src/simix/smx_host.cpp

   1 /* Copyright (c) 2007-2015. The SimGrid Team.
   2  * All rights reserved.                                                     */
   3
   4 /* This program is free software; you can redistribute it and/or modify it
   5  * under the terms of the license (GNU LGPL) which comes with this package. */
   6
   7 #include "smx_private.h"
   8 #include <xbt/ex.hpp>
   9 #include "xbt/sysdep.h"
  10 #include "mc/mc.h"
  11 #include "src/mc/mc_replay.h"
  12 #include "src/surf/virtual_machine.hpp"
  13 #include "src/surf/HostImpl.hpp"
  14
  15 #include "src/kernel/activity/SynchroExec.hpp"
  16 #include "src/kernel/activity/SynchroComm.hpp"
  17
  18 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_host, simix, "SIMIX hosts");
  19
  20 namespace simgrid {
  21   namespace simix {
  22     Host::Host()
  23     {
  24       simgrid::simix::ActorImpl act;
  25       process_list = xbt_swag_new(xbt_swag_offset(act, host_proc_hookup));
  26     }
  27   }
  28 }
  29
  30 /** @brief Internal function to create a SIMIX host. */
  31 void SIMIX_host_create(sg_host_t host)
  32 {
  33   sg_host_simix_set(host, new simgrid::simix::Host());
  34 }
  35
  36 /** @brief Start the host if it is off */
  37 void SIMIX_host_on(sg_host_t h)
  38 {
  39   smx_host_priv_t host = sg_host_simix(h);
  40
  41   xbt_assert((host != nullptr), "Invalid parameters");
  42
  43   if (h->isOff()) {
  44     simgrid::surf::HostImpl* surf_host = h->extension<simgrid::surf::HostImpl>();
  45     surf_host->turnOn();
  46
  47     unsigned int cpt;
  48     smx_process_arg_t arg;
  49     xbt_dynar_foreach(host->boot_processes,cpt,arg) {
  50       XBT_DEBUG("Booting Process %s(%s) right now",
  51         arg->name.c_str(), arg->hostname);
  52       if (simix_global->create_process_function) {
  53         simix_global->create_process_function(arg->name.c_str(),
  54                                               arg->code,
  55                                               nullptr,
  56                                               arg->hostname,
  57                                               arg->kill_time,
  58                                               arg->properties,
  59                                               arg->auto_restart,
  60                                               nullptr);
  61       } else {
  62         simcall_process_create(arg->name.c_str(),
  63                                arg->code,
  64                                nullptr,
  65                                arg->hostname,
  66                                arg->kill_time,
  67                                arg->properties,
  68                                arg->auto_restart);
  69       }
  70     }
  71   }
  72 }
  73
  74 /** @brief Stop the host if it is on */
  75 void SIMIX_host_off(sg_host_t h, smx_actor_t issuer)
  76 {
  77   smx_host_priv_t host = sg_host_simix(h);
  78
  79   xbt_assert((host != nullptr), "Invalid parameters");
  80
  81   if (h->isOn()) {
  82     simgrid::surf::HostImpl* surf_host = h->extension<simgrid::surf::HostImpl>();
  83     surf_host->turnOff();
  84
  85     /* Clean Simulator data */
  86     if (xbt_swag_size(host->process_list) != 0) {
  87       smx_actor_t process = nullptr;
  88       xbt_swag_foreach(process, host->process_list) {
  89         SIMIX_process_kill(process, issuer);
  90         XBT_DEBUG("Killing %s on %s by %s",
  91           process->name.c_str(),  sg_host_get_name(process->host),
  92           issuer->name.c_str());
  93       }
  94     }
  95   } else {
  96     XBT_INFO("Host %s is already off", h->name().c_str());
  97   }
  98 }
  99
 100 /**
 101  * \brief Internal function to destroy a SIMIX host.
