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_workstation_model->extension.workstation.get_properties(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
 147   return surf_workstation_model->extension.workstation.
 148       get_speed(host, 1.0);
 149 }
 150
 151 double SIMIX_pre_host_get_available_speed(smx_simcall_t simcall, smx_host_t host){
 152   return SIMIX_host_get_available_speed(host);
 153 }
 154 double SIMIX_host_get_available_speed(smx_host_t host){
 155   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 156
 157   return surf_workstation_model->extension.workstation.
 158       get_available_speed(host);
 159 }
 160
 161 int SIMIX_pre_host_get_state(smx_simcall_t simcall, smx_host_t host){
 162   return SIMIX_host_get_state(host);
 163 }
 164 int SIMIX_host_get_state(smx_host_t host){
 165   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 166
 167   return surf_workstation_model->extension.workstation.
 168       get_state(host);
 169 }
 170
 171 void* SIMIX_pre_host_self_get_data(smx_simcall_t simcall){
 172   return SIMIX_host_self_get_data();
 173 }
 174 void* SIMIX_host_self_get_data(void)
 175 {
 176   smx_host_t self = SIMIX_host_self();
 177   return SIMIX_host_get_data(self);
 178 }
 179
 180 void SIMIX_host_self_set_data(void *data)
 181 {
 182   smx_host_t self = SIMIX_host_self();
 183   SIMIX_host_set_data(self, data);
 184 }
 185
 186 void* SIMIX_pre_host_get_data(smx_simcall_t simcall,smx_host_t host){
 187   return SIMIX_host_get_data(host);
 188 }
 189 void* SIMIX_host_get_data(smx_host_t host){
 190   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 191
 192   return SIMIX_host_priv(host)->data;
 193 }
 194 void _SIMIX_host_free_process_arg(void *);
 195 void _SIMIX_host_free_process_arg(void *data)
 196 {
 197   smx_process_arg_t arg = *(void**)data;
 198   xbt_free(arg->name);
 199   xbt_free(arg);
 200 }
 201 /**
 202  * \brief Add a process to the list of the processes that the host will restart when it comes back
 203  * This function add a process to the list of the processes that will be restarted when the host comes
 204  * back. It is expected that this function is called when the host is down.
 205  * The processes will only be restarted once, meaning that you will have to register the process
 206  * again to restart the process again.
 207  */
 208 void SIMIX_host_add_auto_restart_process(smx_host_t host,
 209                                          const char *name,
 210                                          xbt_main_func_t code,
 211                                          void *data,
 212                                          const char *hostname,
 213                                          double kill_time,
 214                                          int argc, char **argv,
 215                                          xbt_dict_t properties,
 216                                          int auto_restart)
 217 {
 218   if (!SIMIX_host_priv(host)->auto_restart_processes) {
 219     SIMIX_host_priv(host)->auto_restart_processes = xbt_dynar_new(sizeof(smx_process_arg_t),_SIMIX_host_free_process_arg);
 220   }
 221   smx_process_arg_t arg = xbt_new(s_smx_process_arg_t,1);
 222   arg->name = xbt_strdup(name);
 223   arg->code = code;
 224   arg->data = data;
 225   arg->hostname = hostname;
 226   arg->kill_time = kill_time;
 227   arg->argc = argc;
 228
 229   arg->argv = xbt_new(char*,argc + 1);
 230
 231   int i;
 232   for (i = 0; i < argc; i++) {
 233     arg->argv[i] = xbt_strdup(argv[i]);
 234   }
 235   arg->argv[argc] = NULL;
 236
 237   arg->properties = properties;
 238   arg->auto_restart = auto_restart;
 239
 240   if( SIMIX_host_get_state(host) == SURF_RESOURCE_OFF
 241       && !xbt_dict_get_or_null(watched_hosts_lib,sg_host_name(host))){
 242     xbt_dict_set(watched_hosts_lib,sg_host_name(host),host,NULL);
 243     XBT_DEBUG("Have push host %s to watched_hosts_lib because state == SURF_RESOURCE_OFF",sg_host_name(host));
 244   }
 245   xbt_dynar_push_as(SIMIX_host_priv(host)->auto_restart_processes,smx_process_arg_t,arg);
 246 }
 247 /**
 248  * \brief Restart the list of processes that have been registered to the host
 249  */
 250 void SIMIX_host_restart_processes(smx_host_t host)
 251 {
 252   unsigned int cpt;
 253   smx_process_arg_t arg;
 254   xbt_dynar_foreach(SIMIX_host_priv(host)->auto_restart_processes,cpt,arg) {
 255
 256     smx_process_t process;
 257
 258     XBT_DEBUG("Restarting Process %s(%s) right now", arg->argv[0], arg->hostname);
 259     if (simix_global->create_process_function) {
