src/s4u/s4u_Comm.cpp

   1 /* Copyright (c) 2006-2023. The SimGrid Team. All rights reserved.          */
   2
   3 /* This program is free software; you can redistribute it and/or modify it
   4  * under the terms of the license (GNU LGPL) which comes with this package. */
   5
   6 #include <cmath>
   7 #include <simgrid/Exception.hpp>
   8 #include <simgrid/comm.h>
   9 #include <simgrid/s4u/Comm.hpp>
  10 #include <simgrid/s4u/Engine.hpp>
  11 #include <simgrid/s4u/Mailbox.hpp>
  12
  13 #include "mc/mc.h"
  14 #include "src/kernel/activity/CommImpl.hpp"
  15 #include "src/kernel/actor/ActorImpl.hpp"
  16 #include "src/kernel/actor/SimcallObserver.hpp"
  17 #include "src/mc/mc_replay.hpp"
  18
  19 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(s4u_comm, s4u_activity, "S4U asynchronous communications");
  20
  21 namespace simgrid::s4u {
  22 xbt::signal<void(Comm const&)> Comm::on_send;
  23 xbt::signal<void(Comm const&)> Comm::on_recv;
  24
  25 CommPtr Comm::set_copy_data_callback(const std::function<void(kernel::activity::CommImpl*, void*, size_t)>& callback)
  26 {
  27   copy_data_function_ = callback;
  28   return this;
  29 }
  30
  31 void Comm::copy_buffer_callback(kernel::activity::CommImpl* comm, void* buff, size_t buff_size)
  32 {
  33   XBT_DEBUG("Copy the data over");
  34   memcpy(comm->dst_buff_, buff, buff_size);
  35   if (comm->is_detached()) { // if this is a detached send, the source buffer was duplicated by SMPI sender to make the
  36                              // original buffer available to the application ASAP
  37     xbt_free(buff);
  38     comm->src_buff_ = nullptr;
  39   }
  40 }
  41
  42 void Comm::copy_pointer_callback(kernel::activity::CommImpl* comm, void* buff, size_t buff_size)
  43 {
  44   xbt_assert((buff_size == sizeof(void*)), "Cannot copy %zu bytes: must be sizeof(void*)", buff_size);
  45   *(void**)(comm->dst_buff_) = buff;
  46 }
  47
  48 Comm::~Comm()
  49 {
  50   if (state_ == State::STARTED && not detached_ &&
  51       (pimpl_ == nullptr || pimpl_->get_state() == kernel::activity::State::RUNNING)) {
  52     XBT_INFO("Comm %p freed before its completion. Did you forget to detach it? (state: %s)", this, get_state_str());
  53     if (pimpl_ != nullptr)
  54       XBT_INFO("pimpl_->state: %s", pimpl_->get_state_str());
  55     else
  56       XBT_INFO("pimpl_ is null");
  57     xbt_backtrace_display_current();
  58   }
  59 }
  60
  61 void Comm::send(kernel::actor::ActorImpl* sender, const Mailbox* mbox, double task_size, double rate, void* src_buff,
  62                 size_t src_buff_size,
  63                 const std::function<bool(void*, void*, simgrid::kernel::activity::CommImpl*)>& match_fun,
  64                 const std::function<void(simgrid::kernel::activity::CommImpl*, void*, size_t)>& copy_data_fun,
  65                 void* data, double timeout)
  66 {
  67   /* checking for infinite values */
  68   xbt_assert(std::isfinite(task_size), "task_size is not finite!");
  69   xbt_assert(std::isfinite(rate), "rate is not finite!");
  70   xbt_assert(std::isfinite(timeout), "timeout is not finite!");
  71
  72   xbt_assert(mbox, "No rendez-vous point defined for send");
  73
  74   if (MC_is_active() || MC_record_replay_is_active()) {
  75     /* the model-checker wants two separate simcalls, and wants comm to be nullptr during the simcall */
  76     simgrid::kernel::activity::ActivityImplPtr comm = nullptr;
  77
  78     simgrid::kernel::actor::CommIsendSimcall send_observer{
  79         sender,  mbox->get_impl(), task_size, rate, static_cast<unsigned char*>(src_buff), src_buff_size, match_fun,
  80         nullptr, copy_data_fun,    data,      false};
  81     comm = simgrid::kernel::actor::simcall_answered(
  82         [&send_observer] { return simgrid::kernel::activity::CommImpl::isend(&send_observer); }, &send_observer);
  83
  84     if (simgrid::kernel::actor::ActivityWaitSimcall wait_observer{sender, comm.get(), timeout};
  85         simgrid::kernel::actor::simcall_blocking(
  86             [&wait_observer] {
  87               wait_observer.get_activity()->wait_for(wait_observer.get_issuer(), wait_observer.get_timeout());
