src/simix/smx_host.c

   1 /* Copyright (c) 2007-2012. 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
  13 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_host, simix,
  14                                 "Logging specific to SIMIX (hosts)");
  15
  16 static void SIMIX_execution_finish(smx_action_t action);
  17
  18 /**
  19  * \brief Internal function to create a SIMIX host.
  20  * \param name name of the host to create
  21  * \param workstation the SURF workstation to encapsulate
  22  * \param data some user data (may be NULL)
  23  */
  24 smx_host_t SIMIX_host_create(const char *name,
  25                                void *workstation, void *data)
  26 {
  27   smx_host_t smx_host = xbt_new0(s_smx_host_t, 1);
  28   s_smx_process_t proc;
  29
  30   /* Host structure */
  31   smx_host->name = xbt_strdup(name);
  32   smx_host->data = data;
  33   smx_host->host = workstation;
  34   smx_host->process_list =
  35       xbt_swag_new(xbt_swag_offset(proc, host_proc_hookup));
  36
  37   /* Update global variables */
  38   xbt_lib_set(host_lib,smx_host->name,SIMIX_HOST_LEVEL,smx_host);
  39
  40   return smx_host;
  41 }
  42
  43 /**
  44  * \brief Internal function to destroy a SIMIX host.
  45  *
  46  * \param h the host to destroy (a smx_host_t)
  47  */
  48 void SIMIX_host_destroy(void *h)
  49 {
  50   smx_host_t host = (smx_host_t) h;
  51
  52   xbt_assert((host != NULL), "Invalid parameters");
  53
  54   /* Clean Simulator data */
  55   if (xbt_swag_size(host->process_list) != 0) {
  56     char *msg =
  57         bprintf("Shutting down host %s, but it's not empty:", host->name);
  58     char *tmp;
  59     smx_process_t process = NULL;
  60
  61     xbt_swag_foreach(process, host->process_list) {
  62       tmp = bprintf("%s\n\t%s", msg, process->name);
  63       free(msg);
  64       msg = tmp;
  65     }
  66     SIMIX_display_process_status();
  67     THROWF(arg_error, 0, "%s", msg);
  68   }
  69   xbt_dynar_free(&host->auto_restart_processes);
  70   xbt_swag_free(host->process_list);
  71
  72   /* Clean host structure */
  73   free(host->name);
  74   free(host);
  75
  76   return;
  77 }
  78
  79 ///**
  80 // * \brief Returns a dict of all hosts.
  81 // *
  82 // * \return List of all hosts (as a #xbt_dict_t)
  83 // */
  84 //xbt_dict_t SIMIX_host_get_dict(void)
  85 //{
  86 //  xbt_dict_t host_dict = xbt_dict_new_homogeneous(NULL);
  87 //  xbt_lib_cursor_t cursor = NULL;
  88 //  char *name = NULL;
  89 //  void **host = NULL;
  90 //
  91 //  xbt_lib_foreach(host_lib, cursor, name, host){
  92 //    if(host[SIMIX_HOST_LEVEL])
  93 //            xbt_dict_set(host_dict,name,host[SIMIX_HOST_LEVEL], NULL);
  94 //  }
  95 //  return host_dict;
  96 //}
  97
  98 smx_host_t SIMIX_host_get_by_name(u_smx_scalar_t *args)
  99 {
 100   const char *name = args[0].cc;
 101   xbt_assert(((simix_global != NULL)
 102                && (host_lib != NULL)),
 103               "Environment not set yet");
 104
 105   return xbt_lib_get_or_null(host_lib, name, SIMIX_HOST_LEVEL);
 106 }
 107
 108 smx_host_t SIMIX_host_self(void)
 109 {
 110   smx_process_t process = SIMIX_process_self();
 111   return (process == NULL) ? NULL : SIMIX_process_get_host(process);
 112 }
 113
 114 /* needs to be public and without simcall because it is called
 115    by exceptions and logging events */
 116 const char* SIMIX_host_self_get_name(void)
 117 {
 118   smx_host_t host = SIMIX_host_self();
 119   if (host == NULL || SIMIX_process_self() == simix_global->maestro_process)
 120     return "";
 121
 122   return SIMIX_host_get_name(SIMIX_pack_args(PTR(host)));
 123 }
 124
 125 const char* SIMIX_host_get_name(u_smx_scalar_t *args)
 126 {
 127   smx_host_t host = args[0].p;
 128   xbt_assert((host != NULL), "Invalid parameters");
 129
 130   return host->name;
 131 }
 132
 133 xbt_dict_t SIMIX_host_get_properties(u_smx_scalar_t *args)
 134 {
 135   smx_host_t host = args[0].p;
 136   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 137
 138   return surf_workstation_model->extension.workstation.get_properties(host->host);
 139 }
 140
 141 double SIMIX_host_get_speed(u_smx_scalar_t *args)
 142 {
 143   smx_host_t host = args[0].p;
 144   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 145
 146   return surf_workstation_model->extension.workstation.