 102  *
 103  * \param h the host to destroy (a sg_host_t)
 104  */
 105 void SIMIX_host_destroy(void *h)
 106 {
 107   smx_host_priv_t host = static_cast<smx_host_priv_t>(h);
 108
 109   xbt_assert((host != nullptr), "Invalid parameters");
 110
 111   /* Clean Simulator data */
 112   if (xbt_swag_size(host->process_list) != 0) {
 113     char *msg = xbt_strdup("Shutting down host, but it's not empty:");
 114     char *tmp;
 115     smx_actor_t process = nullptr;
 116
 117     xbt_swag_foreach(process, host->process_list) {
 118       tmp = bprintf("%s\n\t%s", msg, process->name.c_str());
 119       free(msg);
 120       msg = tmp;
 121     }
 122     SIMIX_display_process_status();
 123     THROWF(arg_error, 0, "%s", msg);
 124   }
 125   xbt_dynar_free(&host->auto_restart_processes);
 126   xbt_dynar_free(&host->boot_processes);
 127   xbt_swag_free(host->process_list);
 128
 129   /* Clean host structure */
 130   free(host);
 131   return;
 132 }
 133
 134 sg_host_t SIMIX_host_self()
 135 {
 136   smx_actor_t process = SIMIX_process_self();
 137   return (process == nullptr) ? nullptr : process->host;
 138 }
 139
 140 /* needs to be public and without simcall for exceptions and logging events */
 141 const char* SIMIX_host_self_get_name()
 142 {
 143   sg_host_t host = SIMIX_host_self();
 144   if (host == nullptr || SIMIX_process_self() == simix_global->maestro_process)
 145     return "";
 146
 147   return sg_host_get_name(host);
 148 }
 149
 150 void _SIMIX_host_free_process_arg(void *data)
 151 {
 152   smx_process_arg_t arg = *(static_cast<smx_process_arg_t*>(data));
 153   delete arg;
 154 }
 155 /**
 156  * \brief Add a process to the list of the processes that the host will restart when it comes back
 157  * This function add a process to the list of the processes that will be restarted when the host comes
 158  * back. It is expected that this function is called when the host is down.
 159  * The processes will only be restarted once, meaning that you will have to register the process
 160  * again to restart the process again.
 161  */
 162 void SIMIX_host_add_auto_restart_process(
 163   sg_host_t host, const char *name, std::function<void()> code,
 164   void* data, const char *hostname, double kill_time,
 165   xbt_dict_t properties, int auto_restart)
 166 {
 167   if (!sg_host_simix(host)->auto_restart_processes) {
 168     sg_host_simix(host)->auto_restart_processes = xbt_dynar_new(sizeof(smx_process_arg_t),_SIMIX_host_free_process_arg);
 169   }
 170   smx_process_arg_t arg = new simgrid::simix::ProcessArg();
 171   arg->name = name;
 172   arg->code = std::move(code);
 173   arg->data = data;
 174   arg->hostname = hostname;
 175   arg->kill_time = kill_time;
 176   arg->properties = properties;
 177   arg->auto_restart = auto_restart;
 178
 179   if( host->isOff() && !xbt_dict_get_or_null(watched_hosts_lib,sg_host_get_name(host))){
 180     xbt_dict_set(watched_hosts_lib,sg_host_get_name(host),host,nullptr);
 181     XBT_DEBUG("Push host %s to watched_hosts_lib because state == SURF_RESOURCE_OFF",sg_host_get_name(host));
 182   }
 183   xbt_dynar_push_as(sg_host_simix(host)->auto_restart_processes,smx_process_arg_t,arg);
 184 }
 185 /** @brief Restart the list of processes that have been registered to the host */
 186 void SIMIX_host_autorestart(sg_host_t host)
 187 {
 188   unsigned int cpt;
 189   smx_process_arg_t arg;
 190   xbt_dynar_t process_list = sg_host_simix(host)->auto_restart_processes;
 191   if (!process_list)
