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/sysdep.h"
   9 #include "xbt/log.h"
  10 #include "xbt/dict.h"
  11 #include "mc/mc.h"
  12 #include "src/mc/mc_replay.h"
  13 #include "src/surf/host_interface.hpp"
  14
  15 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_host, simix,
  16                                 "SIMIX hosts");
  17
  18 static void SIMIX_execution_finish(smx_synchro_t synchro);
  19
  20 /**
  21  * \brief Internal function to create a SIMIX host.
  22  * \param name name of the host to create
  23  */
  24 void SIMIX_host_create(sg_host_t host) // FIXME: braindead prototype. Take sg_host as parameter
  25 {
  26   smx_host_priv_t smx_host = xbt_new0(s_smx_host_priv_t, 1);
  27   s_smx_process_t proc;
  28
  29   /* Host structure */
  30   smx_host->process_list =
  31       xbt_swag_new(xbt_swag_offset(proc, host_proc_hookup));
  32
  33   /* Update global variables */
  34   sg_host_simix_set(host, smx_host);
  35 }
  36
  37 /**
  38  * \brief Start the host if it is off
  39  *
  40  */
  41 void SIMIX_host_on(sg_host_t h)
  42 {
  43   smx_host_priv_t host = sg_host_simix(h);
  44
  45   xbt_assert((host != NULL), "Invalid parameters");
  46
  47   if (h->is_off()) {
  48     simgrid::surf::Host* surf_host = h->extension<simgrid::surf::Host>();
  49     surf_host->turnOn();
  50
  51     unsigned int cpt;
  52     smx_process_arg_t arg;
  53     xbt_dynar_foreach(host->boot_processes,cpt,arg) {
  54
  55       char** argv = xbt_new(char*, arg->argc);
  56       for (int i=0; i<arg->argc; i++)
  57         argv[i] = xbt_strdup(arg->argv[i]);
  58
  59       XBT_DEBUG("Booting Process %s(%s) right now", arg->argv[0], arg->hostname);
  60       if (simix_global->create_process_function) {
  61         simix_global->create_process_function(argv[0],
  62                                               arg->code,
  63                                               NULL,
  64                                               arg->hostname,
  65                                               arg->kill_time,
  66                                               arg->argc,
  67                                               argv,
  68                                               arg->properties,
  69                                               arg->auto_restart,
  70                                               NULL);
  71       } else {
  72         simcall_process_create(arg->argv[0],
  73                                arg->code,
  74                                NULL,
  75                                arg->hostname,
  76                                arg->kill_time,
  77                                arg->argc,
  78                                argv,
  79                                arg->properties,
  80                                arg->auto_restart);
  81       }
  82     }
  83   }
  84 }
  85
  86 /**
  87  * \brief Stop the host if it is on
  88  *
  89  */
  90 void SIMIX_host_off(sg_host_t h, smx_process_t issuer)
  91 {
  92   smx_host_priv_t host = sg_host_simix(h);
  93
  94   xbt_assert((host != NULL), "Invalid parameters");
  95
  96   if (h->is_on()) {
  97     simgrid::surf::Host* surf_host = h->extension<simgrid::surf::Host>();
  98     surf_host->turnOff();
  99
 100     /* Clean Simulator data */
 101     if (xbt_swag_size(host->process_list) != 0) {
 102       smx_process_t process = NULL;
 103       xbt_swag_foreach(process, host->process_list) {
 104         SIMIX_process_kill(process, issuer);
 105         XBT_DEBUG("Killing %s on %s by %s", process->name,  sg_host_get_name(process->host), issuer->name);
 106       }
 107     }
 108   } else {
 109     XBT_INFO("Host %s is already off",h->name().c_str());
 110   }
 111 }
 112
 113 /**
 114  * \brief Internal function to destroy a SIMIX host.