  88             },
  89             &wait_observer)) {
  90       throw simgrid::TimeoutException(XBT_THROW_POINT, "Timeouted");
  91     }
  92     comm = nullptr;
  93   } else {
  94     simgrid::kernel::actor::CommIsendSimcall observer(sender, mbox->get_impl(), task_size, rate,
  95                                                       static_cast<unsigned char*>(src_buff), src_buff_size, match_fun,
  96                                                       nullptr, copy_data_fun, data, false);
  97     simgrid::kernel::actor::simcall_blocking([&observer, timeout] {
  98       simgrid::kernel::activity::ActivityImplPtr comm = simgrid::kernel::activity::CommImpl::isend(&observer);
  99       comm->wait_for(observer.get_issuer(), timeout);
 100     });
 101   }
 102 }
 103
 104 void Comm::recv(kernel::actor::ActorImpl* receiver, const Mailbox* mbox, void* dst_buff, size_t* dst_buff_size,
 105                 const std::function<bool(void*, void*, simgrid::kernel::activity::CommImpl*)>& match_fun,
 106                 const std::function<void(simgrid::kernel::activity::CommImpl*, void*, size_t)>& copy_data_fun,
 107                 void* data, double timeout, double rate)
 108 {
 109   xbt_assert(std::isfinite(timeout), "timeout is not finite!");
 110   xbt_assert(mbox, "No rendez-vous point defined for recv");
 111
 112   if (MC_is_active() || MC_record_replay_is_active()) {
 113     /* the model-checker wants two separate simcalls, and wants comm to be nullptr during the simcall */
 114     simgrid::kernel::activity::ActivityImplPtr comm = nullptr;
 115
 116     simgrid::kernel::actor::CommIrecvSimcall observer{receiver,
 117                                                       mbox->get_impl(),
 118                                                       static_cast<unsigned char*>(dst_buff),
 119                                                       dst_buff_size,
 120                                                       match_fun,
 121                                                       copy_data_fun,
 122                                                       data,
 123                                                       rate};
 124     comm = simgrid::kernel::actor::simcall_answered(
 125         [&observer] { return simgrid::kernel::activity::CommImpl::irecv(&observer); }, &observer);
 126
 127     if (simgrid::kernel::actor::ActivityWaitSimcall wait_observer{receiver, comm.get(), timeout};
 128         simgrid::kernel::actor::simcall_blocking(
 129             [&wait_observer] {
 130               wait_observer.get_activity()->wait_for(wait_observer.get_issuer(), wait_observer.get_timeout());
 131             },
 132             &wait_observer)) {
 133       throw simgrid::TimeoutException(XBT_THROW_POINT, "Timeouted");
 134     }
 135     comm = nullptr;
 136   } else {
 137     simgrid::kernel::actor::CommIrecvSimcall observer(receiver, mbox->get_impl(), static_cast<unsigned char*>(dst_buff),
 138                                                       dst_buff_size, match_fun, copy_data_fun, data, rate);
 139     simgrid::kernel::actor::simcall_blocking([&observer, timeout] {
 140       simgrid::kernel::activity::ActivityImplPtr comm = simgrid::kernel::activity::CommImpl::irecv(&observer);
 141       comm->wait_for(observer.get_issuer(), timeout);
 142     });
 143   }
 144 }
 145
 146 CommPtr Comm::sendto_init()
 147 {
 148   CommPtr res(new Comm());
 149   res->pimpl_ = kernel::activity::CommImplPtr(new kernel::activity::CommImpl());
 150   boost::static_pointer_cast<kernel::activity::CommImpl>(res->pimpl_)->detach();
 151   res->sender_ = kernel::actor::ActorImpl::self();
 152   return res;
 153 }
 154
 155 CommPtr Comm::sendto_init(Host* from, Host* to)
 156 {
 157   auto res = Comm::sendto_init()->set_source(from)->set_destination(to);
 158   res->set_state(State::STARTING);
 159   return res;
 160 }
 161
 162 CommPtr Comm::sendto_async(Host* from, Host* to, uint64_t simulated_size_in_bytes)
 163 {
 164   return Comm::sendto_init()->set_payload_size(simulated_size_in_bytes)->set_source(from)->set_destination(to);
