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_priv_t smx_host = xbt_new0(s_smx_host_priv_t, 1);
  28   s_smx_process_t proc;
  29
  30   /* Host structure */
  31   smx_host->data = data;
  32   smx_host->process_list =
  33       xbt_swag_new(xbt_swag_offset(proc, host_proc_hookup));
  34
  35   /* Update global variables */
  36   xbt_lib_set(host_lib,name,SIMIX_HOST_LEVEL,smx_host);
  37
  38   return xbt_lib_get_elm_or_null(host_lib, name);
  39 }
  40
  41 /**
  42  * \brief Internal function to destroy a SIMIX host.
  43  *
  44  * \param h the host to destroy (a smx_host_t)
  45  */
  46 void SIMIX_host_destroy(void *h)
  47 {
  48   smx_host_priv_t host = (smx_host_priv_t) h;
  49
  50   xbt_assert((host != NULL), "Invalid parameters");
  51
  52   /* Clean Simulator data */
  53   if (xbt_swag_size(host->process_list) != 0) {
  54     char *msg = xbt_strdup("Shutting down host, but it's not empty:");
  55     char *tmp;
  56     smx_process_t process = NULL;
  57
  58     xbt_swag_foreach(process, host->process_list) {
  59       tmp = bprintf("%s\n\t%s", msg, process->name);
  60       free(msg);
  61       msg = tmp;
  62     }
  63     SIMIX_display_process_status();
  64     THROWF(arg_error, 0, "%s", msg);
  65   }
  66   xbt_dynar_free(&host->auto_restart_processes);
  67   xbt_swag_free(host->process_list);
  68
  69   /* Clean host structure */
  70   free(host);
  71   return;
  72 }
  73
  74 ///**
  75 // * \brief Returns a dict of all hosts.
  76 // *
  77 // * \return List of all hosts (as a #xbt_dict_t)
  78 // */
  79 //xbt_dict_t SIMIX_host_get_dict(void)
  80 //{
  81 //  xbt_dict_t host_dict = xbt_dict_new_homogeneous(NULL);
  82 //  xbt_lib_cursor_t cursor = NULL;
  83 //  char *name = NULL;
  84 //  void **host = NULL;
  85 //
  86 //  xbt_lib_foreach(host_lib, cursor, name, host){
  87 //    if(host[SIMIX_HOST_LEVEL])
  88 //            xbt_dict_set(host_dict,name,host[SIMIX_HOST_LEVEL], NULL);
  89 //  }
  90 //  return host_dict;
  91 //}
  92 smx_host_t SIMIX_pre_host_get_by_name(smx_simcall_t simcall, const char *name){
  93    return SIMIX_host_get_by_name(name);
  94 }
  95 smx_host_t SIMIX_host_get_by_name(const char *name){
  96   xbt_assert(((simix_global != NULL)
  97                && (host_lib != NULL)),
  98               "Environment not set yet");
  99
 100   return xbt_lib_get_elm_or_null(host_lib, name);
 101 }
 102
 103 smx_host_t SIMIX_host_self(void)
 104 {
 105   smx_process_t process = SIMIX_process_self();
 106   return (process == NULL) ? NULL : SIMIX_process_get_host(process);
 107 }
 108
 109 const char* SIMIX_pre_host_self_get_name(smx_simcall_t simcall){
 110    return SIMIX_host_self_get_name();
 111 }
 112 /* needs to be public and without simcall because it is called
 113    by exceptions and logging events */
 114 const char* SIMIX_host_self_get_name(void)
 115 {
 116   smx_host_t host = SIMIX_host_self();
 117   if (host == NULL || SIMIX_process_self() == simix_global->maestro_process)
 118     return "";
 119
 120   return SIMIX_host_get_name(host);
 121 }
 122
 123 const char* SIMIX_pre_host_get_name(smx_simcall_t simcall, smx_host_t host){
 124    return SIMIX_host_get_name(host);
 125 }
 126 const char* SIMIX_host_get_name(smx_host_t host){
 127   xbt_assert((host != NULL), "Invalid parameters");
 128
 129   return sg_host_name(host);
 130 }
 131
 132 xbt_dict_t SIMIX_pre_host_get_properties(smx_simcall_t simcall, smx_host_t host){
 133   return SIMIX_host_get_properties(host);
 134 }
 135 xbt_dict_t SIMIX_host_get_properties(smx_host_t host){
 136   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 137
 138   return surf_resource_get_properties(surf_workstation_resource_priv(host));
 139 }
 140
 141 double SIMIX_pre_host_get_speed(smx_simcall_t simcall, smx_host_t host){
 142   return SIMIX_host_get_speed(host);
 143 }
 144 double SIMIX_host_get_speed(smx_host_t host){
 145   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 146   return surf_workstation_get_speed(host, 1.0);
