src/simix/smx_vm.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_vm, 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 data some user data (may be NULL)
  22  */
  23 smx_host_t SIMIX_vm_create(const char *name, smx_host_t phys_host)
  24 {
  25
  26   smx_host_priv_t smx_host = xbt_new0(s_smx_host_priv_t, 1);
  27   s_smx_process_t proc;
  28
  29   // TODO check why we do not have any VM here and why we have the host_proc_hookup  ?
  30
  31   /* Host structure */
  32   smx_host->data = NULL;
  33   smx_host->process_list =
  34       xbt_swag_new(xbt_swag_offset(proc, host_proc_hookup));
  35
  36   /* Update global variables */
  37   xbt_lib_set(host_lib,name,SIMIX_HOST_LEVEL,smx_host);
  38
  39   /* Create surf associated resource */
  40   surf_vm_workstation_model->extension.vm_workstation.create();
  41   return xbt_lib_get_elm_or_null(host_lib, name);
  42 }
  43
  44 /**
  45  * \brief Internal function to destroy a SIMIX host.
  46  *
  47  * \param h the host to destroy (a smx_host_t)
  48  */
  49 void SIMIX_vm_host_destroy(void *h)
  50 {
  51   smx_host_priv_t host = (smx_host_priv_t) h;
  52
  53   smx_host_t
  54   xbt_lib_(host_lib,name,SIMIX_HOST_LEVEL,smx_host);
  55
  56   /* jump to vm_ws_destroy() */
  57   surf_vm_workstation_model->extension.vm_workstation.destroy(name);
  58
  59   xbt_assert((host != NULL), "Invalid parameters");
  60
  61   /* Clean Simulator data */
  62   if (xbt_swag_size(host->process_list) != 0) {
  63     char *msg =
  64         bprintf("Shutting down host, but it's not empty:");
  65     char *tmp;
  66     smx_process_t process = NULL;
  67
  68     xbt_swag_foreach(process, host->process_list) {
  69       tmp = bprintf("%s\n\t%s", msg, process->name);
  70       free(msg);
  71       msg = tmp;
  72     }
  73     SIMIX_display_process_status();
  74     THROWF(arg_error, 0, "%s", msg);
  75   }
  76   xbt_dynar_free(&host->auto_restart_processes);
  77   xbt_swag_free(host->process_list);
  78
  79   /* Clean host structure */
  80   free(host);
  81   return;
  82 }
  83
  84 ///**
  85 // * \brief Returns a dict of all hosts.
  86 // *
  87 // * \return List of all hosts (as a #xbt_dict_t)
  88 // */
  89 //xbt_dict_t SIMIX_host_get_dict(void)
  90 //{
  91 //  xbt_dict_t host_dict = xbt_dict_new_homogeneous(NULL);
  92 //  xbt_lib_cursor_t cursor = NULL;
  93 //  char *name = NULL;
  94 //  void **host = NULL;
  95 //
  96 //  xbt_lib_foreach(host_lib, cursor, name, host){
  97 //    if(host[SIMIX_HOST_LEVEL])
  98 //            xbt_dict_set(host_dict,name,host[SIMIX_HOST_LEVEL], NULL);
  99 //  }
 100 //  return host_dict;
 101 //}
 102 smx_host_t SIMIX_pre_vm_create(smx_simcall_t simcall, const char *name, smx_host_t phys_host){
 103    return SIMIX_vm_create(name, phys_host);
 104 }
 105
 106 smx_host_t SIMIX_host_get_by_name(const char *name){
 107   xbt_assert(((simix_global != NULL)
 108                && (host_lib != NULL)),
 109               "Environment not set yet");
 110
 111   return xbt_lib_get_elm_or_null(host_lib, name);
 112 }
 113
 114 smx_host_t SIMIX_host_self(void)
 115 {
 116   smx_process_t process = SIMIX_process_self();
 117   return (process == NULL) ? NULL : SIMIX_process_get_host(process);
 118 }
 119
 120 const char* SIMIX_pre_host_self_get_name(smx_simcall_t simcall){
 121    return SIMIX_host_self_get_name();
 122 }
 123 /* needs to be public and without simcall because it is called
 124    by exceptions and logging events */
 125 const char* SIMIX_host_self_get_name(void)
 126 {
 127   smx_host_t host = SIMIX_host_self();
 128   if (host == NULL || SIMIX_process_self() == simix_global->maestro_process)
 129     return "";
 130
 131   return SIMIX_host_get_name(host);
 132 }
 133
 134 const char* SIMIX_pre_host_get_name(smx_simcall_t simcall, smx_host_t host){
 135    return SIMIX_host_get_name(host);
 136 }
 137 const char* SIMIX_host_get_name(smx_host_t host){
 138   xbt_assert((host != NULL), "Invalid parameters");
 139
 140   return sg_host_name(host);
 141 }
 142
 143 xbt_dict_t SIMIX_pre_host_get_properties(smx_simcall_t simcall, smx_host_t host){
 144   return SIMIX_host_get_properties(host);
 145 }
 146 xbt_dict_t SIMIX_host_get_properties(smx_host_t host){
 147   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 148
 149   return surf_workstation_model->extension.workstation.get_properties(host);
 150 }
 151
 152 double SIMIX_pre_host_get_speed(smx_simcall_t simcall, smx_host_t host){
 153   return SIMIX_host_get_speed(host);
 154 }
 155 double SIMIX_host_get_speed(smx_host_t host){
 156   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 157
 158   return surf_workstation_model->extension.workstation.