 165 }
 166
 167 void Comm::sendto(Host* from, Host* to, uint64_t simulated_size_in_bytes)
 168 {
 169   sendto_async(from, to, simulated_size_in_bytes)->wait();
 170 }
 171
 172 CommPtr Comm::set_source(Host* from)
 173 {
 174   xbt_assert(state_ == State::INITED || state_ == State::STARTING,
 175              "Cannot change the source of a Comm once it's started (state: %s)", to_c_str(state_));
 176   boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->set_source(from);
 177   // Setting 'source' may allow to start the activity, let's try
 178   if (state_ == State::STARTING && remains_ <= 0)
 179     XBT_DEBUG("This communication has a payload size of 0 byte. It cannot start yet");
 180   else
 181     vetoable_start();
 182
 183   return this;
 184 }
 185 Host* Comm::get_source() const
 186 {
 187   return pimpl_ ? boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->get_source() : nullptr;
 188 }
 189
 190 CommPtr Comm::set_destination(Host* to)
 191 {
 192   xbt_assert(state_ == State::INITED || state_ == State::STARTING,
 193              "Cannot change the destination of a Comm once it's started (state: %s)", to_c_str(state_));
 194   boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->set_destination(to);
 195   // Setting 'destination' may allow to start the activity, let's try
 196   if (state_ == State::STARTING && remains_ <= 0)
 197     XBT_DEBUG("This communication has a payload size of 0 byte. It cannot start yet");
 198   else
 199     vetoable_start();
 200
 201   return this;
 202 }
 203
 204 Host* Comm::get_destination() const
 205 {
 206   return pimpl_ ? boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->get_destination() : nullptr;
 207 }
 208
 209 CommPtr Comm::set_rate(double rate)
 210 {
 211   xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
 212              __FUNCTION__);
 213   rate_ = rate;
 214   return this;
 215 }
 216
 217 CommPtr Comm::set_mailbox(Mailbox* mailbox)
 218 {
 219   xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
 220              __FUNCTION__);
 221   mailbox_ = mailbox;
 222   return this;
 223 }
 224
 225 CommPtr Comm::set_src_data(void* buff)
 226 {
 227   xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
 228              __FUNCTION__);
 229   xbt_assert(dst_buff_ == nullptr, "Cannot set the src and dst buffers at the same time");
 230   src_buff_ = buff;
 231   return this;
 232 }
 233
 234 CommPtr Comm::set_src_data_size(size_t size)
 235 {
 236   xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
 237              __FUNCTION__);
 238   src_buff_size_ = size;
 239   return this;
 240 }
 241
 242 CommPtr Comm::set_src_data(void* buff, size_t size)
 243 {
 244   xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
 245              __FUNCTION__);
 246
 247   xbt_assert(dst_buff_ == nullptr, "Cannot set the src and dst buffers at the same time");
 248   src_buff_      = buff;
 249   src_buff_size_ = size;
 250   return this;
 251 }
 252
 253 CommPtr Comm::set_dst_data(void** buff)
 254 {
 255   xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
 256              __FUNCTION__);
 257   xbt_assert(src_buff_ == nullptr, "Cannot set the src and dst buffers at the same time");
 258   dst_buff_ = buff;
 259   return this;
 260 }
 261
 262 CommPtr Comm::set_dst_data(void** buff, size_t size)
 263 {
 264   xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
 265              __FUNCTION__);
 266
 267   xbt_assert(src_buff_ == nullptr, "Cannot set the src and dst buffers at the same time");
 268   dst_buff_      = buff;
 269   dst_buff_size_ = size;
 270   return this;
 271 }
 272
 273 CommPtr Comm::set_payload_size(uint64_t bytes)
 274 {
 275   set_remaining(bytes);
 276   if (pimpl_) {
 277     boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->set_size(bytes);
 278   }
 279   return this;
 280 }
 281
 282 Actor* Comm::get_sender() const
 283 {
 284   kernel::actor::ActorImplPtr sender = nullptr;
 285   if (pimpl_)
 286     sender = boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->src_actor_;
 287   return sender ? sender->get_ciface() : nullptr;
 288 }