 147 }
 148
 149 int SIMIX_pre_host_get_core(smx_simcall_t simcall, smx_host_t host){
 150   return SIMIX_host_get_core(host);
 151 }
 152 int SIMIX_host_get_core(smx_host_t host){
 153   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 154
 155   return surf_workstation_get_core(host);
 156 }
 157
 158
 159
 160 double SIMIX_pre_host_get_available_speed(smx_simcall_t simcall, smx_host_t host){
 161   return SIMIX_host_get_available_speed(host);
 162 }
 163 double SIMIX_host_get_available_speed(smx_host_t host){
 164   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 165
 166   return surf_workstation_get_available_speed(host);
 167 }
 168
 169 int SIMIX_pre_host_get_state(smx_simcall_t simcall, smx_host_t host){
 170   return SIMIX_host_get_state(host);
 171 }
 172 int SIMIX_host_get_state(smx_host_t host){
 173   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 174
 175   return surf_resource_get_state(surf_workstation_resource_priv(host));
 176 }
 177
 178 void* SIMIX_pre_host_self_get_data(smx_simcall_t simcall){
 179   return SIMIX_host_self_get_data();
 180 }
 181 void* SIMIX_host_self_get_data(void)
 182 {
 183   smx_host_t self = SIMIX_host_self();
 184   return SIMIX_host_get_data(self);
 185 }
 186
 187 void SIMIX_host_self_set_data(void *data)
 188 {
 189   smx_host_t self = SIMIX_host_self();
 190   SIMIX_host_set_data(self, data);
 191 }
 192
 193 void* SIMIX_pre_host_get_data(smx_simcall_t simcall,smx_host_t host){
 194   return SIMIX_host_get_data(host);
 195 }
 196 void* SIMIX_host_get_data(smx_host_t host){
 197   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 198
 199   return SIMIX_host_priv(host)->data;
 200 }
 201 void _SIMIX_host_free_process_arg(void *);
 202 void _SIMIX_host_free_process_arg(void *data)
 203 {
 204   smx_process_arg_t arg = *(void**)data;
 205   xbt_free(arg->name);
 206   xbt_free(arg);
 207 }
 208 /**
 209  * \brief Add a process to the list of the processes that the host will restart when it comes back
 210  * This function add a process to the list of the processes that will be restarted when the host comes
 211  * back. It is expected that this function is called when the host is down.
 212  * The processes will only be restarted once, meaning that you will have to register the process
 213  * again to restart the process again.
 214  */
 215 void SIMIX_host_add_auto_restart_process(smx_host_t host,
 216                                          const char *name,
 217                                          xbt_main_func_t code,
 218                                          void *data,
 219                                          const char *hostname,
 220                                          double kill_time,
 221                                          int argc, char **argv,
 222                                          xbt_dict_t properties,
 223                                          int auto_restart)
 224 {
 225   if (!SIMIX_host_priv(host)->auto_restart_processes) {
 226     SIMIX_host_priv(host)->auto_restart_processes = xbt_dynar_new(sizeof(smx_process_arg_t),_SIMIX_host_free_process_arg);
 227   }
 228   smx_process_arg_t arg = xbt_new(s_smx_process_arg_t,1);
 229   arg->name = xbt_strdup(name);
 230   arg->code = code;
 231   arg->data = data;
 232   arg->hostname = hostname;
 233   arg->kill_time = kill_time;
 234   arg->argc = argc;
 235
 236   arg->argv = xbt_new(char*,argc + 1);
 237
 238   int i;
 239   for (i = 0; i < argc; i++) {
 240     arg->argv[i] = xbt_strdup(argv[i]);
 241   }
 242   arg->argv[argc] = NULL;
 243
 244   arg->properties = properties;
 245   arg->auto_restart = auto_restart;
 246
 247   if( SIMIX_host_get_state(host) == SURF_RESOURCE_OFF
 248       && !xbt_dict_get_or_null(watched_hosts_lib,sg_host_name(host))){
 249     xbt_dict_set(watched_hosts_lib,sg_host_name(host),host,NULL);
 250     XBT_DEBUG("Have push host %s to watched_hosts_lib because state == SURF_RESOURCE_OFF",sg_host_name(host));