 159       get_speed(host, 1.0);
 160 }
 161
 162 double SIMIX_pre_host_get_available_speed(smx_simcall_t simcall, smx_host_t host){
 163   return SIMIX_host_get_available_speed(host);
 164 }
 165 double SIMIX_host_get_available_speed(smx_host_t host){
 166   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 167
 168   return surf_workstation_model->extension.workstation.
 169       get_available_speed(host);
 170 }
 171
 172 int SIMIX_pre_host_get_state(smx_simcall_t simcall, smx_host_t host){
 173   return SIMIX_host_get_state(host);
 174 }
 175 int SIMIX_host_get_state(smx_host_t host){
 176   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 177
 178   return surf_workstation_model->extension.workstation.
 179       get_state(host);
 180 }
 181
 182 void* SIMIX_pre_host_self_get_data(smx_simcall_t simcall){
 183   return SIMIX_host_self_get_data();
 184 }
 185 void* SIMIX_host_self_get_data(void)
 186 {
 187   smx_host_t self = SIMIX_host_self();
 188   return SIMIX_host_get_data(self);
 189 }
 190
 191 void SIMIX_host_self_set_data(void *data)
 192 {
 193   smx_host_t self = SIMIX_host_self();
 194   SIMIX_host_set_data(self, data);
 195 }
 196
 197 void* SIMIX_pre_host_get_data(smx_simcall_t simcall,smx_host_t host){
 198   return SIMIX_host_get_data(host);
 199 }
 200 void* SIMIX_host_get_data(smx_host_t host){
 201   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 202
 203   return SIMIX_host_priv(host)->data;
 204 }
 205 void _SIMIX_host_free_process_arg(void *);
 206 void _SIMIX_host_free_process_arg(void *data)
 207 {
 208   smx_process_arg_t arg = *(void**)data;
 209   xbt_free(arg->name);
 210   xbt_free(arg);
 211 }
 212 /**
 213  * \brief Add a process to the list of the processes that the host will restart when it comes back
 214  * This function add a process to the list of the processes that will be restarted when the host comes
 215  * back. It is expected that this function is called when the host is down.
 216  * The processes will only be restarted once, meaning that you will have to register the process
 217  * again to restart the process again.