 192     return;
 193
 194   xbt_dynar_foreach (process_list, cpt, arg) {
 195
 196     XBT_DEBUG("Restarting Process %s(%s) right now", arg->name.c_str(), arg->hostname);
 197     if (simix_global->create_process_function) {
 198       simix_global->create_process_function(arg->name.c_str(), arg->code, nullptr, arg->hostname, arg->kill_time,
 199                                             arg->properties, arg->auto_restart, nullptr);
 200     } else {
 201       simcall_process_create(arg->name.c_str(), arg->code, nullptr, arg->hostname, arg->kill_time, arg->properties,
 202                              arg->auto_restart);
 203     }
 204   }
 205   xbt_dynar_reset(process_list);
 206 }
 207
 208 smx_activity_t simcall_HANDLER_execution_start(smx_simcall_t simcall, const char* name, double flops_amount,
 209                                               double priority, double bound) {
 210   return SIMIX_execution_start(simcall->issuer, name,flops_amount,priority,bound);
 211 }
 212
 213 smx_activity_t SIMIX_execution_start(smx_actor_t issuer, const char *name, double flops_amount, double priority,
 214                                     double bound){
 215
 216   /* alloc structures and initialize */
 217   simgrid::kernel::activity::Exec *exec = new simgrid::kernel::activity::Exec(name, issuer->host);
 218
 219   /* set surf's action */
 220   if (!MC_is_active() && !MC_record_replay_is_active()) {
 221
 222     exec->surf_exec = issuer->host->pimpl_cpu->execution_start(flops_amount);
 223     exec->surf_exec->setData(exec);
 224     exec->surf_exec->setPriority(priority);
 225
 226     if (bound > 0)
 227       static_cast<simgrid::surf::CpuAction*>(exec->surf_exec)->setBound(bound);
 228   }
 229
 230   XBT_DEBUG("Create execute synchro %p: %s", exec, exec->name.c_str());
 231
 232   return exec;
 233 }
 234
 235 smx_activity_t SIMIX_execution_parallel_start(const char *name, int host_nb, sg_host_t *host_list, double *flops_amount,
 236                                              double *bytes_amount, double amount, double rate){
 237
 238   /* alloc structures and initialize */
 239   simgrid::kernel::activity::Exec *exec = new simgrid::kernel::activity::Exec(name, nullptr);
 240
 241   /* set surf's synchro */
 242   sg_host_t *host_list_cpy = xbt_new0(sg_host_t, host_nb);
 243   for (int i = 0; i < host_nb; i++)
 244     host_list_cpy[i] = host_list[i];
 245
 246   /* Check that we are not mixing VMs and PMs in the parallel task */
 247   simgrid::surf::HostImpl *host = host_list[0]->extension<simgrid::surf::HostImpl>();
 248   bool is_a_vm = (nullptr != dynamic_cast<simgrid::surf::VirtualMachine*>(host));
 249   for (int i = 1; i < host_nb; i++) {
 250     bool tmp_is_a_vm = (nullptr != dynamic_cast<simgrid::surf::VirtualMachine*>(host_list[i]->extension<simgrid::surf::HostImpl>()));
 251     xbt_assert(is_a_vm == tmp_is_a_vm, "parallel_execute: mixing VMs and PMs is not supported (yet).");
 252   }
 253
 254   /* set surf's synchro */
 255   if (!MC_is_active() && !MC_record_replay_is_active()) {
 256     exec->surf_exec = surf_host_model->executeParallelTask(host_nb, host_list_cpy, flops_amount, bytes_amount, rate);
 257     exec->surf_exec->setData(exec);
 258   }
 259   XBT_DEBUG("Create parallel execute synchro %p", exec);
 260
 261   return exec;
 262 }
 263
 264 void SIMIX_execution_cancel(smx_activity_t synchro)
 265 {
 266   XBT_DEBUG("Cancel synchro %p", synchro);
 267   simgrid::kernel::activity::Exec *exec = static_cast<simgrid::kernel::activity::Exec *>(synchro);
