host* to) +void ExecImpl::finish() { + while (not simcalls_.empty()) { + smx_simcall_t simcall = simcalls_.front(); + simcalls_.pop_front(); + + /* If a waitany simcall is waiting for this synchro to finish, then remove it from the other synchros in the waitany + * list. Afterwards, get the position of the actual synchro in the waitany list and return it as the result of the + * simcall */ + + if (simcall->call_ == simix::Simcall::NONE) // FIXME: maybe a better way to handle this case + continue; // if process handling comm is killed + if (simcall->call_ == simix::Simcall::EXECUTION_WAITANY_FOR) { + simgrid::kernel::activity::ExecImpl** execs = simcall_execution_waitany_for__get__execs(simcall); + size_t count = simcall_execution_waitany_for__get__count(simcall); + + for (size_t i = 0; i < count; i++) { + // Remove the first occurrence of simcall: + auto* exec = execs[i]; + auto j = boost::range::find(exec->simcalls_, simcall); + if (j != exec->simcalls_.end()) + exec->simcalls_.erase(j); + if (simcall->timeout_cb_) { + simcall->timeout_cb_->remove(); + simcall->timeout_cb_ = nullptr; + } + } + + if (not MC_is_active() && not MC_record_replay_is_active()) { + ExecImpl** element = std::find(execs, execs + count, this); + int rank = (element != execs + count) ? element - execs : -1; + simcall_execution_waitany_for__set__result(simcall, rank); + } + } + + switch (state_) { + case State::DONE: + /* do nothing, synchro done */ + XBT_DEBUG("ExecImpl::finish(): execution successful"); + break; + + case State::FAILED: + XBT_DEBUG("ExecImpl::finish(): host '%s' failed", simcall->issuer_->get_host()->get_cname()); + simcall->issuer_->context_->set_wannadie(); + if (simcall->issuer_->get_host()->is_on()) + simcall->issuer_->exception_ = + std::make_exception_ptr(simgrid::HostFailureException(XBT_THROW_POINT, "Host failed")); + /* else, the actor will be killed with no possibility to survive */ + break; + + case State::CANCELED: + XBT_DEBUG("ExecImpl::finish(): execution canceled"); + simcall->issuer_->exception_ = + std::make_exception_ptr(simgrid::CancelException(XBT_THROW_POINT, "Execution Canceled")); + break; + + case State::TIMEOUT: + XBT_DEBUG("ExecImpl::finish(): execution timeouted"); + simcall->issuer_->exception_ = std::make_exception_ptr(simgrid::TimeoutException(XBT_THROW_POINT, "Timeouted")); + break; + + default: + xbt_die("Internal error in ExecImpl::finish(): unexpected synchro state %d", static_cast(state_)); + } + + 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_->context_->set_wannadie(); + } +} + +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->surf_action_; + resource::Action* new_action = to->pimpl_cpu->execution_start(old_action->get_cost()); 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()); // 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; } - 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_migration; +xbt::signal ExecImpl::on_migration; + +} // namespace activity +} // namespace kernel +} // namespace simgrid