Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
Restarting or killing maestro does not seem to be a cleaver idea
[simgrid.git] / src / kernel / actor / ActorImpl.cpp
index 9e386fa..4569bad 100644 (file)
@@ -133,11 +133,40 @@ void ActorImpl::detach()
 
 void ActorImpl::cleanup()
 {
+  finished_ = true;
+
+  if (has_to_auto_restart() && not get_host()->is_on()) {
+    XBT_DEBUG("Insert host %s to watched_hosts because it's off and %s needs to restart", get_host()->get_cname(),
+              get_cname());
+    watched_hosts.insert(get_host()->get_name());
+  }
+
+  // Execute the termination callbacks
+  smx_process_exit_status_t exit_status = (context_->iwannadie) ? SMX_EXIT_FAILURE : SMX_EXIT_SUCCESS;
+  while (not on_exit.empty()) {
+    s_smx_process_exit_fun_t exit_fun = on_exit.back();
+    on_exit.pop_back();
+    (exit_fun.fun)(exit_status, exit_fun.arg);
+  }
+
+  /* cancel non-blocking activities */
+  for (auto activity : comms)
+    boost::static_pointer_cast<activity::CommImpl>(activity)->cancel();
+  comms.clear();
+
+  XBT_DEBUG("%s@%s(%ld) should not run anymore", get_cname(), get_host()->get_cname(), get_pid());
+
   if (this == simix_global->maestro_process) /* Do not cleanup maestro */
     return;
 
   XBT_DEBUG("Cleanup actor %s (%p), waiting synchro %p", get_cname(), this, waiting_synchro.get());
 
+  /* Unregister from the kill timer if any */
+  if (kill_timer != nullptr) {
+    kill_timer->remove();
+    kill_timer = nullptr;
+  }
+
   simix_global->mutex.lock();
 
   simix_global->process_list.erase(pid_);
@@ -145,13 +174,16 @@ void ActorImpl::cleanup()
     simgrid::xbt::intrusive_erase(host_->pimpl_->process_list_, *this);
   if (not smx_destroy_list_hook.is_linked()) {
 #if SIMGRID_HAVE_MC
-    xbt_dynar_push_as(simix_global->dead_actors_vector, smx_actor_t, this);
+    xbt_dynar_push_as(simix_global->dead_actors_vector, ActorImpl*, this);
 #endif
     simix_global->actors_to_destroy.push_back(*this);
   }
-  context_->iwannadie = false;
 
   simix_global->mutex.unlock();
+
+  context_->iwannadie = false; // don't let the simcall's yield() do a Context::stop(), to avoid infinite loops
+  simgrid::simix::simcall([this] { simgrid::s4u::Actor::on_destruction(iface()); });
+  context_->iwannadie = true;
 }
 
 void ActorImpl::exit()
@@ -309,13 +341,15 @@ void ActorImpl::daemonize()
 
 s4u::Actor* ActorImpl::restart()
 {
+  xbt_assert(this != simix_global->maestro_process, "Restarting maestro is not supported");
+
   XBT_DEBUG("Restarting actor %s on %s", get_cname(), host_->get_cname());
 
   // retrieve the arguments of the old actor
   ProcessArg arg = ProcessArg(host_, this);
 
   // kill the old actor
-  (this == simix_global->maestro_process) ? this->exit() : SIMIX_process_self()->kill(this);
+  context::Context::self()->get_actor()->kill(this);
 
   // start the new actor
   ActorImplPtr actor =
@@ -367,15 +401,15 @@ void ActorImpl::resume()
   XBT_OUT();
 }
 
-activity::ActivityImplPtr ActorImpl::join(smx_actor_t actor, double timeout)
+activity::ActivityImplPtr ActorImpl::join(ActorImpl* actor, double timeout)
 {
   activity::ActivityImplPtr res = this->sleep(timeout);
   intrusive_ptr_add_ref(res.get());
   SIMIX_process_on_exit(actor,
                         [](int, void* arg) {
-                          auto sleep = static_cast<simgrid::kernel::activity::SleepImpl*>(arg);
+                          auto sleep = static_cast<activity::SleepImpl*>(arg);
                           if (sleep->surf_action_)
-                            sleep->surf_action_->finish(simgrid::kernel::resource::Action::State::FINISHED);
+                            sleep->surf_action_->finish(resource::Action::State::FINISHED);
                           intrusive_ptr_release(sleep);
                         },
                         res.get());
@@ -388,8 +422,11 @@ activity::ActivityImplPtr ActorImpl::sleep(double duration)
     throw_exception(std::make_exception_ptr(simgrid::HostFailureException(
         XBT_THROW_POINT, std::string("Host ") + host_->get_cname() + " failed, you cannot sleep there.")));
 