 218  */
 219 void SIMIX_host_add_auto_restart_process(smx_host_t host,
 220                                          const char *name,
 221                                          xbt_main_func_t code,
 222                                          void *data,
 223                                          const char *hostname,
 224                                          double kill_time,
 225                                          int argc, char **argv,
 226                                          xbt_dict_t properties,
 227                                          int auto_restart)
 228 {
 229   if (!SIMIX_host_priv(host)->auto_restart_processes) {
 230     SIMIX_host_priv(host)->auto_restart_processes = xbt_dynar_new(sizeof(smx_process_arg_t),_SIMIX_host_free_process_arg);
 231   }
 232   smx_process_arg_t arg = xbt_new(s_smx_process_arg_t,1);
 233   arg->name = xbt_strdup(name);
 234   arg->code = code;
 235   arg->data = data;
 236   arg->hostname = hostname;
 237   arg->kill_time = kill_time;
 238   arg->argc = argc;
 239
 240   arg->argv = xbt_new(char*,argc + 1);
 241
 242   int i;
 243   for (i = 0; i < argc; i++) {
 244     arg->argv[i] = xbt_strdup(argv[i]);
 245   }
 246   arg->argv[argc] = NULL;
 247
 248   arg->properties = properties;
 249   arg->auto_restart = auto_restart;
 250
 251   if( SIMIX_host_get_state(host) == SURF_RESOURCE_OFF
 252       && !xbt_dict_get_or_null(watched_hosts_lib,sg_host_name(host))){
 253     xbt_dict_set(watched_hosts_lib,sg_host_name(host),host,NULL);
 254     XBT_DEBUG("Have push host %s to watched_hosts_lib because state == SURF_RESOURCE_OFF",sg_host_name(host));
 255   }
 256   xbt_dynar_push_as(SIMIX_host_priv(host)->auto_restart_processes,smx_process_arg_t,arg);
 257 }
 258 /**
 259  * \brief Restart the list of processes that have been registered to the host
 260  */
 261 void SIMIX_host_restart_processes(smx_host_t host)
 262 {
 263   unsigned int cpt;
 264   smx_process_arg_t arg;
 265   xbt_dynar_foreach(SIMIX_host_priv(host)->auto_restart_processes,cpt,arg) {
 266
 267     smx_process_t process;
 268
 269     XBT_DEBUG("Restarting Process %s(%s) right now", arg->argv[0], arg->hostname);
 270     if (simix_global->create_process_function) {
 271       simix_global->create_process_function(&process,
 272                                             arg->argv[0],
 273                                             arg->code,
 274                                             NULL,
 275                                             arg->hostname,
 276                                             arg->kill_time,
 277                                             arg->argc,
 278                                             arg->argv,
 279                                             arg->properties,
 280                                             arg->auto_restart);
 281     }
 282     else {
 283       simcall_process_create(&process,
 284                                             arg->argv[0],
 285                                             arg->code,
 286                                             NULL,
 287                                             arg->hostname,
 288                                             arg->kill_time,
 289                                             arg->argc,
 290                                             arg->argv,
 291                                             arg->properties,
 292                                             arg->auto_restart);
 293
 294     }
 295   }
 296   xbt_dynar_reset(SIMIX_host_priv(host)->auto_restart_processes);
 297 }
 298
 299 void SIMIX_host_autorestart(smx_host_t host)
 300 {
 301   if(simix_global->autorestart)
 302     simix_global->autorestart(host);
 303   else
 304     xbt_die("No function for simix_global->autorestart");
 305 }
 306
 307 void SIMIX_pre_host_set_data(smx_simcall_t simcall, smx_host_t host, void *data) {
 308   SIMIX_host_set_data(host, data);
 309 }
 310 void SIMIX_host_set_data(smx_host_t host, void *data){
 311   xbt_assert((host != NULL), "Invalid parameters");
 312   xbt_assert((SIMIX_host_priv(host)->data == NULL), "Data already set");
 313
 314   SIMIX_host_priv(host)->data = data;
 315 }
 316
 317 smx_action_t SIMIX_pre_host_execute(smx_simcall_t simcall,const char *name,
 318     smx_host_t host, double computation_amount, double priority){
 319   return SIMIX_host_execute(name, host, computation_amount, priority);
 320 }
 321 smx_action_t SIMIX_host_execute(const char *name,
 322     smx_host_t host, double computation_amount, double priority){
 323