 115  *
 116  * \param h the host to destroy (a sg_host_t)
 117  */
 118 void SIMIX_host_destroy(void *h)
 119 {
 120   smx_host_priv_t host = (smx_host_priv_t) h;
 121
 122   xbt_assert((host != NULL), "Invalid parameters");
 123
 124   /* Clean Simulator data */
 125   if (xbt_swag_size(host->process_list) != 0) {
 126     char *msg = xbt_strdup("Shutting down host, but it's not empty:");
 127     char *tmp;
 128     smx_process_t process = NULL;
 129
 130     xbt_swag_foreach(process, host->process_list) {
 131       tmp = bprintf("%s\n\t%s", msg, process->name);
 132       free(msg);
 133       msg = tmp;
 134     }
 135     SIMIX_display_process_status();
 136     THROWF(arg_error, 0, "%s", msg);
 137   }
 138   xbt_dynar_free(&host->auto_restart_processes);
 139   xbt_dynar_free(&host->boot_processes);
 140   xbt_swag_free(host->process_list);
 141
 142   /* Clean host structure */
 143   free(host);
 144   return;
 145 }
 146
 147 sg_host_t SIMIX_host_self(void)
 148 {
 149   smx_process_t process = SIMIX_process_self();
 150   return (process == NULL) ? NULL : SIMIX_process_get_host(process);
 151 }
 152
 153 /* needs to be public and without simcall because it is called
 154    by exceptions and logging events */
 155 const char* SIMIX_host_self_get_name(void)
 156 {
 157   sg_host_t host = SIMIX_host_self();
 158   if (host == NULL || SIMIX_process_self() == simix_global->maestro_process)
 159     return "";
 160
 161   return sg_host_get_name(host);
 162 }
 163
 164 void _SIMIX_host_free_process_arg(void *data)
 165 {
 166   smx_process_arg_t arg = *(smx_process_arg_t*)data;
 167   int i;
 168   for (i = 0; i < arg->argc; i++)
 169     xbt_free(arg->argv[i]);
 170   xbt_free(arg->argv);
 171   xbt_free(arg->name);
 172   xbt_free(arg);
 173 }
 174 /**
 175  * \brief Add a process to the list of the processes that the host will restart when it comes back
 176  * This function add a process to the list of the processes that will be restarted when the host comes
 177  * back. It is expected that this function is called when the host is down.
 178  * The processes will only be restarted once, meaning that you will have to register the process
 179  * again to restart the process again.
 180  */
 181 void SIMIX_host_add_auto_restart_process(sg_host_t host,
 182                                          const char *name,
 183                                          xbt_main_func_t code,
 184                                          void *data,
 185                                          const char *hostname,
 186                                          double kill_time,
 187                                          int argc, char **argv,
 188                                          xbt_dict_t properties,
 189                                          int auto_restart)
 190 {
 191   if (!sg_host_simix(host)->auto_restart_processes) {
 192     sg_host_simix(host)->auto_restart_processes = xbt_dynar_new(sizeof(smx_process_arg_t),_SIMIX_host_free_process_arg);
 193   }
 194   smx_process_arg_t arg = xbt_new(s_smx_process_arg_t,1);
 195   arg->name = xbt_strdup(name);
 196   arg->code = code;
 197   arg->data = data;
 198   arg->hostname = hostname;
 199   arg->kill_time = kill_time;
 200   arg->argc = argc;
 201
 202   arg->argv = xbt_new(char*,argc + 1);
 203
 204   int i;
 205   for (i = 0; i < argc; i++) {
 206     arg->argv[i] = xbt_strdup(argv[i]);
 207   }
 208   arg->argv[argc] = NULL;
 209
 210   arg->properties = properties;
 211   arg->auto_restart = auto_restart;
 212
 213   if( ! sg_host_is_on(host)
 214       && !xbt_dict_get_or_null(watched_hosts_lib,sg_host_get_name(host))){
 215     xbt_dict_set(watched_hosts_lib,sg_host_get_name(host),host,NULL);
 216     XBT_DEBUG("Have pushed host %s to watched_hosts_lib because state == SURF_RESOURCE_OFF",sg_host_get_name(host));
 217   }
 218   xbt_dynar_push_as(sg_host_simix(host)->auto_restart_processes,smx_process_arg_t,arg);
