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(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   return SIMIX_host_get_data(SIMIX_host_self());
 171 }
 172
 173 void SIMIX_host_self_set_data(void *data)
 174 {
 175   SIMIX_host_set_data(SIMIX_host_self(), data);
 176 }
 177
 178 void* SIMIX_host_get_data(u_smx_scalar_t *args)
 179 {
 180   smx_host_t host = args[0].p;
 181   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 182
 183   return host->data;
 184 }
 185 void _SIMIX_host_free_process_arg(void *);
 186 void _SIMIX_host_free_process_arg(void *data)
 187 {
 188   smx_process_arg_t arg = *(void**)data;
 189   xbt_free(arg->name);
 190   xbt_free(arg);
 191 }
 192 /**
 193  * \brief Add a process to the list of the processes that the host will restart when it comes back
 194  * This function add a process to the list of the processes that will be restarted when the host comes
 195  * back. It is expected that this function is called when the host is down.
 196  * The processes will only be restarted once, meaning that you will have to register the process
 197  * again to restart the process again.
 198  */
 199 void SIMIX_host_add_auto_restart_process(smx_host_t host,
 200                                          const char *name,
 201                                          xbt_main_func_t code,
 202                                          void *data,
 203                                          const char *hostname,
 204                                          double kill_time,
 205                                          int argc, char **argv,
 206                                          xbt_dict_t properties,
 207                                          int auto_restart)
 208 {
 209   if (!host->auto_restart_processes) {
 210     host->auto_restart_processes = xbt_dynar_new(sizeof(smx_process_arg_t),_SIMIX_host_free_process_arg);
 211   }
 212   smx_process_arg_t arg = xbt_new(s_smx_process_arg_t,1);
 213   arg->name = xbt_strdup(name);
 214   arg->code = code;
 215   arg->data = data;
 216   arg->hostname = hostname;
 217   arg->kill_time = kill_time;
 218   arg->argc = argc;
 219
 220   arg->argv = xbt_new(char*,argc + 1);
 221
 222   int i;
 223   for (i = 0; i < argc; i++) {
 224     arg->argv[i] = xbt_strdup(argv[i]);
 225   }
 226   arg->argv[argc] = NULL;
 227
 228   arg->properties = properties;
 229   arg->auto_restart = auto_restart;
 230
 231   if( SIMIX_host_get_state(host) == SURF_RESOURCE_OFF
 232       && !xbt_dict_get_or_null(watched_hosts_lib,host->name)){
 233     xbt_dict_set(watched_hosts_lib,host->name,host,NULL);
 234     XBT_DEBUG("Have push host %s to watched_hosts_lib because state == SURF_RESOURCE_OFF",host->name);
 235   }
 236   xbt_dynar_push_as(host->auto_restart_processes,smx_process_arg_t,arg);
 237 }
 238 /**
 239  * \brief Restart the list of processes that have been registered to the host
 240  */
 241 void SIMIX_host_restart_processes(smx_host_t host)
 242 {
 243   unsigned int cpt;
 244   smx_process_arg_t arg;
 245   xbt_dynar_foreach(host->auto_restart_processes,cpt,arg) {
 246
 247     smx_process_t process;
 248
 249     XBT_DEBUG("Restarting Process %s(%s) right now", arg->argv[0], arg->hostname);
 250     if (simix_global->create_process_function) {
 251       simix_global->create_process_function(&process,
 252                                             arg->argv[0],
 253                                             arg->code,
 254                                             NULL,
 255                                             arg->hostname,
 256                                             arg->kill_time,
 257                                             arg->argc,
 258                                             arg->argv,
 259                                             arg->properties,
 260                                             arg->auto_restart);
 261     }
 262     else {
 263       simcall_process_create(&process,
 264                                             arg->argv[0],
 265                                             arg->code,
 266                                             NULL,
 267                                             arg->hostname,
 268                                             arg->kill_time,
 269                                             arg->argc,
 270                                             arg->argv,
 271                                             arg->properties,
 272                                             arg->auto_restart);
