X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/cdef428603e68472f5999bdcc382748f8eb5b7de..d78a7f87e28a15c0f1e71e4510a055554e0e5e9b:/src/kernel/activity/ExecImpl.cpp diff --git a/src/kernel/activity/ExecImpl.cpp b/src/kernel/activity/ExecImpl.cpp index a2ca6a8f6e..0f925c9600 100644 --- a/src/kernel/activity/ExecImpl.cpp +++ b/src/kernel/activity/ExecImpl.cpp @@ -1,11 +1,13 @@ -/* Copyright (c) 2007-2019. 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/kernel/routing/NetPoint.hpp" #include "simgrid/modelchecker.h" +#include "simgrid/s4u/Exec.hpp" #include "src/mc/mc_replay.hpp" #include "src/surf/HostImpl.hpp" #include "src/surf/cpu_interface.hpp" @@ -13,56 +15,61 @@ #include "simgrid/s4u/Host.hpp" +#include + XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(simix_process); -void simcall_HANDLER_execution_wait(smx_simcall_t simcall, simgrid::kernel::activity::ExecImpl* synchro) +void simcall_HANDLER_execution_waitany_for(smx_simcall_t simcall, simgrid::kernel::activity::ExecImpl* execs[], + size_t count, double timeout) { - XBT_DEBUG("Wait for execution of synchro %p, state %d", synchro, (int)synchro->state_); - - /* Associate this simcall to the synchro */ - synchro->simcalls_.push_back(simcall); - simcall->issuer->waiting_synchro = synchro; - - /* set surf's synchro */ - if (MC_is_active() || MC_record_replay_is_active()) { - synchro->state_ = SIMIX_DONE; - synchro->finish(); - return; + 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(); + }); } - /* If the synchro is already finished then perform the error handling */ - if (synchro->state_ != SIMIX_RUNNING) - synchro->finish(); -} + for (size_t i = 0; i < count; i++) { + /* associate this simcall to the the synchro */ + auto* exec = execs[i]; + exec->simcalls_.push_back(simcall); -void simcall_HANDLER_execution_test(smx_simcall_t simcall, simgrid::kernel::activity::ExecImpl* synchro) -{ - bool res = (synchro->state_ != SIMIX_WAITING && synchro->state_ != SIMIX_RUNNING); - if (res) { - synchro->simcalls_.push_back(simcall); - synchro->finish(); - } else { - SIMIX_simcall_answer(simcall); + /* 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; + } } - simcall_execution_test__set__result(simcall, res); } namespace simgrid { namespace kernel { namespace activity { -ExecImpl::~ExecImpl() +ExecImpl::ExecImpl() { - if (timeout_detector_) - timeout_detector_->unref(); - XBT_DEBUG("Destroy exec %p", this); + piface_ = new s4u::Exec(this); + actor::ActorImpl* self = actor::ActorImpl::self(); + if (self) { + actor_ = self; + self->activities_.emplace_back(this); + } } ExecImpl& ExecImpl::set_host(s4u::Host* host) { - if (not hosts_.empty()) - hosts_.clear(); - hosts_.push_back(host); + hosts_.assign(1, host); return *this; } @@ -74,8 +81,8 @@ ExecImpl& ExecImpl::set_hosts(const std::vector& hosts) ExecImpl& ExecImpl::set_timeout(double timeout) { - if (timeout > 0 && not MC_is_active() && not MC_record_replay_is_active()) { - timeout_detector_ = hosts_.front()->pimpl_cpu->sleep(timeout); + 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; @@ -83,9 +90,7 @@ ExecImpl& ExecImpl::set_timeout(double timeout) ExecImpl& ExecImpl::set_flops_amount(double flops_amount) { - if (not flops_amounts_.empty()) - flops_amounts_.clear(); - flops_amounts_.push_back(flops_amount); + flops_amounts_.assign(1, flops_amount); return *this; } @@ -104,23 +109,26 @@ ExecImpl& ExecImpl::set_bytes_amounts(const std::vector& bytes_amounts) ExecImpl* ExecImpl::start() { - state_ = SIMIX_RUNNING; + 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_priority(priority_); + surf_action_->set_sharing_penalty(sharing_penalty_); surf_action_->set_category(get_tracing_category()); - if (bound_ > 0) + if (bound_ > 0) { surf_action_->set_bound(bound_); + surf_action_->set_user_bound(bound_); + } } else { - surf_action_ = surf_host_model->execute_parallel(hosts_, flops_amounts_.data(), bytes_amounts_.data(), -1); + // FIXME[donassolo]: verify if