 289
 290 bool Comm::is_assigned() const
 291 {
 292   return (pimpl_ && boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->is_assigned()) ||
 293          mailbox_ != nullptr;
 294 }
 295
 296 Comm* Comm::start()
 297 {
 298   xbt_assert(get_state() == State::INITED || get_state() == State::STARTING,
 299              "You cannot use %s() once your communication started (not implemented)", __FUNCTION__);
 300   if (get_source() != nullptr || get_destination() != nullptr) {
 301     xbt_assert(is_assigned(), "When either from_ or to_ is specified, both must be.");
 302     xbt_assert(src_buff_ == nullptr && dst_buff_ == nullptr,
 303                "Direct host-to-host communications cannot carry any data.");
 304     XBT_DEBUG("host-to-host Comm. Pimpl already created and set, just start it.");
 305     kernel::actor::simcall_answered([this] {
 306       pimpl_->set_state(kernel::activity::State::READY);
 307       boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->start();
 308     });
 309   } else if (src_buff_ != nullptr) { // Sender side
 310     on_send(*this);
 311     kernel::actor::CommIsendSimcall observer{sender_,
 312                                              mailbox_->get_impl(),
 313                                              remains_,
 314                                              rate_,
 315                                              static_cast<unsigned char*>(src_buff_),
 316                                              src_buff_size_,
 317                                              match_fun_,
 318                                              clean_fun_,
 319                                              copy_data_function_,
 320                                              get_data<void>(),
 321                                              detached_};
 322     pimpl_ = kernel::actor::simcall_answered([&observer] { return kernel::activity::CommImpl::isend(&observer); },
 323                                              &observer);
 324   } else if (dst_buff_ != nullptr) { // Receiver side
 325     xbt_assert(not detached_, "Receive cannot be detached");
 326     on_recv(*this);
 327     kernel::actor::CommIrecvSimcall observer{receiver_,
 328                                              mailbox_->get_impl(),
 329                                              static_cast<unsigned char*>(dst_buff_),
 330                                              &dst_buff_size_,
 331                                              match_fun_,
 332                                              copy_data_function_,
 333                                              get_data<void>(),
 334                                              rate_};
 335     pimpl_ = kernel::actor::simcall_answered([&observer] { return kernel::activity::CommImpl::irecv(&observer); },
 336                                              &observer);
 337   } else {
 338     xbt_die("Cannot start a communication before specifying whether we are the sender or the receiver");
 339   }
 340
 341   if (suspended_)
 342     pimpl_->suspend();
 343
 344   if (not detached_) {
 345     pimpl_->set_iface(this);
 346     pimpl_->set_actor(sender_);
 347   }
 348
 349   state_ = State::STARTED;
 350   return this;
 351 }
 352
 353 Comm* Comm::detach()
 354 {
 355   xbt_assert(state_ == State::INITED || state_ == State::STARTING,
 356              "You cannot use %s() once your communication is %s (not implemented)", __FUNCTION__, get_state_str());
 357   xbt_assert(dst_buff_ == nullptr && dst_buff_size_ == 0, "You can only detach sends, not recvs");
 358   detached_ = true;
 359   vetoable_start();
 360   return this;
 361 }
 362
 363 ssize_t Comm::test_any(const std::vector<CommPtr>& comms)
 364 {
 365   std::vector<ActivityPtr> activities;
 366   for (const auto& comm : comms)
 367     activities.push_back(boost::dynamic_pointer_cast<Activity>(comm));
 368   return Activity::test_any(activities);
 369 }
 370
 371 /** @brief Block the calling actor until the communication is finished, or until timeout
 372  *
 373  * On timeout, an exception is thrown and the communication is invalidated.