 273
 274     }
 275   }
 276   xbt_dynar_reset(host->auto_restart_processes);
 277 }
 278
 279 void SIMIX_host_autorestart(smx_host_t host)
 280 {
 281   if(simix_global->autorestart)
 282     simix_global->autorestart(host);
 283   else
 284     xbt_die("No function for simix_global->autorestart");
 285 }
 286
 287 void SIMIX_host_set_data(u_smx_scalar_t *args)
 288 {
 289   smx_host_t host = args[0].p;
 290   void *data = args[1].p;
 291   xbt_assert((host != NULL), "Invalid parameters");
 292   xbt_assert((host->data == NULL), "Data already set");
 293
 294   host->data = data;
 295 }
 296
 297 smx_action_t SIMIX_host_execute(u_smx_scalar_t args[])
 298 {
 299   const char *name = args[0].p;
 300   smx_host_t host = args[1].p;
 301   double computation_amount = args[2].d;
 302   double priority = args[3].d;
 303
 304   /* alloc structures and initialize */
 305   smx_action_t action = xbt_mallocator_get(simix_global->action_mallocator);
 306   action->type = SIMIX_ACTION_EXECUTE;
 307   action->name = xbt_strdup(name);
 308   action->state = SIMIX_RUNNING;
 309   action->execution.host = host;
 310
 311 #ifdef HAVE_TRACING
 312   action->category = NULL;
 313 #endif
 314
 315   /* set surf's action */
 316   if (!MC_is_active()) {
 317     action->execution.surf_exec =
 318       surf_workstation_model->extension.workstation.execute(host->host,
 319     computation_amount);
 320     surf_workstation_model->action_data_set(action->execution.surf_exec, action);
 321     surf_workstation_model->set_priority(action->execution.surf_exec, priority);
 322   }
 323
 324   XBT_DEBUG("Create execute action %p", action);
 325
 326   return action;
 327 }
 328
 329 smx_action_t SIMIX_host_parallel_execute(u_smx_scalar_t *args)
 330 {
 331   const char *name = args[0].cc;
 332   int host_nb = args[1].i;
 333   smx_host_t *host_list = args[2].p;
 334   double *computation_amount = args[3].p;
 335   double *communication_amount = args[4].p;
 336   double amount = args[5].d;
 337   double rate = args[6].d;
 338   void **workstation_list = NULL;
 339   int i;
 340
 341   /* alloc structures and initialize */
 342   smx_action_t action = xbt_mallocator_get(simix_global->action_mallocator);
 343   action->type = SIMIX_ACTION_PARALLEL_EXECUTE;
 344   action->name = xbt_strdup(name);
 345   action->state = SIMIX_RUNNING;
 346   action->execution.host = NULL; /* FIXME: do we need the list of hosts? */
 347
 348 #ifdef HAVE_TRACING
 349   action->category = NULL;
 350 #endif
 351
 352   /* set surf's action */
 353   workstation_list = xbt_new0(void *, host_nb);
 354   for (i = 0; i < host_nb; i++)
 355     workstation_list[i] = host_list[i]->host;
 356
 357   /* set surf's action */
 358   if (!MC_is_active()) {
 359     action->execution.surf_exec =
 360       surf_workstation_model->extension.workstation.
 361       execute_parallel_task(host_nb, workstation_list, computation_amount,
 362                       communication_amount, rate);
 363
 364     surf_workstation_model->action_data_set(action->execution.surf_exec, action);
 365   }
 366   XBT_DEBUG("Create parallel execute action %p", action);
 367
 368   return action;
 369 }
 370
 371 void SIMIX_host_execution_destroy(u_smx_scalar_t *args)
 372 {
 373   smx_action_t action = args[0].p;
 374   XBT_DEBUG("Destroy action %p", action);
 375
 376   if (action->execution.surf_exec) {
 377     surf_workstation_model->action_unref(action->execution.surf_exec);
 378     action->execution.surf_exec = NULL;
 379   }
 380   xbt_free(action->name);
 381   xbt_mallocator_release(simix_global->action_mallocator, action);
 382 }
 383
 384 void SIMIX_host_execution_cancel(u_smx_scalar_t *args)
 385 {
 386   smx_action_t action = args[0].p;
 387   XBT_DEBUG("Cancel action %p", action);
 388
 389   if (action->execution.surf_exec)
 390     surf_workstation_model->action_cancel(action->execution.surf_exec);
 391 }
 392
 393 double SIMIX_host_execution_get_remains(u_smx_scalar_t *args)
 394 {
 395   smx_action_t action = args[0].p;
 396   double result = 0.0;
 397
 398   if (action->state == SIMIX_RUNNING)
 399     result = surf_workstation_model->get_remains(action->execution.surf_exec);
 400
 401   return result;
 402 }
 403
 404 e_smx_state_t SIMIX_host_execution_get_state(u_smx_scalar_t *args)
 405 {
 406   smx_action_t action = args[0].p;
 407   return action->state;
 408 }
 409
 410 void SIMIX_host_execution_set_priority(u_smx_scalar_t *args)
 411 {
 412   smx_action_t action = args[0].p;
 413   double priority = args[1].d;