 251   }
 252   xbt_dynar_push_as(SIMIX_host_priv(host)->auto_restart_processes,smx_process_arg_t,arg);
 253 }
 254 /**
 255  * \brief Restart the list of processes that have been registered to the host
 256  */
 257 void SIMIX_host_restart_processes(smx_host_t host)
 258 {
 259   unsigned int cpt;
 260   smx_process_arg_t arg;
 261   xbt_dynar_foreach(SIMIX_host_priv(host)->auto_restart_processes,cpt,arg) {
 262
 263     smx_process_t process;
 264
 265     XBT_DEBUG("Restarting Process %s(%s) right now", arg->argv[0], arg->hostname);
 266     if (simix_global->create_process_function) {
 267       simix_global->create_process_function(&process,
 268                                             arg->argv[0],
 269                                             arg->code,
 270                                             NULL,
 271                                             arg->hostname,
 272                                             arg->kill_time,
 273                                             arg->argc,
 274                                             arg->argv,
 275                                             arg->properties,
 276                                             arg->auto_restart);
 277     }
 278     else {
 279       simcall_process_create(&process,
 280                                             arg->argv[0],
 281                                             arg->code,
 282                                             NULL,
 283                                             arg->hostname,
 284                                             arg->kill_time,
 285                                             arg->argc,
 286                                             arg->argv,
 287                                             arg->properties,
 288                                             arg->auto_restart);
 289
 290     }
 291   }
 292   xbt_dynar_reset(SIMIX_host_priv(host)->auto_restart_processes);
 293 }
 294
 295 void SIMIX_host_autorestart(smx_host_t host)
 296 {
 297   if(simix_global->autorestart)
 298     simix_global->autorestart(host);
 299   else
 300     xbt_die("No function for simix_global->autorestart");
 301 }
 302
 303 void SIMIX_pre_host_set_data(smx_simcall_t simcall, smx_host_t host, void *data) {
 304   SIMIX_host_set_data(host, data);
 305 }
 306 void SIMIX_host_set_data(smx_host_t host, void *data){
 307   xbt_assert((host != NULL), "Invalid parameters");
 308   xbt_assert((SIMIX_host_priv(host)->data == NULL), "Data already set");
 309
 310   SIMIX_host_priv(host)->data = data;
 311 }
 312
 313 smx_action_t SIMIX_pre_host_execute(smx_simcall_t simcall,const char *name,
 314     smx_host_t host, double computation_amount, double priority){
 315   return SIMIX_host_execute(name, host, computation_amount, priority);
 316 }
 317 smx_action_t SIMIX_host_execute(const char *name,
 318     smx_host_t host, double computation_amount, double priority){
 319
 320   /* alloc structures and initialize */
 321   smx_action_t action = xbt_mallocator_get(simix_global->action_mallocator);
 322   action->type = SIMIX_ACTION_EXECUTE;
 323   action->name = xbt_strdup(name);
 324   action->state = SIMIX_RUNNING;
 325   action->execution.host = host;
 326
 327 #ifdef HAVE_TRACING
 328   action->category = NULL;
 329 #endif
 330
 331   /* set surf's action */
 332   if (!MC_is_active()) {
 333     action->execution.surf_exec =
 334       surf_workstation_execute(host, computation_amount);
 335     surf_action_set_data(action->execution.surf_exec, action);
 336     surf_action_set_priority(action->execution.surf_exec, priority);
 337   }
 338
 339   XBT_DEBUG("Create execute action %p", action);
 340
 341   return action;
 342 }
 343
 344 smx_action_t SIMIX_pre_host_parallel_execute(smx_simcall_t simcall, const char *name,
 345     int host_nb, smx_host_t *host_list,
 346     double *computation_amount, double *communication_amount,
 347     double amount, double rate){