 268
 269   if (exec->surf_exec)
 270     exec->surf_exec->cancel();
 271 }
 272
 273 void SIMIX_execution_set_priority(smx_activity_t synchro, double priority)
 274 {
 275   simgrid::kernel::activity::Exec *exec = static_cast<simgrid::kernel::activity::Exec *>(synchro);
 276   if(exec->surf_exec)
 277     exec->surf_exec->setPriority(priority);
 278 }
 279
 280 void SIMIX_execution_set_bound(smx_activity_t synchro, double bound)
 281 {
 282   simgrid::kernel::activity::Exec *exec = static_cast<simgrid::kernel::activity::Exec *>(synchro);
 283   if(exec->surf_exec)
 284     static_cast<simgrid::surf::CpuAction*>(exec->surf_exec)->setBound(bound);
 285 }
 286
 287 void simcall_HANDLER_execution_wait(smx_simcall_t simcall, smx_activity_t synchro)
 288 {
 289   simgrid::kernel::activity::Exec *exec = static_cast<simgrid::kernel::activity::Exec *>(synchro);
 290   XBT_DEBUG("Wait for execution of synchro %p, state %d", synchro, (int)synchro->state);
 291
 292   /* Associate this simcall to the synchro */
 293   synchro->simcalls.push_back(simcall);
 294   simcall->issuer->waiting_synchro = synchro;
 295
 296   /* set surf's synchro */
 297   if (MC_is_active() || MC_record_replay_is_active()) {
 298     synchro->state = SIMIX_DONE;
 299     SIMIX_execution_finish(exec);
 300     return;
 301   }
 302
 303   /* If the synchro is already finished then perform the error handling */
 304   if (synchro->state != SIMIX_RUNNING)
 305     SIMIX_execution_finish(exec);
 306 }
 307
 308 void SIMIX_execution_finish(simgrid::kernel::activity::Exec *exec)
 309 {
 310   for (smx_simcall_t simcall : exec->simcalls) {
 311     switch (exec->state) {
 312
 313       case SIMIX_DONE:
 314         /* do nothing, synchro done */
 315         XBT_DEBUG("SIMIX_execution_finished: execution successful");
 316         break;
 317
 318       case SIMIX_FAILED:
 319         XBT_DEBUG("SIMIX_execution_finished: host '%s' failed", sg_host_get_name(simcall->issuer->host));
 320         simcall->issuer->context->iwannadie = 1;
 321         SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
 322         break;
 323
 324       case SIMIX_CANCELED:
 325         XBT_DEBUG("SIMIX_execution_finished: execution canceled");
 326         SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Canceled");
 327         break;
 328
 329       default:
 330         xbt_die("Internal error in SIMIX_execution_finish: unexpected synchro state %d",
 331             (int)exec->state);
 332     }
 333     /* Fail the process if the host is down */
 334     if (simcall->issuer->host->isOff())
 335       simcall->issuer->context->iwannadie = 1;
 336
 337     simcall->issuer->waiting_synchro = nullptr;
 338     simcall_execution_wait__set__result(simcall, exec->state);
 339     SIMIX_simcall_answer(simcall);
 340   }
 341
 342   /* We no longer need it */
 343   exec->unref();
 344 }
 345
 346 void SIMIX_set_category(smx_activity_t synchro, const char *category)
 347 {
 348   if (synchro->state != SIMIX_RUNNING)
 349     return;
 350
 351   simgrid::kernel::activity::Exec *exec = dynamic_cast<simgrid::kernel::activity::Exec *>(synchro);
 352   if (exec != nullptr) {
 353     exec->surf_exec->setCategory(category);
 354     return;
 355   }
 356
 357   simgrid::kernel::activity::Comm *comm = dynamic_cast<simgrid::kernel::activity::Comm *>(synchro);
 358   if (comm != nullptr) {
 359     comm->surf_comm->setCategory(category);
 360     return;
 361   }
 362 }