 147       get_speed(host->host, 1.0);
 148 }
 149
 150 double SIMIX_host_get_available_speed(u_smx_scalar_t *args)
 151 {
 152   smx_host_t host = args[0].p;
 153   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 154
 155   return surf_workstation_model->extension.workstation.
 156       get_available_speed(host->host);
 157 }
 158
 159 int SIMIX_host_get_state(u_smx_scalar_t *args)
 160 {
 161   smx_host_t host = args[0].p;
 162   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 163
 164   return surf_workstation_model->extension.workstation.
 165       get_state(host->host);
 166 }
 167
 168 void* SIMIX_host_self_get_data(void)
 169 {
 170   smx_host_t self = SIMIX_host_self();
 171   return SIMIX_host_get_data(SIMIX_pack_args(PTR(self)));
 172 }
 173
 174 void SIMIX_host_self_set_data(void *data)
 175 {
 176   smx_host_t self = SIMIX_host_self();
 177   SIMIX_host_set_data(SIMIX_pack_args(PTR(self), PTR(data)));
 178 }
 179
 180 void* SIMIX_host_get_data(u_smx_scalar_t *args)
 181 {
 182   smx_host_t host = args[0].p;
 183   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 184
 185   return host->data;
 186 }
 187 void _SIMIX_host_free_process_arg(void *);
 188 void _SIMIX_host_free_process_arg(void *data)
 189 {
 190   smx_process_arg_t arg = *(void**)data;
 191   xbt_free(arg->name);
 192   xbt_free(arg);
 193 }
 194 /**
 195  * \brief Add a process to the list of the processes that the host will restart when it comes back
 196  * This function add a process to the list of the processes that will be restarted when the host comes
 197  * back. It is expected that this function is called when the host is down.
 198  * The processes will only be restarted once, meaning that you will have to register the process
 199  * again to restart the process again.
