* under the terms of the license (GNU LGPL) which comes with this package. */
#include "mc/mc.h"
+#include "simgrid/Exception.hpp"
#include "smx_private.hpp"
#include "src/kernel/activity/CommImpl.hpp"
#include "src/kernel/activity/ExecImpl.hpp"
#include "src/simix/smx_synchro_private.hpp"
#include "src/surf/HostImpl.hpp"
#include "src/surf/cpu_interface.hpp"
-#include "xbt/ex.hpp"
#ifdef HAVE_SMPI
#include "src/smpi/include/private.hpp"
unsigned long simix_process_maxpid = 0;
/**
- * \brief Returns the current agent.
+ * @brief Returns the current agent.
*
* This functions returns the currently running SIMIX process.
*
- * \return The SIMIX process
+ * @return The SIMIX process
*/
smx_actor_t SIMIX_process_self()
{
}
/**
- * \brief Returns whether a process has pending asynchronous communications.
- * \return true if there are asynchronous communications in this process
+ * @brief Returns whether a process has pending asynchronous communications.
+ * @return true if there are asynchronous communications in this process
*/
int SIMIX_process_has_pending_comms(smx_actor_t process) {
}
/**
- * \brief Moves a process to the list of processes to destroy.
+ * @brief Moves a process to the list of processes to destroy.
*/
void SIMIX_process_cleanup(smx_actor_t process)
{
process->kill_timer = nullptr;
}
- xbt_os_mutex_acquire(simix_global->mutex);
+ simix_global->mutex.lock();
/* cancel non-blocking communications */
while (not process->comms.empty()) {
comm->cancel();
}
- XBT_DEBUG("%p should not be run anymore",process);
+ XBT_DEBUG("%s@%s(%ld) should not run anymore", process->get_cname(), process->iface()->get_host()->get_cname(),
+ process->pid_);
simix_global->process_list.erase(process->pid_);
if (process->host_ && process->host_process_list_hook.is_linked())
simgrid::xbt::intrusive_erase(process->host_->pimpl_->process_list_, *process);
#endif
simix_global->process_to_destroy.push_back(*process);
}
- process->context_->iwannadie = 0;
+ process->context_->iwannadie = false;
- xbt_os_mutex_release(simix_global->mutex);
+ simix_global->mutex.unlock();
}
/**
SIMIX_process_kill(this, (this == simix_global->maestro_process) ? this : SIMIX_process_self());
// start the new process
- ActorImpl* actor = simix_global->create_process_function(arg.name.c_str(), std::move(arg.code), arg.data, arg.host,
- arg.properties.get(), nullptr);
+ ActorImpl* actor =
+ SIMIX_process_create(arg.name.c_str(), std::move(arg.code), arg.data, arg.host, arg.properties.get(), nullptr);
simcall_process_set_kill_time(actor, arg.kill_time);
actor->set_auto_restart(arg.auto_restart);
return nullptr;
}
- suspended_ = 1;
+ suspended_ = true;
/* If we are suspending another actor that is waiting on a sync, suspend its synchronization. */
if (this != issuer) {
return nullptr;
} else {
- return SIMIX_execution_start("suspend", 0.0, 1.0, 0.0, this->host_);
+ return SIMIX_execution_start("suspend", "", 0.0, 1.0, 0.0, this->host_);
}
}
if (not suspended_)
return;
- suspended_ = 0;
+ suspended_ = false;
/* resume the synchronization that was blocking the resumed actor. */
if (waiting_synchro)
smx_activity_t ActorImpl::sleep(double duration)
{
if (host_->is_off())
- THROWF(host_error, 0, "Host %s failed, you cannot sleep there.", host_->get_cname());
+ throw_exception(std::make_exception_ptr(simgrid::HostFailureException(
+ XBT_THROW_POINT, std::string("Host ") + std::string(host_->get_cname()) + " failed, you cannot sleep there.")));
simgrid::kernel::activity::SleepImpl* synchro = new simgrid::kernel::activity::SleepImpl();
synchro->host = host_;
return synchro;
}
+void ActorImpl::throw_exception(std::exception_ptr e)
+{
+ exception = e;
+
+ if (suspended_)
+ resume();
+
+ /* cancel the blocking synchro if any */
+ if (waiting_synchro) {
+
+ simgrid::kernel::activity::ExecImplPtr exec =
+ boost::dynamic_pointer_cast<simgrid::kernel::activity::ExecImpl>(waiting_synchro);
+ if (exec != nullptr && exec->surf_action_)
+ exec->surf_action_->cancel();
+
+ simgrid::kernel::activity::CommImplPtr comm =
+ boost::dynamic_pointer_cast<simgrid::kernel::activity::CommImpl>(waiting_synchro);
+ if (comm != nullptr) {
+ comms.remove(comm);
+ comm->cancel();
+ }
+
+ simgrid::kernel::activity::SleepImplPtr sleep =
+ boost::dynamic_pointer_cast<simgrid::kernel::activity::SleepImpl>(waiting_synchro);
+ if (sleep != nullptr) {
+ SIMIX_process_sleep_destroy(waiting_synchro);
+ if (std::find(begin(simix_global->process_to_run), end(simix_global->process_to_run), this) ==
+ end(simix_global->process_to_run) &&
+ this != SIMIX_process_self()) {
+ XBT_DEBUG("Inserting %s in the to_run list", get_cname());
+ simix_global->process_to_run.push_back(this);
+ }
+ }
+
+ simgrid::kernel::activity::RawImplPtr raw =
+ boost::dynamic_pointer_cast<simgrid::kernel::activity::RawImpl>(waiting_synchro);
+ if (raw != nullptr) {
+ SIMIX_synchro_stop_waiting(this, &simcall);
+ }
+
+ simgrid::kernel::activity::IoImplPtr io =
+ boost::dynamic_pointer_cast<simgrid::kernel::activity::IoImpl>(waiting_synchro);
+ if (io != nullptr) {
+ delete io.get();
+ }
+ }
+ waiting_synchro = nullptr;
+}
+
void create_maestro(simgrid::simix::ActorCode code)
{
smx_actor_t maestro = nullptr;
}
/**
- * \brief Internal function to create a process.