 374  *
 375  * @param timeout the amount of seconds to wait for the comm termination.
 376  *                Negative values denote infinite wait times. 0 as a timeout returns immediately. */
 377 Comm* Comm::wait_for(double timeout)
 378 {
 379   XBT_DEBUG("Calling Comm::wait_for with state %s", get_state_str());
 380   kernel::actor::ActorImpl* issuer = nullptr;
 381   switch (state_) {
 382     case State::FINISHED:
 383       break;
 384     case State::FAILED:
 385       throw NetworkFailureException(XBT_THROW_POINT, "Cannot wait for a failed communication");
 386     case State::INITED:
 387     case State::STARTING: // It's not started yet. Do it in one simcall if it's a regular communication
 388       if (get_source() != nullptr || get_destination() != nullptr) {
 389         return vetoable_start()->wait_for(timeout); // In the case of host2host comm, do it in two simcalls
 390       } else if (src_buff_ != nullptr) {
 391         on_send(*this);
 392         send(sender_, mailbox_, remains_, rate_, src_buff_, src_buff_size_, match_fun_, copy_data_function_,
 393              get_data<void>(), timeout);
 394
 395       } else { // Receiver
 396         on_recv(*this);
 397         recv(receiver_, mailbox_, dst_buff_, &dst_buff_size_, match_fun_, copy_data_function_, get_data<void>(),
 398              timeout, rate_);
 399       }
 400       break;
 401     case State::STARTED:
 402       try {
 403         issuer = kernel::actor::ActorImpl::self();
 404         kernel::actor::ActivityWaitSimcall observer{issuer, pimpl_.get(), timeout};
 405         if (kernel::actor::simcall_blocking(
 406                 [&observer] { observer.get_activity()->wait_for(observer.get_issuer(), observer.get_timeout()); },
 407                 &observer)) {
 408           throw TimeoutException(XBT_THROW_POINT, "Timeouted");
 409         }
 410       } catch (const NetworkFailureException& e) {
 411         issuer->simcall_.observer_ = nullptr; // Comm failed on network failure, reset the observer to nullptr
 412         complete(State::FAILED);
 413         e.rethrow_nested(XBT_THROW_POINT, boost::core::demangle(typeid(e).name()) + " raised in kernel mode.");
 414       }
 415       break;
 416
 417     case State::CANCELED:
 418       throw CancelException(XBT_THROW_POINT, "Communication canceled");
 419
 420     default:
 421       THROW_IMPOSSIBLE;
 422   }
 423   complete(State::FINISHED);
 424   return this;
 425 }
 426
 427 ssize_t Comm::wait_any_for(const std::vector<CommPtr>& comms, double timeout)
 428 {
 429   std::vector<ActivityPtr> activities;
 430   for (const auto& comm : comms)
 431     activities.push_back(boost::dynamic_pointer_cast<Activity>(comm));
 432   ssize_t changed_pos;
 433   try {
 434     changed_pos = Activity::wait_any_for(activities, timeout);
 435   } catch (const NetworkFailureException& e) {
 436     changed_pos = -1;
 437     for (auto c : comms) {
 438       if (c->pimpl_->get_state() == kernel::activity::State::FAILED) {
 439         c->complete(State::FAILED);
 440       }
 441     }
 442     e.rethrow_nested(XBT_THROW_POINT, boost::core::demangle(typeid(e).name()) + " raised in kernel mode.");
 443   }
 444   return changed_pos;
 445 }
 446