 219 }
 220 /**
 221  * \brief Restart the list of processes that have been registered to the host
 222  */
 223 void SIMIX_host_restart_processes(sg_host_t host)
 224 {
 225   unsigned int cpt;
 226   smx_process_arg_t arg;
 227   xbt_dynar_t process_list = sg_host_simix(host)->auto_restart_processes;
 228   if (!process_list)
 229     return;
 230
 231   xbt_dynar_foreach (process_list, cpt, arg) {
 232
 233     XBT_DEBUG("Restarting Process %s(%s) right now", arg->argv[0], arg->hostname);
 234     if (simix_global->create_process_function) {
 235       simix_global->create_process_function(arg->argv[0],
 236                                             arg->code,
 237                                             NULL,
 238                                             arg->hostname,
 239                                             arg->kill_time,
 240                                             arg->argc,
 241                                             arg->argv,
 242                                             arg->properties,
 243                                             arg->auto_restart,
 244                                             NULL);
 245     } else {
 246       simcall_process_create(arg->argv[0],
 247                              (xbt_main_func_t) arg->code,
 248                              NULL,
 249                              arg->hostname,
 250                              arg->kill_time,
 251                              arg->argc,
 252                              arg->argv,
 253                              arg->properties,
 254                              arg->auto_restart);
 255
 256     }
 257     /* arg->argv is used by the process created above.  Hide it to
 258      * _SIMIX_host_free_process_arg() which is called by xbt_dynar_reset()
 259      * below. */
 260     arg->argc = 0;
 261     arg->argv = NULL;
 262   }
 263   xbt_dynar_reset(process_list);
 264 }
 265
 266 void SIMIX_host_autorestart(sg_host_t host)
 267 {
 268   if(simix_global->autorestart)
 269     simix_global->autorestart(host);
 270   else
 271     xbt_die("No function for simix_global->autorestart");
 272 }
 273 smx_synchro_t simcall_HANDLER_process_execute(smx_simcall_t simcall,
 274                 const char* name, double flops_amount, double priority, double bound, unsigned long affinity_mask) {
 275         return SIMIX_process_execute(simcall->issuer, name,flops_amount,priority,bound,affinity_mask);
 276 }
 277 smx_synchro_t SIMIX_process_execute(smx_process_t issuer, const char *name,
 278      double flops_amount, double priority, double bound, unsigned long affinity_mask){
 279
 280   /* alloc structures and initialize */
 281   smx_synchro_t synchro = (smx_synchro_t) xbt_mallocator_get(simix_global->synchro_mallocator);
 282   synchro->type = SIMIX_SYNC_EXECUTE;
 283   synchro->name = xbt_strdup(name);
 284   synchro->state = SIMIX_RUNNING;
 285   synchro->execution.host = issuer->host;
 286   synchro->category = NULL;
 287
 288   /* set surf's action */
 289   if (!MC_is_active() && !MC_record_replay_is_active()) {
 290
 291     synchro->execution.surf_exec = issuer->host->pimpl_cpu->execute(flops_amount);
 292     synchro->execution.surf_exec->setData(synchro);
 293     synchro->execution.surf_exec->setPriority(priority);
 294
 295     if (bound != 0)
 296       static_cast<simgrid::surf::CpuAction*>(synchro->execution.surf_exec)
 297         ->setBound(bound);
 298
 299     if (affinity_mask != 0) {
 300       /* just a double check to confirm that this host is the host where this task is running. */
 301       xbt_assert(synchro->execution.host == issuer->host);
 302       static_cast<simgrid::surf::CpuAction*>(synchro->execution.surf_exec)
 303         ->setAffinity(issuer->host->pimpl_cpu, affinity_mask);
 304     }
 305   }
 306
 307   XBT_DEBUG("Create execute synchro %p: %s", synchro, synchro->name);
 308
 309   return synchro;
 310 }
 311