+ * @brief Internal function to create a process.
*
* This function actually creates the process.
* It may be called when a SIMCALL_PROCESS_CREATE simcall occurs,
* or directly for SIMIX internal purposes. The sure thing is that it's called from maestro context.
*
- * \return the process created
+ * @return the process created
*/
smx_actor_t SIMIX_process_create(std::string name, simgrid::simix::ActorCode code, void* data, simgrid::s4u::Host* host,
std::unordered_map<std::string, std::string>* properties, smx_actor_t parent_process)
process->code = code;
XBT_VERB("Create context %s", process->get_cname());
- process->context_ = SIMIX_context_new(std::move(code), simix_global->cleanup_process_function, process);
+ process->context_ = SIMIX_context_new(std::move(code), &SIMIX_process_cleanup, process);
/* Add properties */
if (properties != nullptr)
XBT_VERB("Create context %s", process->get_cname());
if (not simix_global)
xbt_die("simix is not initialized, please call MSG_init first");
- process->context_ = simix_global->context_factory->attach(simix_global->cleanup_process_function, process);
+ process->context_ = simix_global->context_factory->attach(&SIMIX_process_cleanup, process);
/* Add properties */
if (properties != nullptr)
xbt_die("Not a suitable context");
auto process = context->process();
- simix_global->cleanup_process_function(process);
+ SIMIX_process_cleanup(process);
context->attach_stop();
}
/**
- * \brief Executes the processes from simix_global->process_to_run.
+ * @brief Executes the processes from simix_global->process_to_run.
*
* The processes of simix_global->process_to_run are run (in parallel if
* possible). On exit, simix_global->process_to_run is empty, and
}
/**
- * \brief Internal function to kill a SIMIX process.
+ * @brief Internal function to kill a SIMIX process.
*
* This function may be called when a SIMCALL_PROCESS_KILL simcall occurs,
* or directly for SIMIX internal purposes.
*
- * \param process poor victim
- * \param issuer the process which has sent the PROCESS_KILL. Important to not schedule twice the same process.
+ * @param process poor victim
+ * @param issuer the process which has sent the PROCESS_KILL. Important to not schedule twice the same process.
*/
void SIMIX_process_kill(smx_actor_t process, smx_actor_t issuer) {
(issuer->host_ == nullptr ? "(null)" : issuer->host_->get_cname()), process->get_cname(),
process->host_->get_cname());
- process->context_->iwannadie = 1;
- process->blocked_ = 0;
- process->suspended_ = 0;
+ process->context_->iwannadie = true;
+ process->blocked_ = false;
+ process->suspended_ = false;
process->exception = nullptr;
+ // Forcefully kill the actor if its host is turned off. Not an HostFailureException because you should not survive that
+ if (process->host_->is_off()) {
+ /* HORRIBLE HACK: Don't throw an StopRequest exception in Java, because it breaks sometimes.
+ *
+ * It seems to break for the actors started from the Java world, with new Process()
+ * while it works for the ones started from the C world, with the deployment file.
+ * When it happens, the simulation stops brutally with a message "untrapped exception StopRequest".
+ *
+ * From what I understand, it works for the native actors because they have a nice try/catch block around their main
+ * but I fail to have something like that for pure Java actors. That's probably a story of C->Java vs Java->C
+ * calling conventions. The right solution may be to have try/catch(StopRequest) blocks around each native call in
+ * JNI. ie, protect every Java->C++ call from C++ exceptions. But this sounds long and painful to do before we
+ * switch to an automatic generator such as SWIG. For now, we don't throw here that exception that we sometimes fail
+ * to catch.
+ *
+ * One of the unfortunate outcome is that the threads started from the deployment file are not stopped anymore.