 447 void Comm::wait_all(const std::vector<CommPtr>& comms)
 448 {
 449   // TODO: this should be a simcall or something
 450   for (auto& comm : comms)
 451     comm->wait();
 452 }
 453
 454 size_t Comm::wait_all_for(const std::vector<CommPtr>& comms, double timeout)
 455 {
 456   if (timeout < 0.0) {
 457     wait_all(comms);
 458     return comms.size();
 459   }
 460
 461   double deadline = Engine::get_clock() + timeout;
 462   std::vector<CommPtr> waited_comm(1, nullptr);
 463   for (size_t i = 0; i < comms.size(); i++) {
 464     double wait_timeout = std::max(0.0, deadline - Engine::get_clock());
 465     waited_comm[0]      = comms[i];
 466     // Using wait_any_for() here (and not wait_for) because we don't want comms to be invalidated on timeout
 467     if (wait_any_for(waited_comm, wait_timeout) == -1) {
 468       XBT_DEBUG("Timeout (%g): i = %zu", wait_timeout, i);
 469       return i;
 470     }
 471   }
 472   return comms.size();
 473 }
 474 } // namespace simgrid::s4u
 475 /* **************************** Public C interface *************************** */
 476 void sg_comm_detach(sg_comm_t comm, void (*clean_function)(void*))
 477 {
 478   comm->detach(clean_function);
 479   comm->unref();
 480 }
 481 void sg_comm_unref(sg_comm_t comm)
 482 {
 483   comm->unref();
 484 }
 485 int sg_comm_test(sg_comm_t comm)
 486 {
 487   bool finished = comm->test();
 488   if (finished)
 489     comm->unref();
 490   return finished;
 491 }
 492
 493 sg_error_t sg_comm_wait(sg_comm_t comm)
 494 {
 495   return sg_comm_wait_for(comm, -1);
 496 }
 497
 498 sg_error_t sg_comm_wait_for(sg_comm_t comm, double timeout)
 499 {
 500   sg_error_t status = SG_OK;
 501
 502   simgrid::s4u::CommPtr s4u_comm(comm, false);
 503   try {
 504     s4u_comm->wait_for(timeout);
 505   } catch (const simgrid::TimeoutException&) {
 506     status = SG_ERROR_TIMEOUT;
 507   } catch (const simgrid::CancelException&) {
 508     status = SG_ERROR_CANCELED;
 509   } catch (const simgrid::NetworkFailureException&) {
 510     status = SG_ERROR_NETWORK;
 511   }
 512   return status;
 513 }
 514
 515 void sg_comm_wait_all(sg_comm_t* comms, size_t count)
 516 {
 517   sg_comm_wait_all_for(comms, count, -1);
 518 }
 519
 520 size_t sg_comm_wait_all_for(sg_comm_t* comms, size_t count, double timeout)
 521 {
 522   std::vector<simgrid::s4u::CommPtr> s4u_comms;
 523   for (size_t i = 0; i < count; i++)
 524     s4u_comms.emplace_back(comms[i], false);
 525
 526   size_t pos = simgrid::s4u::Comm::wait_all_for(s4u_comms, timeout);
 527   for (size_t i = pos; i < count; i++)
 528     s4u_comms[i]->add_ref();
 529   return pos;
 530 }
 531
 532 ssize_t sg_comm_wait_any(sg_comm_t* comms, size_t count)
 533 {
 534   return sg_comm_wait_any_for(comms, count, -1);
 535 }
 536
 537 ssize_t sg_comm_wait_any_for(sg_comm_t* comms, size_t count, double timeout)
 538 {
 539   std::vector<simgrid::s4u::CommPtr> s4u_comms;
 540   for (size_t i = 0; i < count; i++)
 541     s4u_comms.emplace_back(comms[i], false);
 542
 543   ssize_t pos = simgrid::s4u::Comm::wait_any_for(s4u_comms, timeout);
 544   for (size_t i = 0; i < count; i++) {
 545     if (pos != -1 && static_cast<size_t>(pos) != i)
 546       s4u_comms[i]->add_ref();
 547   }
 548   return pos;
 549 }