Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
Move hosts_ to private, and deal with consequences
[simgrid.git] / src / kernel / activity / ExecImpl.cpp
index a687d1d..fdaf2dc 100644 (file)
@@ -1,54 +1,51 @@
-/* Copyright (c) 2007-2021. The SimGrid Team. All rights reserved.          */
+/* Copyright (c) 2007-2022. 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 <simgrid/Exception.hpp>
+#include <simgrid/kernel/routing/NetPoint.hpp>
+#include <simgrid/modelchecker.h>
+#include <simgrid/s4u/Engine.hpp>
+
 #include "src/kernel/activity/ExecImpl.hpp"
-#include "simgrid/Exception.hpp"
-#include "simgrid/kernel/routing/NetPoint.hpp"
-#include "simgrid/modelchecker.h"
-#include "simgrid/s4u/Engine.hpp"
-#include "simgrid/s4u/Exec.hpp"
+#include "src/kernel/actor/ActorImpl.hpp"
 #include "src/kernel/actor/SimcallObserver.hpp"
+#include "src/kernel/resource/CpuImpl.hpp"
 #include "src/mc/mc_replay.hpp"
 #include "src/surf/HostImpl.hpp"
-#include "src/surf/cpu_interface.hpp"
-#include "src/surf/surf_interface.hpp"
-
-#include "simgrid/s4u/Host.hpp"
 
-XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(simix_process);
+XBT_LOG_NEW_DEFAULT_SUBCATEGORY(ker_cpu, kernel, "Kernel cpu-related synchronization");
 
-namespace simgrid {
-namespace kernel {
-namespace activity {
+namespace simgrid::kernel::activity {
 
 ExecImpl::ExecImpl()
 {
   piface_                = new s4u::Exec(this);
   actor::ActorImpl* self = actor::ActorImpl::self();
   if (self) {
-    actor_ = self;
+    set_actor(self);
     self->activities_.emplace_back(this);
   }
 }
 
 ExecImpl& ExecImpl::set_host(s4u::Host* host)
 {
-  hosts_.assign(1, host);
+  clear_hosts();
+  add_host(host);
   return *this;
 }
 
 ExecImpl& ExecImpl::set_hosts(const std::vector<s4u::Host*>& hosts)
 {
-  hosts_ = hosts;
+  ActivityImpl::set_hosts(hosts);
   return *this;
 }
 
 ExecImpl& ExecImpl::set_timeout(double 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_.reset(get_host()->get_cpu()->sleep(timeout));
     timeout_detector_->set_activity(this);
   }
   return *this;
@@ -72,41 +69,58 @@ ExecImpl& ExecImpl::set_bytes_amounts(const std::vector<double>& bytes_amounts)
 
   return *this;
 }
+ExecImpl& ExecImpl::set_thread_count(int thread_count)
+{
+  thread_count_ = thread_count;
+
+  return *this;
+}
 
 ExecImpl* ExecImpl::start()
 {
-  state_ = State::RUNNING;
+  set_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_);
-        surf_action_->set_user_bound(bound_);
+    if (get_hosts().size() == 1) {
+      if (thread_count_ == 1) {
+        surf_action_ = get_host()->get_cpu()->execution_start(flops_amounts_.front(), bound_);
+        surf_action_->set_sharing_penalty(sharing_penalty_);
+      } else {
+        auto host_model = get_host()->get_netpoint()->get_englobing_zone()->get_host_model();
+        surf_action_    = host_model->execute_thread(get_host(), flops_amounts_.front(), thread_count_);
       }
+      surf_action_->set_category(get_tracing_category());
     } else {
-      // 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);
+      // 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();
+      surf_action_    = host_model->execute_parallel(get_hosts(), flops_amounts_.data(), bytes_amounts_.data(), -1);
     }
     surf_action_->set_activity(this);
-    start_time_ = surf_action_->get_start_time();
+    set_start_time(surf_action_->get_start_time());
   }
 
   XBT_DEBUG("Create execute synchro %p: %s", this, get_cname());
   return this;
 }
 
+double ExecImpl::get_remaining() const
+{
+  if (get_state() == State::WAITING || get_state() == State::FAILED)
+    return flops_amounts_.front();
+  return ActivityImpl::get_remaining();
+}
+
 double ExecImpl::get_seq_remaining_ratio()
 {
+  if (get_state() == State::WAITING)
+    return 1;
   return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains() / surf_action_->get_cost();
 }
 
 double ExecImpl::get_par_remaining_ratio()
 {
   // parallel task: their remain is already between 0 and 1
+  if (get_state() == State::WAITING)
+    return 1;
   return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains();
 }
 
@@ -122,106 +136,106 @@ ExecImpl& ExecImpl::set_sharing_penalty(double sharing_penalty)
   return *this;
 }
 