 260       simix_global->create_process_function(&process,
 261                                             arg->argv[0],
 262                                             arg->code,
 263                                             NULL,
 264                                             arg->hostname,
 265                                             arg->kill_time,
 266                                             arg->argc,
 267                                             arg->argv,
 268                                             arg->properties,
 269                                             arg->auto_restart);
 270     }
 271     else {
 272       simcall_process_create(&process,
 273                                             arg->argv[0],
 274                                             arg->code,
 275                                             NULL,
 276                                             arg->hostname,
 277                                             arg->kill_time,
 278                                             arg->argc,
 279                                             arg->argv,
 280                                             arg->properties,
 281                                             arg->auto_restart);
 282
 283     }
 284   }
 285   xbt_dynar_reset(SIMIX_host_priv(host)->auto_restart_processes);
 286 }
 287
 288 void SIMIX_host_autorestart(smx_host_t host)
 289 {
 290   if(simix_global->autorestart)
 291     simix_global->autorestart(host);
 292   else
 293     xbt_die("No function for simix_global->autorestart");
 294 }
 295
 296 void SIMIX_pre_host_set_data(smx_simcall_t simcall, smx_host_t host, void *data) {
 297   SIMIX_host_set_data(host, data);
 298 }
 299 void SIMIX_host_set_data(smx_host_t host, void *data){
 300   xbt_assert((host != NULL), "Invalid parameters");
 301   xbt_assert((SIMIX_host_priv(host)->data == NULL), "Data already set");
 302
 303   SIMIX_host_priv(host)->data = data;
 304 }
 305
 306 smx_action_t SIMIX_pre_host_execute(smx_simcall_t simcall,const char *name,
 307     smx_host_t host, double computation_amount, double priority){
 308   return SIMIX_host_execute(name, host, computation_amount, priority);
 309 }
 310 smx_action_t SIMIX_host_execute(const char *name,
 311     smx_host_t host, double computation_amount, double priority){
 312
 313   /* alloc structures and initialize */
 314   smx_action_t action = xbt_mallocator_get(simix_global->action_mallocator);
 315   action->type = SIMIX_ACTION_EXECUTE;
 316   action->name = xbt_strdup(name);
 317   action->state = SIMIX_RUNNING;
 318   action->execution.host = host;
 319
 320 #ifdef HAVE_TRACING
 321   action->category = NULL;
 322 #endif
 323
 324   /* set surf's action */
 325   if (!MC_is_active()) {
 326     action->execution.surf_exec =
 327       surf_workstation_model->extension.workstation.execute(host,
 328     computation_amount);
 329     surf_workstation_model->action_data_set(action->execution.surf_exec, action);
 330     surf_workstation_model->set_priority(action->execution.surf_exec, priority);
 331   }
 332
 333   XBT_DEBUG("Create execute action %p", action);
 334
 335   return action;
 336 }
 337
 338 smx_action_t SIMIX_pre_host_parallel_execute(smx_simcall_t simcall, const char *name,
 339     int host_nb, smx_host_t *host_list,
 340     double *computation_amount, double *communication_amount,
 341     double amount, double rate){
 342   return SIMIX_host_parallel_execute(name, host_nb, host_list, computation_amount,
 343                                      communication_amount, amount, rate);
 344 }
 345 smx_action_t SIMIX_host_parallel_execute(const char *name,
 346     int host_nb, smx_host_t *host_list,
 347     double *computation_amount, double *communication_amount,
 348     double amount, double rate){
 349
 350   void **workstation_list = NULL;
 351   int i;
 352
 353   /* alloc structures and initialize */
 354   smx_action_t action = xbt_mallocator_get(simix_global->action_mallocator);
 355   action->type = SIMIX_ACTION_PARALLEL_EXECUTE;
 356   action->name = xbt_strdup(name);
 357   action->state = SIMIX_RUNNING;
 358   action->execution.host = NULL; /* FIXME: do we need the list of hosts? */
 359
 360 #ifdef HAVE_TRACING
 361   action->category = NULL;
 362 #endif
 363
 364   /* set surf's action */
 365   workstation_list = xbt_new0(void *, host_nb);
 366   for (i = 0; i < host_nb; i++)
 367     workstation_list[i] = host_list[i];
 368
 369   /* set surf's action */
 370   if (!MC_is_active()) {
 371     action->execution.surf_exec =
 372       surf_workstation_model->extension.workstation.
