X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/32892d0df7206a7c14ace5a13f985df1e0a9249a..a539cc1790aed8a1549a6cde5cdddc0876615851:/src/simix/ActorImpl.cpp diff --git a/src/simix/ActorImpl.cpp b/src/simix/ActorImpl.cpp index 2f2b67c3d7..ea437030f4 100644 --- a/src/simix/ActorImpl.cpp +++ b/src/simix/ActorImpl.cpp @@ -19,10 +19,6 @@ #include "src/surf/HostImpl.hpp" #include "src/surf/cpu_interface.hpp" -#ifdef HAVE_SMPI -#include "src/smpi/include/private.hpp" -#endif - #include XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_process, simix, "Logging specific to SIMIX (process)"); @@ -101,7 +97,8 @@ 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_->pimpl_->process_list_, *process); @@ -139,6 +136,12 @@ namespace simgrid { namespace kernel { namespace actor { +ActorImpl::ActorImpl(simgrid::xbt::string name, simgrid::s4u::Host* host) : name_(name), host_(host), piface_(this) +{ + pid_ = simix_process_maxpid++; + simcall.issuer = this; +} + ActorImpl::~ActorImpl() { delete this->context_; @@ -177,8 +180,8 @@ simgrid::s4u::Actor* ActorImpl::restart() 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, 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); @@ -229,7 +232,8 @@ 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_; @@ -240,14 +244,59 @@ 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; /* Create maestro process and initialize it */ - maestro = new simgrid::kernel::actor::ActorImpl(); - maestro->pid_ = simix_process_maxpid++; - maestro->name_ = ""; - maestro->set_user_data(nullptr); + smx_actor_t maestro = new simgrid::kernel::actor::ActorImpl(simgrid::xbt::string(""), /*host*/ nullptr); if (not code) { maestro->context_ = SIMIX_context_new(simgrid::simix::ActorCode(), nullptr, maestro); @@ -284,31 +333,25 @@ smx_actor_t SIMIX_process_create(std::string name, simgrid::simix::ActorCode cod std::unordered_map* properties, smx_actor_t parent_process) { - XBT_DEBUG("Start process %s on host '%s'", name.c_str(), host->get_cname()); + XBT_DEBUG("Start actor %s@'%s'", name.c_str(), host->get_cname()); if (host->is_off()) { XBT_WARN("Cannot launch process '%s' on failed host '%s'", name.c_str(), host->get_cname()); return nullptr; } - smx_actor_t process = new simgrid::kernel::actor::ActorImpl(); + smx_actor_t process = new simgrid::kernel::actor::ActorImpl(simgrid::xbt::string(name), host); xbt_assert(code && host != nullptr, "Invalid parameters"); /* Process data */ - process->pid_ = simix_process_maxpid++; - process->name_ = simgrid::xbt::string(name); - process->host_ = host; process->set_user_data(data); - process->simcall.issuer = process; + process->code = code; - if (parent_process != nullptr) { + if (parent_process != nullptr) process->ppid_ = parent_process->pid_; - } - - 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) @@ -347,25 +390,17 @@ smx_actor_t SIMIX_process_attach(const char* name, void* data, const char* hostn return nullptr; } - smx_actor_t process = new simgrid::kernel::actor::ActorImpl(); + smx_actor_t process = new simgrid::kernel::actor::ActorImpl(simgrid::xbt::string(name), host); /* Process data */ - process->pid_ = simix_process_maxpid++; - process->name_ = std::string(name); - process->host_ = host; process->set_user_data(data); - process->simcall.issuer = process; + process->code = nullptr; - if (parent_process != nullptr) { + if (parent_process != nullptr) process->ppid_ = parent_process->pid_; - } - - /* Process data for auto-restart */ - process->code = 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); + xbt_assert(simix_global != nullptr, "simix is not initialized, please call MSG_init first"); + process->context_ = simix_global->context_factory->attach(&SIMIX_process_cleanup, process); /* Add properties */ if (properties != nullptr) @@ -382,13 +417,11 @@ smx_actor_t SIMIX_process_attach(const char* name, void* data, const char* hostn intrusive_ptr_add_ref(process); auto* context = dynamic_cast(process->context_); - if (not context) - xbt_die("Not a suitable context"); - + xbt_assert(nullptr != context, "Not a suitable context"); context->attach_start(); - /* The onCreation() signal must be delayed until there, where the pid and everything is set */ - simgrid::s4u::ActorPtr tmp = process->iface(); // Passing this directly to onCreation will lead to crashes + /* The on_creation() signal must be delayed until there, where the pid and everything is set */ + simgrid::s4u::ActorPtr tmp = process->iface(); // Passing this directly to on_creation will lead to crashes simgrid::s4u::Actor::on_creation(tmp); return process; @@ -401,7 +434,7 @@ void SIMIX_process_detach() xbt_die("Not a suitable context"); auto process = context->process(); - simix_global->cleanup_process_function(process); + SIMIX_process_cleanup(process); context->attach_stop(); } @@ -448,11 +481,35 @@ void SIMIX_process_kill(smx_actor_t process, smx_actor_t issuer) { 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) { - if (process->host_->is_off()) { - SMX_EXCEPTION(process, host_error, 0, "Host failed"); - } + simgrid::kernel::activity::ExecImplPtr exec = boost::dynamic_pointer_cast(process->waiting_synchro); simgrid::kernel::activity::CommImplPtr comm = @@ -487,7 +544,9 @@ void SIMIX_process_kill(smx_actor_t process, smx_actor_t issuer) { } else if (io != nullptr) { 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; @@ -500,13 +559,8 @@ 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); @@ -710,6 +764,8 @@ void SIMIX_process_yield(smx_actor_t self) 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_) { @@ -779,21 +835,8 @@ void SIMIX_process_on_exit(smx_actor_t process, std::function * 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 */ -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) -{ - auto wrapped_code = simgrid::xbt::wrap_main(code, argc, argv); - for (int i = 0; i != argc; ++i) - xbt_free(argv[i]); - xbt_free(argv); - smx_actor_t res = simcall_process_create(name, std::move(wrapped_code), data, host, properties); - return res; -} - smx_actor_t simcall_process_create(std::string name, simgrid::simix::ActorCode code, void* data, sg_host_t host, std::unordered_map* properties) {