 324   /* alloc structures and initialize */
 325   smx_action_t action = xbt_mallocator_get(simix_global->action_mallocator);
 326   action->type = SIMIX_ACTION_EXECUTE;
 327   action->name = xbt_strdup(name);
 328   action->state = SIMIX_RUNNING;
 329   action->execution.host = host;
 330
 331 #ifdef HAVE_TRACING
 332   action->category = NULL;
 333 #endif
 334
 335   /* set surf's action */
 336   if (!MC_is_active()) {
 337     action->execution.surf_exec =
 338       surf_workstation_model->extension.workstation.execute(host,
 339     computation_amount);
 340     surf_workstation_model->action_data_set(action->execution.surf_exec, action);
 341     surf_workstation_model->set_priority(action->execution.surf_exec, priority);
 342   }
 343
 344   XBT_DEBUG("Create execute action %p", action);
 345
 346   return action;
 347 }
 348
 349 smx_action_t SIMIX_pre_host_parallel_execute(smx_simcall_t simcall, const char *name,
 350     int host_nb, smx_host_t *host_list,
 351     double *computation_amount, double *communication_amount,
 352     double amount, double rate){
 353   return SIMIX_host_parallel_execute(name, host_nb, host_list, computation_amount,
 354                                      communication_amount, amount, rate);
 355 }
 356 smx_action_t SIMIX_host_parallel_execute(const char *name,
 357     int host_nb, smx_host_t *host_list,
 358     double *computation_amount, double *communication_amount,
 359     double amount, double rate){
 360
 361   void **workstation_list = NULL;
 362   int i;
 363
 364   /* alloc structures and initialize */
 365   smx_action_t action = xbt_mallocator_get(simix_global->action_mallocator);
 366   action->type = SIMIX_ACTION_PARALLEL_EXECUTE;
 367   action->name = xbt_strdup(name);
 368   action->state = SIMIX_RUNNING;
 369   action->execution.host = NULL; /* FIXME: do we need the list of hosts? */
 370
 371 #ifdef HAVE_TRACING
 372   action->category = NULL;
 373 #endif
 374
 375   /* set surf's action */
 376   workstation_list = xbt_new0(void *, host_nb);
 377   for (i = 0; i < host_nb; i++)
 378     workstation_list[i] = host_list[i];
 379
 380   /* set surf's action */
 381   if (!MC_is_active()) {
 382     action->execution.surf_exec =
 383       surf_workstation_model->extension.workstation.
 384       execute_parallel_task(host_nb, workstation_list, computation_amount,
 385                       communication_amount, rate);
 386
 387     surf_workstation_model->action_data_set(action->execution.surf_exec, action);
 388   }
 389   XBT_DEBUG("Create parallel execute action %p", action);
 390
 391   return action;
 392 }
 393
 394 void SIMIX_pre_host_execution_destroy(smx_simcall_t simcall, smx_action_t action){
 395   SIMIX_host_execution_destroy(action);
 396 }
 397 void SIMIX_host_execution_destroy(smx_action_t action){
 398   XBT_DEBUG("Destroy action %p", action);
 399
 400   if (action->execution.surf_exec) {
 401     surf_workstation_model->action_unref(action->execution.surf_exec);
 402     action->execution.surf_exec = NULL;
 403   }
 404   xbt_free(action->name);
 405   xbt_mallocator_release(simix_global->action_mallocator, action);
 406 }
 407
 408 void SIMIX_pre_host_execution_cancel(smx_simcall_t simcall, smx_action_t action){
 409   SIMIX_host_execution_cancel(action);
 410 }
 411 void SIMIX_host_execution_cancel(smx_action_t action){
 412   XBT_DEBUG("Cancel action %p", action);
 413
 414   if (action->execution.surf_exec)
 415     surf_workstation_model->action_cancel(action->execution.surf_exec);
 416 }
 417
 418 double SIMIX_pre_host_execution_get_remains(smx_simcall_t simcall, smx_action_t action){
 419   return SIMIX_host_execution_get_remains(action);
 420 }
 421 double SIMIX_host_execution_get_remains(smx_action_t action){
 422   double result = 0.0;
 423
 424   if (action->state == SIMIX_RUNNING)
 425     result = surf_workstation_model->get_remains(action->execution.surf_exec);
 426
 427   return result;
 428 }
 429
 430 e_smx_state_t SIMIX_pre_host_execution_get_state(smx_simcall_t simcall, smx_action_t action){
 431   return SIMIX_host_execution_get_state(action);
 432 }
 433 e_smx_state_t SIMIX_host_execution_get_state(smx_action_t action){
 434   return action->state;
 435 }
 436
 437 void SIMIX_pre_host_execution_set_priority(smx_simcall_t simcall, smx_action_t action,
 438                                         double priority){
 439   return SIMIX_host_execution_set_priority(action, priority);
 440 }
 441 void SIMIX_host_execution_set_priority(smx_action_t action, double priority){
 442   if(action->execution.surf_exec)