 200  */
 201 void SIMIX_host_add_auto_restart_process(smx_host_t host,
 202                                          const char *name,
 203                                          xbt_main_func_t code,
 204                                          void *data,
 205                                          const char *hostname,
 206                                          double kill_time,
 207                                          int argc, char **argv,
 208                                          xbt_dict_t properties,
 209                                          int auto_restart)
 210 {
 211   if (!host->auto_restart_processes) {
 212     host->auto_restart_processes = xbt_dynar_new(sizeof(smx_process_arg_t),_SIMIX_host_free_process_arg);
 213   }
 214   smx_process_arg_t arg = xbt_new(s_smx_process_arg_t,1);
 215   arg->name = xbt_strdup(name);
 216   arg->code = code;
 217   arg->data = data;
 218   arg->hostname = hostname;
 219   arg->kill_time = kill_time;
 220   arg->argc = argc;
 221
 222   arg->argv = xbt_new(char*,argc + 1);
 223
 224   int i;
 225   for (i = 0; i < argc; i++) {
 226     arg->argv[i] = xbt_strdup(argv[i]);
 227   }
 228   arg->argv[argc] = NULL;
 229
 230   arg->properties = properties;
 231   arg->auto_restart = auto_restart;
 232
 233   if( SIMIX_host_get_state(SIMIX_pack_args(PTR(host))) == SURF_RESOURCE_OFF
 234       && !xbt_dict_get_or_null(watched_hosts_lib,host->name)){
 235     xbt_dict_set(watched_hosts_lib,host->name,host,NULL);
 236     XBT_DEBUG("Have push host %s to watched_hosts_lib because state == SURF_RESOURCE_OFF",host->name);
 237   }
 238   xbt_dynar_push_as(host->auto_restart_processes,smx_process_arg_t,arg);
 239 }
 240 /**
 241  * \brief Restart the list of processes that have been registered to the host
 242  */
 243 void SIMIX_host_restart_processes(smx_host_t host)
 244 {
 245   unsigned int cpt;
 246   smx_process_arg_t arg;
 247   xbt_dynar_foreach(host->auto_restart_processes,cpt,arg) {
 248
 249     smx_process_t process;
 250
 251     XBT_DEBUG("Restarting Process %s(%s) right now", arg->argv[0], arg->hostname);
 252     if (simix_global->create_process_function) {
 253       simix_global->create_process_function(&process,
 254                                             arg->argv[0],
 255                                             arg->code,
 256                                             NULL,
 257                                             arg->hostname,
 258                                             arg->kill_time,
 259                                             arg->argc,
 260                                             arg->argv,
 261                                             arg->properties,
 262                                             arg->auto_restart);
 263     }
 264     else {
 265       simcall_process_create(&process,
 266                                             arg->argv[0],
 267                                             arg->code,
 268                                             NULL,
 269                                             arg->hostname,
 270                                             arg->kill_time,
 271                                             arg->argc,
 272                                             arg->argv,
 273                                             arg->properties,
 274                                             arg->auto_restart);
 275
 276     }
 277   }
 278   xbt_dynar_reset(host->auto_restart_processes);
 279 }
 280
 281 void SIMIX_host_autorestart(smx_host_t host)
 282 {
 283   if(simix_global->autorestart)
 284     simix_global->autorestart(host);
 285   else
 286     xbt_die("No function for simix_global->autorestart");
 287 }
 288
 289 void SIMIX_host_set_data(u_smx_scalar_t *args)
 290 {
 291   smx_host_t host = args[0].p;
 292   void *data = args[1].p;
 293   xbt_assert((host != NULL), "Invalid parameters");
 294   xbt_assert((host->data == NULL), "Data already set");
 295
 296   host->data = data;
 297 }
 298
 299 smx_action_t SIMIX_host_execute(u_smx_scalar_t args[])
 300 {
 301   const char *name = args[0].p;
 302   smx_host_t host = args[1].p;