+ExecImpl& ExecImpl::update_sharing_penalty(double sharing_penalty)
+{
+  sharing_penalty_ = sharing_penalty;
+  surf_action_->set_sharing_penalty(sharing_penalty);
+  return *this;
+}
+
 void ExecImpl::post()
 {
   xbt_assert(surf_action_ != nullptr);
-  if (std::any_of(hosts_.begin(), hosts_.end(), [](const s4u::Host* host) { return not host->is_on(); })) {
+  auto hosts=get_hosts();
+  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;
+    set_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;
+    set_state(State::CANCELED);
   } else if (timeout_detector_ && timeout_detector_->get_state() == resource::Action::State::FINISHED) {
     if (surf_action_->get_remains() > 0.0) {
       surf_action_->set_state(resource::Action::State::FAILED);
-      state_ = State::TIMEOUT;
+      set_state(State::TIMEOUT);
     } else {
-      state_ = State::DONE;
+      set_state(State::DONE);
     }
   } else {
-    state_ = State::DONE;
+    set_state(State::DONE);
   }
 
-  finish_time_ = surf_action_->get_finish_time();
-
   clean_action();
   timeout_detector_.reset();
-  if (actor_) {
-    actor_->activities_.remove(this);
-    actor_ = nullptr;
+  if (get_actor() != nullptr) {
+    get_actor()->activities_.remove(this);
   }
+  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<s4u::Exec*>(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", to_c_str(state_));
+  XBT_DEBUG("ExecImpl::finish() in state %s", get_state_str());
   while (not simcalls_.empty()) {
-    smx_simcall_t simcall = simcalls_.front();
+    actor::Simcall* 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 (auto* observer =
-            dynamic_cast<kernel::actor::ExecutionWaitanySimcall*>(simcall->observer_)) { // simcall is a wait_any?
-      const auto& execs = observer->get_execs();
-
-      for (auto* exec : execs) {
-        exec->unregister_simcall(simcall);
+    if (simcall->call_ == actor::Simcall::Type::NONE) // FIXME: maybe a better way to handle this case
+      continue;                                       // if process handling comm is killed
 
-        if (simcall->timeout_cb_) {
-          simcall->timeout_cb_->remove();
-          simcall->timeout_cb_ = nullptr;
-        }
-      }
+    handle_activity_waitany(simcall);
 
-      if (not MC_is_active() && not MC_record_replay_is_active()) {
-        auto element = std::find(execs.begin(), execs.end(), this);
-        int rank     = element != execs.end() ? static_cast<int>(std::distance(execs.begin(), element)) : -1;
-        observer->set_result(rank);
-      }
-    }
-    switch (state_) {
-      case State::FAILED:
-        simcall->issuer_->context_->set_wannadie();
-        if (simcall->issuer_->get_host()->is_on())
-          simcall->issuer_->exception_ = std::make_exception_ptr(HostFailureException(XBT_THROW_POINT, "Host failed"));
-        /* else, the actor will be killed with no possibility to survive */
-        break;
-
-      case State::CANCELED:
-        simcall->issuer_->exception_ = std::make_exception_ptr(CancelException(XBT_THROW_POINT, "Execution Canceled"));
-        break;
-
-      case State::TIMEOUT:
-        simcall->issuer_->exception_ = std::make_exception_ptr(TimeoutException(XBT_THROW_POINT, "Timeouted"));
-        break;
-
-      default:
-        xbt_assert(state_ == State::DONE, "Internal error in ExecImpl::finish(): unexpected synchro state %s",
-                   to_c_str(state_));
-    }
+    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_->context_->set_wannadie();
+      simcall->issuer_->set_wannadie();
   }
 }
 
+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()) {
     resource::Action* old_action = this->surf_action_;
-    resource::Action* new_action = to->pimpl_cpu->execution_start(old_action->get_cost());
+    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_activity(this);
     new_action->set_sharing_penalty(old_action->get_sharing_penalty());
@@ -237,37 +251,9 @@ ActivityImpl* ExecImpl::migrate(s4u::Host* to)
   return this;
 }
 
-void ExecImpl::wait_any_for(actor::ActorImpl* issuer, const std::vector<ExecImpl*>& execs, double timeout)
-{
-  if (timeout < 0.0) {
-    issuer->simcall_.timeout_cb_ = nullptr;
-  } else {
-    issuer->simcall_.timeout_cb_ = timer::Timer::set(s4u::Engine::get_clock() + timeout, [issuer, &execs]() {
-      issuer->simcall_.timeout_cb_ = nullptr;
-      for (auto* exec : execs)
-        exec->unregister_simcall(&issuer->simcall_);
-      // default result (-1) is set in actor::ExecutionWaitanySimcall
-      issuer->simcall_answer();
-    });
-  }
-
-  for (auto* exec : execs) {
-    /* associate this simcall to the the synchro */
-    exec->simcalls_.push_back(&issuer->simcall_);
-
-    /* see if the synchro is already finished */
-    if (exec->state_ != State::WAITING && exec->state_ != State::RUNNING) {
-      exec->finish();
-      break;
-    }
-  }
-}
-
 /*************
  * Callbacks *
  *************/
 xbt::signal<void(ExecImpl const&, s4u::Host*)> ExecImpl::on_migration;
 
-} // namespace activity
-} // namespace kernel
-} // namespace simgrid
+} // namespace simgrid::kernel::activity