-  return simgrid::kernel::activity::SleepImplPtr(new simgrid::kernel::activity::SleepImpl("sleep", host_))
-      ->start(duration);
+  return activity::SleepImplPtr(new activity::SleepImpl())
+      ->set_name("sleep")
+      ->set_host(host_)
+      ->set_duration(duration)
+      ->start();
 }
 
 void ActorImpl::throw_exception(std::exception_ptr e)
@@ -414,7 +451,7 @@ void ActorImpl::throw_exception(std::exception_ptr e)
 
     activity::SleepImplPtr sleep = boost::dynamic_pointer_cast<activity::SleepImpl>(waiting_synchro);
     if (sleep != nullptr) {
-      SIMIX_process_sleep_destroy(waiting_synchro);
+      SIMIX_process_sleep_destroy(sleep);
       if (std::find(begin(simix_global->actors_to_run), end(simix_global->actors_to_run), this) ==
               end(simix_global->actors_to_run) &&
           this != SIMIX_process_self()) {
@@ -438,14 +475,14 @@ void ActorImpl::throw_exception(std::exception_ptr e)
 
 void ActorImpl::set_host(s4u::Host* dest)
 {
-  simgrid::xbt::intrusive_erase(host_->pimpl_->process_list_, *this);
+  xbt::intrusive_erase(host_->pimpl_->process_list_, *this);
   host_ = dest;
   dest->pimpl_->process_list_.push_back(*this);
 }
 
 ActorImplPtr ActorImpl::init(const std::string& name, s4u::Host* host)
 {
-  ActorImpl* actor = new ActorImpl(simgrid::xbt::string(name), host);
+  ActorImpl* actor = new ActorImpl(xbt::string(name), host);
   actor->set_ppid(this->pid_);
 
   intrusive_ptr_add_ref(actor);
@@ -490,9 +527,9 @@ ActorImplPtr ActorImpl::create(const std::string& name, const simix::ActorCode&
 
   ActorImplPtr actor;
   if (parent_actor != nullptr)
-    actor = parent_actor->init(simgrid::xbt::string(name), host);
+    actor = parent_actor->init(xbt::string(name), host);
   else
-    actor = SIMIX_process_self()->init(simgrid::xbt::string(name), host);
+    actor = SIMIX_process_self()->init(xbt::string(name), host);
 
   /* actor data */
   actor->set_user_data(data);
@@ -565,7 +602,7 @@ void SIMIX_process_throw(smx_actor_t actor, xbt_errcat_t cat, int value, const c
     simgrid::kernel::activity::SleepImplPtr sleep =
         boost::dynamic_pointer_cast<simgrid::kernel::activity::SleepImpl>(actor->waiting_synchro);
     if (sleep != nullptr) {
-      SIMIX_process_sleep_destroy(actor->waiting_synchro);
+      SIMIX_process_sleep_destroy(sleep);
       if (std::find(begin(simix_global->actors_to_run), end(simix_global->actors_to_run), actor) ==
               end(simix_global->actors_to_run) &&
           actor != SIMIX_process_self()) {
@@ -666,11 +703,9 @@ void simcall_HANDLER_process_sleep(smx_simcall_t simcall, double duration)
   simcall->issuer->waiting_synchro = sync;
 }
 
-void SIMIX_process_sleep_destroy(smx_activity_t synchro)
+void SIMIX_process_sleep_destroy(simgrid::kernel::activity::SleepImplPtr sleep)
 {
-  XBT_DEBUG("Destroy sleep synchro %p", synchro.get());
-  simgrid::kernel::activity::SleepImplPtr sleep =
-      boost::static_pointer_cast<simgrid::kernel::activity::SleepImpl>(synchro);
+  XBT_DEBUG("Destroy sleep synchro %p", sleep.get());
 
   if (sleep->surf_action_) {
     sleep->surf_action_->unref();