 303   double computation_amount = args[2].d;
 304   double priority = args[3].d;
 305
 306   /* alloc structures and initialize */
 307   smx_action_t action = xbt_mallocator_get(simix_global->action_mallocator);
 308   action->type = SIMIX_ACTION_EXECUTE;
 309   action->name = xbt_strdup(name);
 310   action->state = SIMIX_RUNNING;
 311   action->execution.host = host;
 312
 313 #ifdef HAVE_TRACING
 314   action->category = NULL;
 315 #endif
 316
 317   /* set surf's action */
 318   if (!MC_is_active()) {
 319     action->execution.surf_exec =
 320       surf_workstation_model->extension.workstation.execute(host->host,
 321     computation_amount);
 322     surf_workstation_model->action_data_set(action->execution.surf_exec, action);
 323     surf_workstation_model->set_priority(action->execution.surf_exec, priority);
 324   }
 325
 326   XBT_DEBUG("Create execute action %p", action);
 327
 328   return action;
 329 }
 330
 331 smx_action_t SIMIX_host_parallel_execute(u_smx_scalar_t *args)
 332 {
 333   const char *name = args[0].cc;
 334   int host_nb = args[1].i;
 335   smx_host_t *host_list = args[2].p;
 336   double *computation_amount = args[3].p;
 337   double *communication_amount = args[4].p;
 338   double amount = args[5].d;
 339   double rate = args[6].d;
 340   void **workstation_list = NULL;
 341   int i;
 342
 343   /* alloc structures and initialize */
 344   smx_action_t action = xbt_mallocator_get(simix_global->action_mallocator);
 345   action->type = SIMIX_ACTION_PARALLEL_EXECUTE;
 346   action->name = xbt_strdup(name);
 347   action->state = SIMIX_RUNNING;
 348   action->execution.host = NULL; /* FIXME: do we need the list of hosts? */
 349
 350 #ifdef HAVE_TRACING
 351   action->category = NULL;
 352 #endif
 353
 354   /* set surf's action */
 355   workstation_list = xbt_new0(void *, host_nb);
 356   for (i = 0; i < host_nb; i++)
 357     workstation_list[i] = host_list[i]->host;
 358
 359   /* set surf's action */
 360   if (!MC_is_active()) {
 361     action->execution.surf_exec =
 362       surf_workstation_model->extension.workstation.
 363       execute_parallel_task(host_nb, workstation_list, computation_amount,
 364                       communication_amount, rate);
 365
 366     surf_workstation_model->action_data_set(action->execution.surf_exec, action);
 367   }
 368   XBT_DEBUG("Create parallel execute action %p", action);
 369
 370   return action;
 371 }
 372
 373 void SIMIX_host_execution_destroy(u_smx_scalar_t *args)
 374 {
 375   smx_action_t action = args[0].p;
 376   XBT_DEBUG("Destroy action %p", action);
 377
 378   if (action->execution.surf_exec) {
 379     surf_workstation_model->action_unref(action->execution.surf_exec);
 380     action->execution.surf_exec = NULL;
 381   }
 382   xbt_free(action->name);
 383   xbt_mallocator_release(simix_global->action_mallocator, action);
 384 }
 385
 386 void SIMIX_host_execution_cancel(u_smx_scalar_t *args)
 387 {
 388   smx_action_t action = args[0].p;
 389   XBT_DEBUG("Cancel action %p", action);
 390
 391   if (action->execution.surf_exec)
 392     surf_workstation_model->action_cancel(action->execution.surf_exec);
 393 }
 394
 395 double SIMIX_host_execution_get_remains(u_smx_scalar_t *args)
 396 {
 397   smx_action_t action = args[0].p;
 398   double result = 0.0;
 399
 400   if (action->state == SIMIX_RUNNING)
 401     result = surf_workstation_model->get_remains(action->execution.surf_exec);
 402
 403   return result;
 404 }
 405
 406 e_smx_state_t SIMIX_host_execution_get_state(u_smx_scalar_t *args)
 407 {
 408   smx_action_t action = args[0].p;
 409   return action->state;
 410 }
 411
 412 void SIMIX_host_execution_set_priority(u_smx_scalar_t *args)
 413 {
 414   smx_action_t action = args[0].p;
 415   double priority = args[1].d;
 416   if(action->execution.surf_exec)
 417     surf_workstation_model->set_priority(action->execution.surf_exec, priority);
 418 }
 419
 420 void SIMIX_pre_host_execution_wait(u_smx_scalar_t *args)
 421 {
 422   smx_action_t action = args[0].p;
 423
 424   XBT_DEBUG("Wait for execution of action %p, state %d", action, (int)action->state);
 425
