-/* 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"
XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(simix_process);
-void simcall_HANDLER_execution_wait(smx_simcall_t simcall, simgrid::kernel::activity::ExecImpl* synchro, double timeout)
-{
- XBT_DEBUG("Wait for execution of synchro %p, state %d", synchro, (int)synchro->state_);
- xbt_assert(std::isfinite(timeout), "timeout is not finite!");
-
- /* Associate this simcall to the synchro */
- synchro->register_simcall(simcall);
-
- /* set surf's synchro */
- if (MC_is_active() || MC_record_replay_is_active()) {
- int idx = SIMCALL_GET_MC_VALUE(*simcall);
- if (idx == 0) {
- synchro->state_ = simgrid::kernel::activity::State::DONE;
- } else {
- /* If we reached this point, the wait simcall must have a timeout */
- /* Otherwise it shouldn't be enabled and executed by the MC */
- if (timeout < 0.0)
- THROW_IMPOSSIBLE;
- synchro->state_ = simgrid::kernel::activity::State::TIMEOUT;
- }
- synchro->finish();
- return;
- }
-
- /* If the synchro is already finished then perform the error handling */
- if (synchro->state_ != simgrid::kernel::activity::State::RUNNING) {
- synchro->finish();
- } else { /* we need a sleep action (even when there is no timeout) to be notified of host failures */
- synchro->set_timeout(timeout);
- }
-}
-
-void simcall_HANDLER_execution_test(smx_simcall_t simcall, simgrid::kernel::activity::ExecImpl* synchro)
-{
- bool res = (synchro->state_ != simgrid::kernel::activity::State::WAITING &&
- synchro->state_ != simgrid::kernel::activity::State::RUNNING);
- if (res) {
- synchro->simcalls_.push_back(simcall);
- synchro->finish();
- } else {
- simcall->issuer_->simcall_answer();
- }
- simcall_execution_test__set__result(simcall, res);
-}
-
void simcall_HANDLER_execution_waitany_for(smx_simcall_t simcall, simgrid::kernel::activity::ExecImpl* execs[],
size_t count, double timeout)
{
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];
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;
}
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;
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;
}
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);
}
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_ = 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 */
+ } 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_ = State::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_ = State::DONE;
}
- clean_action();
+ get_iface()->set_finish_time(surf_action_->get_finish_time());
- if (timeout_detector_) {
- timeout_detector_->unref();
- timeout_detector_ = nullptr;
+ clean_action();
+ timeout_detector_.reset();
+ if (actor_) {
+ actor_->activities_.remove(this);
+ actor_ = nullptr;
}
-
/* Answer all simcalls associated with the synchro */
finish();
}
* 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_ == SIMCALL_NONE) // FIXME: maybe a better way to handle this case
- continue; // if process handling comm is killed
- if (simcall->call_ == SIMCALL_EXECUTION_WAITANY_FOR) {
+ 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);
simcall_execution_waitany_for__set__result(simcall, rank);
}
}
-
switch (state_) {
-
case State::DONE:
/* do nothing, synchro done */
XBT_DEBUG("ExecImpl::finish(): execution successful");
case State::FAILED:
XBT_DEBUG("ExecImpl::finish(): host '%s' failed", simcall->issuer_->get_host()->get_cname());
- simcall->issuer_->context_->iwannadie = true;
+ 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"));
xbt_die("Internal error in ExecImpl::finish(): unexpected synchro state %d", static_cast<int>(state_));
}
- simcall->issuer_->waiting_synchro = nullptr;
+ 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_->iwannadie = true;
+ simcall->issuer_->context_->set_wannadie();
}
}
new_action->set_remains(old_action->get_remains());
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.
+ new_action->set_user_bound(old_action->get_user_bound());
old_action->set_activity(nullptr);
old_action->cancel();