 348   return SIMIX_host_parallel_execute(name, host_nb, host_list, computation_amount,
 349                                      communication_amount, amount, rate);
 350 }
 351 smx_action_t SIMIX_host_parallel_execute(const char *name,
 352     int host_nb, smx_host_t *host_list,
 353     double *computation_amount, double *communication_amount,
 354     double amount, double rate){
 355
 356   void **workstation_list = NULL;
 357   int i;
 358
 359   /* alloc structures and initialize */
 360   smx_action_t action = xbt_mallocator_get(simix_global->action_mallocator);
 361   action->type = SIMIX_ACTION_PARALLEL_EXECUTE;
 362   action->name = xbt_strdup(name);
 363   action->state = SIMIX_RUNNING;
 364   action->execution.host = NULL; /* FIXME: do we need the list of hosts? */
 365
 366 #ifdef HAVE_TRACING
 367   action->category = NULL;
 368 #endif
 369
 370   /* set surf's action */
 371   workstation_list = xbt_new0(void *, host_nb);
 372   for (i = 0; i < host_nb; i++)
 373     workstation_list[i] = surf_workstation_resource_priv(host_list[i]);
 374
 375   /* set surf's action */
 376   if (!MC_is_active()) {
 377     action->execution.surf_exec =
 378       surf_workstation_model_execute_parallel_task((surf_workstation_model_t)surf_workstation_model,
 379                   host_nb, workstation_list, computation_amount, communication_amount, rate);
 380
 381     surf_action_set_data(action->execution.surf_exec, action);
 382   }
 383   XBT_DEBUG("Create parallel execute action %p", action);
 384
 385   return action;
 386 }
 387
 388 void SIMIX_pre_host_execution_destroy(smx_simcall_t simcall, smx_action_t action){
 389   SIMIX_host_execution_destroy(action);
 390 }
 391 void SIMIX_host_execution_destroy(smx_action_t action){
 392   XBT_DEBUG("Destroy action %p", action);
 393
 394   if (action->execution.surf_exec) {
 395     surf_action_unref(action->execution.surf_exec);
 396     action->execution.surf_exec = NULL;
 397   }
 398   xbt_free(action->name);
 399   xbt_mallocator_release(simix_global->action_mallocator, action);
 400 }
 401
 402 void SIMIX_pre_host_execution_cancel(smx_simcall_t simcall, smx_action_t action){
 403   SIMIX_host_execution_cancel(action);
 404 }
 405 void SIMIX_host_execution_cancel(smx_action_t action){
 406   XBT_DEBUG("Cancel action %p", action);
 407
 408   if (action->execution.surf_exec)
 409     surf_action_cancel(action->execution.surf_exec);
 410 }
 411
 412 double SIMIX_pre_host_execution_get_remains(smx_simcall_t simcall, smx_action_t action){
 413   return SIMIX_host_execution_get_remains(action);
 414 }
 415 double SIMIX_host_execution_get_remains(smx_action_t action){
 416   double result = 0.0;
 417
 418   if (action->state == SIMIX_RUNNING)
 419     result = surf_action_get_remains(action->execution.surf_exec);
 420
 421   return result;
 422 }
 423
 424 e_smx_state_t SIMIX_pre_host_execution_get_state(smx_simcall_t simcall, smx_action_t action){
 425   return SIMIX_host_execution_get_state(action);
 426 }
 427 e_smx_state_t SIMIX_host_execution_get_state(smx_action_t action){
 428   return action->state;
 429 }
 430
 431 void SIMIX_pre_host_execution_set_priority(smx_simcall_t simcall, smx_action_t action,
 432                                         double priority){
 433   return SIMIX_host_execution_set_priority(action, priority);
 434 }
 435 void SIMIX_host_execution_set_priority(smx_action_t action, double priority){
 436   if(action->execution.surf_exec)
 437         surf_action_set_priority(action->execution.surf_exec, priority);
 438 }
 439
 440 void SIMIX_pre_host_execution_wait(smx_simcall_t simcall, smx_action_t action){
 441
 442   XBT_DEBUG("Wait for execution of action %p, state %d", action, (int)action->state);
 443
 444   /* Associate this simcall to the action */
 445   xbt_fifo_push(action->simcalls, simcall);
 446   simcall->issuer->waiting_action = action;
 447
 448   /* set surf's action */
 449   if (MC_is_active()) {
 450     action->state = SIMIX_DONE;