all hosts belongs to the same netZone? + auto host_model = hosts_.front()->get_netpoint()->get_englobing_zone()->get_host_model(); + surf_action_ = 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()); - ExecImpl::on_creation(*this); return this; } @@ -141,39 +149,43 @@ ExecImpl& ExecImpl::set_bound(double bound) return *this; } -ExecImpl& ExecImpl::set_priority(double priority) +ExecImpl& ExecImpl::set_sharing_penalty(double sharing_penalty) { - priority_ = priority; + 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 + xbt_assert(surf_action_ != nullptr); + if (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() == resource::Action::State::FAILED) { - /* If the host running the synchro didn't fail, then the synchro was canceled */ - state_ = SIMIX_CANCELED; + state_ = State::FAILED; + } else if (surf_action_->get_state() == resource::Action::State::FAILED) { + /* If all the hosts are running the synchro didn't fail, then the synchro was canceled */ + state_ = State::CANCELED; } else if (timeout_detector_ && timeout_detector_->get_state() == resource::Action::State::FINISHED) { - state_ = SIMIX_TIMEOUT; + if (surf_action_->get_remains() > 0.0) { + surf_action_->set_state(resource::Action::State::FAILED); + state_ = State::TIMEOUT; + } else { + state_ = State::DONE; + } } else { - state_ = SIMIX_DONE; + state_ = State::DONE; } - on_completion(*this); + get_iface()->set_finish_time(surf_action_->get_finish_time()); clean_action(); - - if (timeout_detector_) { - timeout_detector_->unref(); - timeout_detector_ = nullptr; + 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()) - finish(); + /* Answer all simcalls associated with the synchro */ + finish(); } void ExecImpl::finish() @@ -181,45 +193,72 @@ void ExecImpl::finish() while (not simcalls_.empty()) { smx_simcall_t simcall = simcalls_.front(); simcalls_.pop_front(); - switch (state_) { - case SIMIX_DONE: + /* 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 SIMIX_FAILED: - XBT_DEBUG("ExecImpl::finish(): host '%s' failed", simcall->issuer->get_host()->get_cname()); - simcall->issuer->context_->iwannadie = true; - if (simcall->issuer->get_host()->is_on()) - simcall->issuer->exception_ = + 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 SIMIX_CANCELED: + case State::CANCELED: XBT_DEBUG("ExecImpl::finish(): execution canceled"); - simcall->issuer->exception_ = + simcall->issuer_->exception_ = std::make_exception_ptr(simgrid::CancelException(XBT_THROW_POINT, "Execution Canceled")); break; - case SIMIX_TIMEOUT: + case State::TIMEOUT: XBT_DEBUG("ExecImpl::finish(): execution timeouted"); - simcall->issuer->exception_ = std::make_exception_ptr(simgrid::TimeoutError(XBT_THROW_POINT, "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; - simcall_execution_wait__set__result(simcall, state_); - + simcall->issuer_->waiting_synchro_ = nullptr; /* Fail the process if the host is down */ - if (simcall->issuer->get_host()->is_on()) - SIMIX_simcall_answer(simcall); + if (simcall->issuer_->get_host()->is_on()) + simcall->issuer_->simcall_answer(); else - simcall->issuer->context_->iwannadie = true; + simcall->issuer_->context_->set_wannadie(); } } @@ -230,10 +269,8 @@ ActivityImpl* ExecImpl::migrate(s4u::Host* to) resource::Action* new_action = to->pimpl_cpu->execution_start(old_action->get_cost()); new_action->set_remains(old_action->get_remains()); new_action->set_activity(this); - new_action->set_priority(old_action->get_priority()); - - // 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. + new_action->set_sharing_penalty(old_action->get_sharing_penalty()); + new_action->set_user_bound(old_action->get_user_bound()); old_action->set_activity(nullptr); old_action->cancel(); @@ -248,8 +285,6 @@ ActivityImpl* ExecImpl::migrate(s4u::Host* to) /************* * Callbacks * *************/ -xbt::signal ExecImpl::on_creation; -xbt::signal ExecImpl::on_completion; xbt::signal ExecImpl::on_migration; } // namespace activity