 312 smx_synchro_t SIMIX_process_parallel_execute(const char *name,
 313     int host_nb, sg_host_t *host_list,
 314     double *flops_amount, double *bytes_amount,
 315     double amount, double rate){
 316
 317   sg_host_t*host_list_cpy = NULL;
 318   int i;
 319
 320   /* alloc structures and initialize */
 321   smx_synchro_t synchro = (smx_synchro_t) xbt_mallocator_get(simix_global->synchro_mallocator);
 322   synchro->type = SIMIX_SYNC_PARALLEL_EXECUTE;
 323   synchro->name = xbt_strdup(name);
 324   synchro->state = SIMIX_RUNNING;
 325   synchro->execution.host = NULL; /* FIXME: do we need the list of hosts? */
 326   synchro->category = NULL;
 327
 328   /* set surf's synchro */
 329   host_list_cpy = xbt_new0(sg_host_t, host_nb);
 330   for (i = 0; i < host_nb; i++)
 331     host_list_cpy[i] = host_list[i];
 332
 333
 334   /* FIXME: what happens if host_list contains VMs and PMs. If
 335    * execute_parallel_task() does not change the state of the model, we can mix
 336    * them. */
 337   surf_host_model_t ws_model =
 338     host_list[0]->extension<simgrid::surf::Host>()->getModel();
 339   for (i = 1; i < host_nb; i++) {
 340     surf_host_model_t ws_model_tmp =
 341       host_list[i]->extension<simgrid::surf::Host>()->getModel();
 342     if (ws_model_tmp != ws_model) {
 343       XBT_CRITICAL("mixing VMs and PMs is not supported");
 344       DIE_IMPOSSIBLE;
 345     }
 346   }
 347
 348   /* set surf's synchro */
 349   if (!MC_is_active() && !MC_record_replay_is_active()) {
 350     synchro->execution.surf_exec =
 351       surf_host_model->executeParallelTask(
 352                   host_nb, host_list_cpy, flops_amount, bytes_amount, rate);
 353
 354     synchro->execution.surf_exec->setData(synchro);
 355   }
 356   XBT_DEBUG("Create parallel execute synchro %p", synchro);
 357
 358   return synchro;
 359 }
 360
 361 void SIMIX_process_execution_destroy(smx_synchro_t synchro){
 362   XBT_DEBUG("Destroy synchro %p", synchro);
 363
 364   if (synchro->execution.surf_exec) {
 365     synchro->execution.surf_exec->unref();
 366     synchro->execution.surf_exec = NULL;
 367   }
 368   xbt_free(synchro->name);
 369   xbt_mallocator_release(simix_global->synchro_mallocator, synchro);
 370 }
 371
 372 void SIMIX_process_execution_cancel(smx_synchro_t synchro){
 373   XBT_DEBUG("Cancel synchro %p", synchro);
 374
 375   if (synchro->execution.surf_exec)
 376     synchro->execution.surf_exec->cancel();
 377 }
 378
 379 double SIMIX_process_execution_get_remains(smx_synchro_t synchro){
 380   double result = 0.0;
 381
 382   if (synchro->state == SIMIX_RUNNING)
 383     result = surf_action_get_remains(synchro->execution.surf_exec);
 384
 385   return result;
 386 }
 387
 388 e_smx_state_t SIMIX_process_execution_get_state(smx_synchro_t synchro){
 389   return synchro->state;
 390 }
 391
 392 void SIMIX_process_execution_set_priority(smx_synchro_t synchro, double priority){
 393
 394   if(synchro->execution.surf_exec)
 395         synchro->execution.surf_exec->setPriority(priority);
 396 }
 397
 398 void SIMIX_process_execution_set_bound(smx_synchro_t synchro, double bound){
 399
 400   if(synchro->execution.surf_exec)
 401         static_cast<simgrid::surf::CpuAction*>(synchro->execution.surf_exec)
 402     ->setBound(bound);
 403 }
 404
 405 void SIMIX_process_execution_set_affinity(smx_synchro_t synchro, sg_host_t host, unsigned long mask){
 406   xbt_assert(synchro->type == SIMIX_SYNC_EXECUTE);
 407
 408   if (synchro->execution.surf_exec) {
 409     /* just a double check to confirm that this host is the host where this task is running. */
 410     xbt_assert(synchro->execution.host == host);
 411     static_cast<simgrid::surf::CpuAction*>(synchro->execution.surf_exec)
 412       ->setAffinity(host->pimpl_cpu, mask);
 413   }
 414 }
 415
 416 void simcall_HANDLER_process_execution_wait(smx_simcall_t simcall, smx_synchro_t synchro){
 417