 451     SIMIX_execution_finish(action);
 452     return;
 453   }
 454
 455   /* If the action is already finished then perform the error handling */
 456   if (action->state != SIMIX_RUNNING)
 457     SIMIX_execution_finish(action);
 458 }
 459
 460 void SIMIX_host_execution_suspend(smx_action_t action)
 461 {
 462   if(action->execution.surf_exec)
 463     surf_action_suspend(action->execution.surf_exec);
 464 }
 465
 466 void SIMIX_host_execution_resume(smx_action_t action)
 467 {
 468   if(action->execution.surf_exec)
 469     surf_action_resume(action->execution.surf_exec);
 470 }
 471
 472 void SIMIX_execution_finish(smx_action_t action)
 473 {
 474   xbt_fifo_item_t item;
 475   smx_simcall_t simcall;
 476
 477   xbt_fifo_foreach(action->simcalls, item, simcall, smx_simcall_t) {
 478
 479     switch (action->state) {
 480
 481       case SIMIX_DONE:
 482         /* do nothing, action done */
 483   XBT_DEBUG("SIMIX_execution_finished: execution successful");
 484         break;
 485
 486       case SIMIX_FAILED:
 487         XBT_DEBUG("SIMIX_execution_finished: host '%s' failed", sg_host_name(simcall->issuer->smx_host));
 488         simcall->issuer->context->iwannadie = 1;
 489         //SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
 490         break;
 491
 492       case SIMIX_CANCELED:
 493         XBT_DEBUG("SIMIX_execution_finished: execution canceled");
 494         SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Canceled");
 495         break;
 496
 497       default:
 498         xbt_die("Internal error in SIMIX_execution_finish: unexpected action state %d",
 499             (int)action->state);
 500     }
 501     /* check if the host is down */
 502
 503     if (surf_resource_get_state(surf_workstation_resource_priv(simcall->issuer->smx_host)) != SURF_RESOURCE_ON) {
 504       simcall->issuer->context->iwannadie = 1;
 505     }
 506
 507     simcall->issuer->waiting_action =    NULL;
 508     simcall_host_execution_wait__set__result(simcall, action->state);
 509     SIMIX_simcall_answer(simcall);
 510   }
 511
 512   /* We no longer need it */
 513   SIMIX_host_execution_destroy(action);
 514 }
 515
 516 void SIMIX_post_host_execute(smx_action_t action)
 517 {
 518   if (action->type == SIMIX_ACTION_EXECUTE && /* FIMXE: handle resource failure
 519                                                * for parallel tasks too */
 520       surf_resource_get_state(surf_workstation_resource_priv(action->execution.host)) == SURF_RESOURCE_OFF) {
 521     /* If the host running the action failed, notice it so that the asking
 522      * process can be killed if it runs on that host itself */
 523     action->state = SIMIX_FAILED;
 524   } else if (surf_action_get_state(action->execution.surf_exec) == SURF_ACTION_FAILED) {
 525     /* If the host running the action didn't fail, then the action was
 526      * canceled */
 527     action->state = SIMIX_CANCELED;
 528   } else {
 529     action->state = SIMIX_DONE;
 530   }
 531
 532   if (action->execution.surf_exec) {
 533     surf_action_unref(action->execution.surf_exec);
 534     action->execution.surf_exec = NULL;
 535   }
 536
 537   /* If there are simcalls associated with the action, then answer them */
 538   if (xbt_fifo_size(action->simcalls)) {
 539     SIMIX_execution_finish(action);
 540   }
 541 }
 542
 543
 544 #ifdef HAVE_TRACING
 545 void SIMIX_pre_set_category(smx_simcall_t simcall, smx_action_t action,
 546                             const char *category){
 547   SIMIX_set_category(action, category);
 548 }
 549 void SIMIX_set_category(smx_action_t action, const char *category)
 550 {
 551   if (action->state != SIMIX_RUNNING) return;
 552   if (action->type == SIMIX_ACTION_EXECUTE){
 553     surf_action_set_category(action->execution.surf_exec, category);
 554   }else if (action->type == SIMIX_ACTION_COMMUNICATE){
 555     surf_action_set_category(action->comm.surf_comm, category);
 556   }
 557 }
 558 #endif
 559