+ * Or maybe this is the actors stopping gracefully as opposed to the killed ones? Or maybe this is absolutely all
+ * actors of the Java simulation? I'm not sure. Anyway. Because of them, the simulation hangs at the end, waiting
+ * for them to stop but they won't. The current answer to that is very brutal:
+ * we do a "exit(0)" to kill the JVM from the C code after the call to MSG_run(). Definitely unpleasant.
+ */
+
+ if (simgrid::kernel::context::factory_initializer == nullptr) // Only Java sets a factory_initializer, for now
+ process->throw_exception(std::make_exception_ptr(simgrid::kernel::context::Context::StopRequest("Host failed")));
+ }
+
/* destroy the blocking synchro if any */
if (process->waiting_synchro != nullptr) {
SIMIX_synchro_stop_waiting(process, &process->simcall);
} else if (io != nullptr) {
- SIMIX_io_destroy(process->waiting_synchro);
+ delete io.get();
} else {
- xbt_die("Unknown type of activity");
+ simgrid::kernel::activity::ActivityImplPtr activity = process->waiting_synchro;
+ xbt_die("Activity %s is of unknown type %s", activity->name_.c_str(),
+ simgrid::xbt::demangle(typeid(activity).name()).get());
}
process->waiting_synchro = nullptr;
}
}
-/** @brief Ask another process to raise the given exception
- *
- * @param process The process that should raise that exception
- * @param cat category of exception
- * @param value value associated to the exception
- * @param msg string information associated to the exception
- */
+/** @deprecated When this function gets removed, also remove the xbt_ex class, that is only there to help users to
+ * transition */
void SIMIX_process_throw(smx_actor_t process, xbt_errcat_t cat, int value, const char *msg) {
SMX_EXCEPTION(process, cat, value, msg);
simgrid::kernel::activity::IoImplPtr io =
boost::dynamic_pointer_cast<simgrid::kernel::activity::IoImpl>(process->waiting_synchro);
if (io != nullptr) {
- SIMIX_io_destroy(process->waiting_synchro);
+ delete io.get();
}
}
process->waiting_synchro = nullptr;
}
/**
- * \brief Kills all running processes.
- * \param issuer this one will not be killed
+ * @brief Kills all running processes.
+ * @param issuer this one will not be killed
*/
void SIMIX_process_killall(smx_actor_t issuer)
{
}
/**
- * \brief Calling this function makes the process to yield.
+ * @brief Calling this function makes the process to yield.
*
* Only the current process can call this function, giving back the control to maestro.
*
- * \param self the current process
+ * @param self the current process
*/
void SIMIX_process_yield(smx_actor_t self)
{
self->finished_ = true;
/* execute the on_exit functions */
SIMIX_process_on_exit_runall(self);
- /* Add the process to the list of process to restart, only if the host is down */
- if (self->auto_restart_ && self->host_->is_off()) {
- SIMIX_host_add_auto_restart_process(self->host_, self);
+
+ if (self->auto_restart_ && self->host_->is_off() &&
+ watched_hosts.find(self->host_->get_cname()) == watched_hosts.end()) {
+ XBT_DEBUG("Push host %s to watched_hosts because it's off and %s needs to restart", self->host_->get_cname(),
+ self->get_cname());
+ watched_hosts.insert(self->host_->get_cname());
}
+
XBT_DEBUG("Process %s@%s is dead", self->get_cname(), self->host_->get_cname());
self->context_->stop();
+ xbt_backtrace_display_current();
+ xbt_die("I should be dead by now.");
}
if (self->suspended_) {
XBT_DEBUG("Hey! I'm suspended.");
xbt_assert(self->exception != nullptr, "Gasp! This exception may be lost by subsequent calls.");
- self->suspended_ = 0;
+ self->suspended_ = false;
self->suspend(self);
}
/** @brief Restart a process, starting it again from the beginning. */
/**
- * \ingroup simix_process_management
- * \brief Creates and runs a new SIMIX process.
+ * @ingroup simix_process_management
+ * @brief Creates and runs a new SIMIX process.
*
* The structure and the corresponding thread are created and put in the list of ready processes.
*
- * \param name a name for the process. It is for user-level information and can be nullptr.
- * \param code the main function of the process
- * \param data a pointer to any data one may want to attach to the new object. It is for user-level information and can
+ * @param name a name for the process. It is for user-level information and can be nullptr.
+ * @param code the main function of the process
+ * @param data a pointer to any data one may want to attach to the new object. It is for user-level information and can
* be nullptr.
* It can be retrieved with the method ActorImpl::getUserData().
- * \param host where the new agent is executed.
- * \param argc first argument passed to \a code
- * \param argv second argument passed to \a code
- * \param properties the properties of the process
+ * @param host where the new agent is executed.
+ * @param argc first argument passed to @a code
+ * @param argv second argument passed to @a code
+ * @param properties the properties of the process
*/
smx_actor_t simcall_process_create(std::string name, xbt_main_func_t code, void* data, sg_host_t host, int argc,
char** argv, std::unordered_map<std::string, std::string>* properties)