 414   if(action->execution.surf_exec)
 415     surf_workstation_model->set_priority(action->execution.surf_exec, priority);
 416 }
 417
 418 void SIMIX_pre_host_execution_wait(u_smx_scalar_t *args)
 419 {
 420   smx_simcall_t simcall = args[0].p;
 421   smx_action_t action = simcall->host_execution_wait.execution;
 422
 423   XBT_DEBUG("Wait for execution of action %p, state %d", action, (int)action->state);
 424
 425   /* Associate this simcall to the action */
 426   xbt_fifo_push(action->simcalls, simcall);
 427   simcall->issuer->waiting_action = action;
 428
 429   /* set surf's action */
 430   if (MC_is_active()) {
 431     action->state = SIMIX_DONE;
 432     SIMIX_execution_finish(action);
 433     return;
 434   }
 435
 436   /* If the action is already finished then perform the error handling */
 437   if (action->state != SIMIX_RUNNING)
 438     SIMIX_execution_finish(action);
 439 }
 440
 441 void SIMIX_host_execution_suspend(smx_action_t action)
 442 {
 443   if(action->execution.surf_exec)
 444     surf_workstation_model->suspend(action->execution.surf_exec);
 445 }
 446
 447 void SIMIX_host_execution_resume(smx_action_t action)
 448 {
 449   if(action->execution.surf_exec)
 450     surf_workstation_model->resume(action->execution.surf_exec);
 451 }
 452
 453 void SIMIX_execution_finish(smx_action_t action)
 454 {
 455   xbt_fifo_item_t item;
 456   smx_simcall_t simcall;
 457
 458   xbt_fifo_foreach(action->simcalls, item, simcall, smx_simcall_t) {
 459
 460     switch (action->state) {
 461
 462       case SIMIX_DONE:
 463         /* do nothing, action done */
 464   XBT_DEBUG("SIMIX_execution_finished: execution successful");
 465         break;
 466
 467       case SIMIX_FAILED:
 468         XBT_DEBUG("SIMIX_execution_finished: host '%s' failed", simcall->issuer->smx_host->name);
 469         simcall->issuer->context->iwannadie = 1;
 470         //SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
 471         break;
 472
 473       case SIMIX_CANCELED:
 474         XBT_DEBUG("SIMIX_execution_finished: execution canceled");
 475         SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Canceled");
 476         break;
 477
 478       default:
 479         xbt_die("Internal error in SIMIX_execution_finish: unexpected action state %d",
 480             (int)action->state);
 481     }
 482     /* check if the host is down */
 483     if (surf_workstation_model->extension.
 484         workstation.get_state(simcall->issuer->smx_host->host) != SURF_RESOURCE_ON) {
 485       simcall->issuer->context->iwannadie = 1;
 486     }
 487
 488     simcall->issuer->waiting_action =    NULL;
 489     simcall->host_execution_wait.result = action->state;
 490     SIMIX_simcall_answer(simcall);
 491   }
 492
 493   /* We no longer need it */
 494   SIMIX_host_execution_destroy(action);
 495 }
 496
 497 void SIMIX_post_host_execute(smx_action_t action)
 498 {
 499   if (action->type == SIMIX_ACTION_EXECUTE && /* FIMXE: handle resource failure
 500                                                * for parallel tasks too */
 501       surf_workstation_model->extension.workstation.get_state(action->execution.host->host) == SURF_RESOURCE_OFF) {
 502     /* If the host running the action failed, notice it so that the asking
 503      * process can be killed if it runs on that host itself */
 504     action->state = SIMIX_FAILED;
 505   } else if (surf_workstation_model->action_state_get(action->execution.surf_exec) == SURF_ACTION_FAILED) {
 506     /* If the host running the action didn't fail, then the action was
 507      * canceled */
 508     action->state = SIMIX_CANCELED;
 509   } else {
 510     action->state = SIMIX_DONE;
 511   }
 512
 513   if (action->execution.surf_exec) {
 514     surf_workstation_model->action_unref(action->execution.surf_exec);
 515     action->execution.surf_exec = NULL;
 516   }
 517
 518   /* If there are simcalls associated with the action, then answer them */
 519   if (xbt_fifo_size(action->simcalls)) {
 520     SIMIX_execution_finish(action);
 521   }
 522 }
 523
 524
 525 #ifdef HAVE_TRACING
 526 void SIMIX_set_category(smx_action_t action, const char *category)
 527 {
 528   if (action->state != SIMIX_RUNNING) return;
 529   if (action->type == SIMIX_ACTION_EXECUTE){
 530     surf_workstation_model->set_category(action->execution.surf_exec, category);
 531   }else if (action->type == SIMIX_ACTION_COMMUNICATE){
 532     surf_workstation_model->set_category(action->comm.surf_comm, category);
 533   }
 534 }
 535 #endif
 536