 418   XBT_DEBUG("Wait for execution of synchro %p, state %d", synchro, (int)synchro->state);
 419
 420   /* Associate this simcall to the synchro */
 421   xbt_fifo_push(synchro->simcalls, simcall);
 422   simcall->issuer->waiting_synchro = synchro;
 423
 424   /* set surf's synchro */
 425   if (MC_is_active() || MC_record_replay_is_active()) {
 426     synchro->state = SIMIX_DONE;
 427     SIMIX_execution_finish(synchro);
 428     return;
 429   }
 430
 431   /* If the synchro is already finished then perform the error handling */
 432   if (synchro->state != SIMIX_RUNNING)
 433     SIMIX_execution_finish(synchro);
 434 }
 435
 436 void SIMIX_host_execution_suspend(smx_synchro_t synchro)
 437 {
 438   if(synchro->execution.surf_exec)
 439     synchro->execution.surf_exec->suspend();
 440 }
 441
 442 void SIMIX_host_execution_resume(smx_synchro_t synchro)
 443 {
 444   if(synchro->execution.surf_exec)
 445     synchro->execution.surf_exec->resume();
 446 }
 447
 448 void SIMIX_execution_finish(smx_synchro_t synchro)
 449 {
 450   xbt_fifo_item_t item;
 451   smx_simcall_t simcall;
 452
 453   xbt_fifo_foreach(synchro->simcalls, item, simcall, smx_simcall_t) {
 454
 455     switch (synchro->state) {
 456
 457       case SIMIX_DONE:
 458         /* do nothing, synchro done */
 459         XBT_DEBUG("SIMIX_execution_finished: execution successful");
 460         break;
 461
 462       case SIMIX_FAILED:
 463         XBT_DEBUG("SIMIX_execution_finished: host '%s' failed", sg_host_get_name(simcall->issuer->host));
 464         simcall->issuer->context->iwannadie = 1;
 465         SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
 466         break;
 467
 468       case SIMIX_CANCELED:
 469         XBT_DEBUG("SIMIX_execution_finished: execution canceled");
 470         SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Canceled");
 471         break;
 472
 473       default:
 474         xbt_die("Internal error in SIMIX_execution_finish: unexpected synchro state %d",
 475             (int)synchro->state);
 476     }
 477     /* check if the host is down */
 478     if (simcall->issuer->host->is_off()) {
 479       simcall->issuer->context->iwannadie = 1;
 480     }
 481
 482     simcall->issuer->waiting_synchro =    NULL;
 483     simcall_process_execution_wait__set__result(simcall, synchro->state);
 484     SIMIX_simcall_answer(simcall);
 485   }
 486
 487   /* We no longer need it */
 488   SIMIX_process_execution_destroy(synchro);
 489 }
 490
 491
 492 void SIMIX_post_host_execute(smx_synchro_t synchro)
 493 {
 494   if (synchro->type == SIMIX_SYNC_EXECUTE && /* FIMXE: handle resource failure
 495                                                * for parallel tasks too */
 496       synchro->execution.host->is_off()) {
 497     /* If the host running the synchro failed, notice it so that the asking
 498      * process can be killed if it runs on that host itself */
 499     synchro->state = SIMIX_FAILED;
 500   } else if (synchro->execution.surf_exec->getState() == SURF_ACTION_FAILED) {
 501     /* If the host running the synchro didn't fail, then the synchro was
 502      * canceled */
 503     synchro->state = SIMIX_CANCELED;
 504   } else {
 505     synchro->state = SIMIX_DONE;
 506   }
 507
 508   if (synchro->execution.surf_exec) {
 509     synchro->execution.surf_exec->unref();
 510     synchro->execution.surf_exec = NULL;
 511   }
 512
 513   /* If there are simcalls associated with the synchro, then answer them */
 514   if (xbt_fifo_size(synchro->simcalls)) {
 515     SIMIX_execution_finish(synchro);
 516   }
 517 }
 518
 519
 520 void SIMIX_set_category(smx_synchro_t synchro, const char *category)
 521 {
 522   if (synchro->state != SIMIX_RUNNING) return;
 523   if (synchro->type == SIMIX_SYNC_EXECUTE){
 524     synchro->execution.surf_exec->setCategory(category);
 525   }else if (synchro->type == SIMIX_SYNC_COMMUNICATE){
 526     synchro->comm.surf_comm->setCategory(category);
 527   }
 528 }