X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/1485fe26372c540b07a9a86a5d08155d06d77e81..8265fae7843fa515935916b005f15daa93ff2f26:/src/simix/ActorImpl.cpp diff --git a/src/simix/ActorImpl.cpp b/src/simix/ActorImpl.cpp index 38b423db28..8c9e22a3c5 100644 --- a/src/simix/ActorImpl.cpp +++ b/src/simix/ActorImpl.cpp @@ -4,7 +4,10 @@ * 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/kernel/activity/IoImpl.hpp" #include "src/kernel/activity/SleepImpl.hpp" #include "src/kernel/activity/SynchroRaw.hpp" @@ -13,8 +16,8 @@ #include "src/simix/smx_host_private.hpp" #include "src/simix/smx_io_private.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" @@ -27,11 +30,11 @@ XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_process, simix, "Logging specific to SIMIX 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() { @@ -41,8 +44,8 @@ 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) { @@ -50,7 +53,7 @@ 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) { @@ -66,7 +69,7 @@ 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()) { @@ -98,19 +101,20 @@ void SIMIX_process_cleanup(smx_actor_t process) 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_->extension()->process_list, *process); + simgrid::xbt::intrusive_erase(process->host_->pimpl_->process_list_, *process); if (not process->smx_destroy_list_hook.is_linked()) { #if SIMGRID_HAVE_MC xbt_dynar_push_as(simix_global->dead_actors_vector, smx_actor_t, 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(); } /** @@ -168,15 +172,7 @@ simgrid::s4u::Actor* ActorImpl::restart() XBT_DEBUG("Restarting process %s on %s", get_cname(), host_->get_cname()); // retrieve the arguments of the old process - // FIXME: Factorize this with SIMIX_host_add_auto_restart_process ? - simgrid::kernel::actor::ProcessArg arg; - arg.name = name_; - arg.code = code; - arg.host = host_; - arg.kill_time = SIMIX_timer_get_date(kill_timer); - arg.data = userdata_; - arg.properties = nullptr; - arg.auto_restart = auto_restart_; + simgrid::kernel::actor::ProcessArg arg = ProcessArg(host_, this); // kill the old process SIMIX_process_kill(this, (this == simix_global->maestro_process) ? this : SIMIX_process_self()); @@ -184,10 +180,8 @@ simgrid::s4u::Actor* ActorImpl::restart() // 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); - if (arg.kill_time >= 0) - simcall_process_set_kill_time(actor, arg.kill_time); - if (arg.auto_restart) - actor->auto_restart_ = arg.auto_restart; + simcall_process_set_kill_time(actor, arg.kill_time); + actor->set_auto_restart(arg.auto_restart); return actor->ciface(); } @@ -199,7 +193,7 @@ smx_activity_t ActorImpl::suspend(ActorImpl* issuer) return nullptr; } - suspended_ = 1; + suspended_ = true; /* If we are suspending another actor that is waiting on a sync, suspend its synchronization. */ if (this != issuer) { @@ -209,7 +203,7 @@ smx_activity_t ActorImpl::suspend(ActorImpl* 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_); } } @@ -224,7 +218,7 @@ void ActorImpl::resume() if (not suspended_) return; - suspended_ = 0; + suspended_ = false; /* resume the synchronization that was blocking the resumed actor. */ if (waiting_synchro) @@ -236,7 +230,8 @@ void ActorImpl::resume() 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_; @@ -247,6 +242,55 @@ smx_activity_t ActorImpl::sleep(double duration) 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(waiting_synchro); + if (exec != nullptr && exec->surf_action_) + exec->surf_action_->cancel(); + + simgrid::kernel::activity::CommImplPtr comm = + boost::dynamic_pointer_cast(waiting_synchro); + if (comm != nullptr) { + comms.remove(comm); + comm->cancel(); + } + + simgrid::kernel::activity::SleepImplPtr sleep = + boost::dynamic_pointer_cast(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(waiting_synchro); + if (raw != nullptr) { + SIMIX_synchro_stop_waiting(this, &simcall); + } + + simgrid::kernel::activity::IoImplPtr io = + boost::dynamic_pointer_cast(waiting_synchro); + if (io != nullptr) { + delete io.get(); + } + } + waiting_synchro = nullptr; +} + void create_maestro(simgrid::simix::ActorCode code) { smx_actor_t maestro = nullptr; @@ -279,22 +323,22 @@ void SIMIX_maestro_create(void (*code)(void*), void* data) } /** - * \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(const char* name, simgrid::simix::ActorCode code, void* data, simgrid::s4u::Host* host, +smx_actor_t SIMIX_process_create(std::string name, simgrid::simix::ActorCode code, void* data, simgrid::s4u::Host* host, std::unordered_map* properties, smx_actor_t parent_process) { - XBT_DEBUG("Start process %s on host '%s'", name, host->get_cname()); + XBT_DEBUG("Start process %s on host '%s'", name.c_str(), host->get_cname()); if (host->is_off()) { - XBT_WARN("Cannot launch process '%s' on failed host '%s'", name, host->get_cname()); + XBT_WARN("Cannot launch process '%s' on failed host '%s'", name.c_str(), host->get_cname()); return nullptr; } @@ -322,12 +366,8 @@ smx_actor_t SIMIX_process_create(const char* name, simgrid::simix::ActorCode cod for (auto const& kv : *properties) process->set_property(kv.first, kv.second); - /* Make sure that the process is initialized for simix, in case we are called from the Host::onCreation signal */ - if (host->extension() == nullptr) - host->extension_set(new simgrid::simix::Host()); - /* Add the process to its host's process list */ - host->extension()->process_list.push_back(*process); + host->pimpl_->process_list_.push_back(*process); XBT_DEBUG("Start context '%s'", process->get_cname()); @@ -384,7 +424,7 @@ smx_actor_t SIMIX_process_attach(const char* name, void* data, const char* hostn process->set_property(kv.first, kv.second); /* Add the process to it's host process list */ - host->extension()->process_list.push_back(*process); + host->pimpl_->process_list_.push_back(*process); /* Now insert it in the global process list and in the process to run list */ simix_global->process_list[process->pid_] = process; @@ -417,7 +457,7 @@ void SIMIX_process_detach() } /** - * \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 @@ -434,13 +474,13 @@ void SIMIX_process_runall() } /** - * \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) { @@ -450,14 +490,18 @@ void SIMIX_process_kill(smx_actor_t process, smx_actor_t issuer) { return; } - XBT_DEBUG("Actor '%s'@%s is killing actor '%s'@%s", issuer->get_cname(), issuer->host_->get_cname(), - process->get_cname(), process->host_->get_cname()); + XBT_DEBUG("Actor '%s'@%s is killing actor '%s'@%s", issuer->get_cname(), + (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; + if (process->host_->is_off()) + process->throw_exception(std::make_exception_ptr(simgrid::HostFailureException(XBT_THROW_POINT, "Host failed"))); + /* destroy the blocking synchro if any */ if (process->waiting_synchro != nullptr) { @@ -486,15 +530,18 @@ void SIMIX_process_kill(smx_actor_t process, smx_actor_t issuer) { if (i != process->waiting_synchro->simcalls_.end()) process->waiting_synchro->simcalls_.remove(&process->simcall); } else if (sleep != nullptr) { - SIMIX_process_sleep_destroy(process->waiting_synchro); - + if (sleep->surf_sleep) + sleep->surf_sleep->cancel(); + sleep->post(); } else if (raw != 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; @@ -507,13 +554,8 @@ void SIMIX_process_kill(smx_actor_t process, smx_actor_t issuer) { } } -/** @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); @@ -556,7 +598,7 @@ void SIMIX_process_throw(smx_actor_t process, xbt_errcat_t cat, int value, const simgrid::kernel::activity::IoImplPtr io = boost::dynamic_pointer_cast(process->waiting_synchro); if (io != nullptr) { - SIMIX_io_destroy(process->waiting_synchro); + delete io.get(); } } process->waiting_synchro = nullptr; @@ -564,8 +606,8 @@ void SIMIX_process_throw(smx_actor_t process, xbt_errcat_t cat, int value, const } /** - * \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) { @@ -577,9 +619,9 @@ void SIMIX_process_killall(smx_actor_t issuer) void SIMIX_process_change_host(smx_actor_t actor, sg_host_t dest) { xbt_assert((actor != nullptr), "Invalid parameters"); - simgrid::xbt::intrusive_erase(actor->host_->extension()->process_list, *actor); + simgrid::xbt::intrusive_erase(actor->host_->pimpl_->process_list_, *actor); actor->host_ = dest; - dest->extension()->process_list.push_back(*actor); + dest->pimpl_->process_list_.push_back(*actor); } void simcall_HANDLER_process_suspend(smx_simcall_t simcall, smx_actor_t process) @@ -686,11 +728,11 @@ void SIMIX_process_sleep_destroy(smx_activity_t synchro) } /** - * \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) { @@ -707,18 +749,23 @@ 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_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); } @@ -771,26 +818,24 @@ void SIMIX_process_on_exit(smx_actor_t process, std::function /** @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(const char* name, xbt_main_func_t code, void* data, sg_host_t host, int argc, +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* properties) { - if (name == nullptr) - name = ""; auto wrapped_code = simgrid::xbt::wrap_main(code, argc, argv); for (int i = 0; i != argc; ++i) xbt_free(argv[i]); @@ -799,11 +844,9 @@ smx_actor_t simcall_process_create(const char* name, xbt_main_func_t code, void* return res; } -smx_actor_t simcall_process_create(const char* name, simgrid::simix::ActorCode code, void* data, sg_host_t host, +smx_actor_t simcall_process_create(std::string name, simgrid::simix::ActorCode code, void* data, sg_host_t host, std::unordered_map* properties) { - if (name == nullptr) - name = ""; smx_actor_t self = SIMIX_process_self(); return simgrid::simix::simcall([name, code, data, host, properties, self] { return SIMIX_process_create(name, std::move(code), data, host, properties, self);