 443     surf_workstation_model->set_priority(action->execution.surf_exec, priority);
 444 }
 445
 446 void SIMIX_pre_host_execution_wait(smx_simcall_t simcall, smx_action_t action){
 447
 448   XBT_DEBUG("Wait for execution of action %p, state %d", action, (int)action->state);
 449
 450   /* Associate this simcall to the action */
 451   xbt_fifo_push(action->simcalls, simcall);
 452   simcall->issuer->waiting_action = action;
 453
 454   /* set surf's action */
 455   if (MC_is_active()) {
 456     action->state = SIMIX_DONE;
 457     SIMIX_execution_finish(action);
 458     return;
 459   }
 460
 461   /* If the action is already finished then perform the error handling */
 462   if (action->state != SIMIX_RUNNING)
 463     SIMIX_execution_finish(action);
 464 }
 465
 466 void SIMIX_host_execution_suspend(smx_action_t action)
 467 {
 468   if(action->execution.surf_exec)
 469     surf_workstation_model->suspend(action->execution.surf_exec);
 470 }
 471
 472 void SIMIX_host_execution_resume(smx_action_t action)
 473 {
 474   if(action->execution.surf_exec)
 475     surf_workstation_model->resume(action->execution.surf_exec);
 476 }
 477
 478 void SIMIX_execution_finish(smx_action_t action)
 479 {
 480   xbt_fifo_item_t item;
 481   smx_simcall_t simcall;
 482
 483   xbt_fifo_foreach(action->simcalls, item, simcall, smx_simcall_t) {
 484
 485     switch (action->state) {
 486
 487       case SIMIX_DONE:
 488         /* do nothing, action done */
 489   XBT_DEBUG("SIMIX_execution_finished: execution successful");
 490         break;
 491
 492       case SIMIX_FAILED:
 493         XBT_DEBUG("SIMIX_execution_finished: host '%s' failed", sg_host_name(simcall->issuer->smx_host));
 494         simcall->issuer->context->iwannadie = 1;
 495         //SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
 496         break;
 497
 498       case SIMIX_CANCELED:
 499         XBT_DEBUG("SIMIX_execution_finished: execution canceled");
 500         SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Canceled");
 501         break;
 502
 503       default:
 504         xbt_die("Internal error in SIMIX_execution_finish: unexpected action state %d",
 505             (int)action->state);
 506     }
 507     /* check if the host is down */
 508     if (surf_workstation_model->extension.
 509         workstation.get_state(simcall->issuer->smx_host) != SURF_RESOURCE_ON) {
 510       simcall->issuer->context->iwannadie = 1;
 511     }
 512
 513     simcall->issuer->waiting_action =    NULL;
 514     simcall_host_execution_wait__set__result(simcall, action->state);
 515     SIMIX_simcall_answer(simcall);
 516   }
 517
 518   /* We no longer need it */
 519   SIMIX_host_execution_destroy(action);
 520 }
 521
 522 void SIMIX_post_host_execute(smx_action_t action)
 523 {
 524   if (action->type == SIMIX_ACTION_EXECUTE && /* FIMXE: handle resource failure
 525                                                * for parallel tasks too */
 526       surf_workstation_model->extension.workstation.get_state(action->execution.host) == SURF_RESOURCE_OFF) {
 527     /* If the host running the action failed, notice it so that the asking
 528      * process can be killed if it runs on that host itself */
 529     action->state = SIMIX_FAILED;
 530   } else if (surf_workstation_model->action_state_get(action->execution.surf_exec) == SURF_ACTION_FAILED) {
 531     /* If the host running the action didn't fail, then the action was
 532      * canceled */
 533     action->state = SIMIX_CANCELED;
 534   } else {
 535     action->state = SIMIX_DONE;
 536   }
 537
 538   if (action->execution.surf_exec) {
 539     surf_workstation_model->action_unref(action->execution.surf_exec);
 540     action->execution.surf_exec = NULL;
 541   }
 542
 543   /* If there are simcalls associated with the action, then answer them */
 544   if (xbt_fifo_size(action->simcalls)) {
 545     SIMIX_execution_finish(action);
 546   }
 547 }
 548
 549
 550 #ifdef HAVE_TRACING
 551 void SIMIX_pre_set_category(smx_simcall_t simcall, smx_action_t action,
 552                             const char *category){
 553   SIMIX_set_category(action, category);
 554 }
 555 void SIMIX_set_category(smx_action_t action, const char *category)
 556 {
 557   if (action->state != SIMIX_RUNNING) return;
 558   if (action->type == SIMIX_ACTION_EXECUTE){
 559     surf_workstation_model->set_category(action->execution.surf_exec, category);
 560   }else if (action->type == SIMIX_ACTION_COMMUNICATE){
 561     surf_workstation_model->set_category(action->comm.surf_comm, category);
 562   }
 563 }
 564 #endif
 565