 426   /* Associate this simcall to the action */
 427   xbt_fifo_push(action->simcalls, simcall);
 428   simcall->issuer->waiting_action = action;
 429
 430   /* set surf's action */
 431   if (MC_is_active()) {
 432     action->state = SIMIX_DONE;
 433     SIMIX_execution_finish(action);
 434     return;
 435   }
 436
 437   /* If the action is already finished then perform the error handling */
 438   if (action->state != SIMIX_RUNNING)
 439     SIMIX_execution_finish(action);
 440 }
 441
 442 void SIMIX_host_execution_suspend(smx_action_t action)
 443 {
 444   if(action->execution.surf_exec)
 445     surf_workstation_model->suspend(action->execution.surf_exec);
 446 }
 447
 448 void SIMIX_host_execution_resume(smx_action_t action)
 449 {
 450   if(action->execution.surf_exec)
 451     surf_workstation_model->resume(action->execution.surf_exec);
 452 }
 453
 454 void SIMIX_execution_finish(smx_action_t action)
 455 {
 456   xbt_fifo_item_t item;
 457   smx_simcall_t simcall;
 458
 459   xbt_fifo_foreach(action->simcalls, item, simcall, smx_simcall_t) {
 460
 461     switch (action->state) {
 462
 463       case SIMIX_DONE:
 464         /* do nothing, action done */
 465   XBT_DEBUG("SIMIX_execution_finished: execution successful");
 466         break;
 467
 468       case SIMIX_FAILED:
 469         XBT_DEBUG("SIMIX_execution_finished: host '%s' failed", simcall->issuer->smx_host->name);
 470         simcall->issuer->context->iwannadie = 1;
 471         //SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
 472         break;
 473
 474       case SIMIX_CANCELED:
 475         XBT_DEBUG("SIMIX_execution_finished: execution canceled");
 476         SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Canceled");
 477         break;
 478
 479       default:
 480         xbt_die("Internal error in SIMIX_execution_finish: unexpected action state %d",
 481             (int)action->state);
 482     }
 483     /* check if the host is down */
 484     if (surf_workstation_model->extension.
 485         workstation.get_state(simcall->issuer->smx_host->host) != SURF_RESOURCE_ON) {
 486       simcall->issuer->context->iwannadie = 1;
 487     }
 488
 489     simcall->issuer->waiting_action =    NULL;
 490     simcall->host_execution_wait.result = action->state;
 491     SIMIX_simcall_answer(simcall);
 492   }
 493
 494   /* We no longer need it */
 495   SIMIX_host_execution_destroy(SIMIX_pack_args(PTR(action)));
 496 }
 497
 498 void SIMIX_post_host_execute(smx_action_t action)
 499 {
 500   if (action->type == SIMIX_ACTION_EXECUTE && /* FIMXE: handle resource failure
 501                                                * for parallel tasks too */
 502       surf_workstation_model->extension.workstation.get_state(action->execution.host->host) == SURF_RESOURCE_OFF) {
 503     /* If the host running the action failed, notice it so that the asking
 504      * process can be killed if it runs on that host itself */
 505     action->state = SIMIX_FAILED;
 506   } else if (surf_workstation_model->action_state_get(action->execution.surf_exec) == SURF_ACTION_FAILED) {
 507     /* If the host running the action didn't fail, then the action was
 508      * canceled */
 509     action->state = SIMIX_CANCELED;
 510   } else {
 511     action->state = SIMIX_DONE;
 512   }
 513
 514   if (action->execution.surf_exec) {
 515     surf_workstation_model->action_unref(action->execution.surf_exec);
 516     action->execution.surf_exec = NULL;
 517   }
 518
 519   /* If there are simcalls associated with the action, then answer them */
 520   if (xbt_fifo_size(action->simcalls)) {
 521     SIMIX_execution_finish(action);
 522   }
 523 }
 524
 525
 526 #ifdef HAVE_TRACING
 527 void SIMIX_set_category(smx_action_t action, const char *category)
 528 {
 529   if (action->state != SIMIX_RUNNING) return;
 530   if (action->type == SIMIX_ACTION_EXECUTE){
 531     surf_workstation_model->set_category(action->execution.surf_exec, category);
 532   }else if (action->type == SIMIX_ACTION_COMMUNICATE){
 533     surf_workstation_model->set_category(action->comm.surf_comm, category);
 534   }
 535 }
 536 #endif
 537