 373       execute_parallel_task(host_nb, workstation_list, computation_amount,
 374                       communication_amount, rate);
 375
 376     surf_workstation_model->action_data_set(action->execution.surf_exec, action);
 377   }
 378   XBT_DEBUG("Create parallel execute action %p", action);
 379
 380   return action;
 381 }
 382
 383 void SIMIX_pre_host_execution_destroy(smx_simcall_t simcall, smx_action_t action){
 384   SIMIX_host_execution_destroy(action);
 385 }
 386 void SIMIX_host_execution_destroy(smx_action_t action){
 387   XBT_DEBUG("Destroy action %p", action);
 388
 389   if (action->execution.surf_exec) {
 390     surf_workstation_model->action_unref(action->execution.surf_exec);
 391     action->execution.surf_exec = NULL;
 392   }
 393   xbt_free(action->name);
 394   xbt_mallocator_release(simix_global->action_mallocator, action);
 395 }
 396
 397 void SIMIX_pre_host_execution_cancel(smx_simcall_t simcall, smx_action_t action){
 398   SIMIX_host_execution_cancel(action);
 399 }
 400 void SIMIX_host_execution_cancel(smx_action_t action){
 401   XBT_DEBUG("Cancel action %p", action);
 402
 403   if (action->execution.surf_exec)
 404     surf_workstation_model->action_cancel(action->execution.surf_exec);
 405 }
 406
 407 double SIMIX_pre_host_execution_get_remains(smx_simcall_t simcall, smx_action_t action){
 408   return SIMIX_host_execution_get_remains(action);
 409 }
 410 double SIMIX_host_execution_get_remains(smx_action_t action){
 411   double result = 0.0;
 412
 413   if (action->state == SIMIX_RUNNING)
 414     result = surf_workstation_model->get_remains(action->execution.surf_exec);
 415
 416   return result;
 417 }
 418
 419 e_smx_state_t SIMIX_pre_host_execution_get_state(smx_simcall_t simcall, smx_action_t action){
 420   return SIMIX_host_execution_get_state(action);
 421 }
 422 e_smx_state_t SIMIX_host_execution_get_state(smx_action_t action){
 423   return action->state;
 424 }
 425
 426 void SIMIX_pre_host_execution_set_priority(smx_simcall_t simcall, smx_action_t action,
 427                                         double priority){
 428   return SIMIX_host_execution_set_priority(action, priority);
 429 }
 430 void SIMIX_host_execution_set_priority(smx_action_t action, double priority){
 431   if(action->execution.surf_exec)
 432     surf_workstation_model->set_priority(action->execution.surf_exec, priority);
 433 }
 434
 435 void SIMIX_pre_host_execution_wait(smx_simcall_t simcall, smx_action_t action){
 436
 437   XBT_DEBUG("Wait for execution of action %p, state %d", action, (int)action->state);
 438
 439   /* Associate this simcall to the action */
 440   xbt_fifo_push(action->simcalls, simcall);
 441   simcall->issuer->waiting_action = action;
 442
 443   /* set surf's action */
 444   if (MC_is_active()) {
 445     action->state = SIMIX_DONE;
 446     SIMIX_execution_finish(action);
 447     return;
 448   }
 449
 450   /* If the action is already finished then perform the error handling */
 451   if (action->state != SIMIX_RUNNING)
 452     SIMIX_execution_finish(action);
 453 }
 454
 455 void SIMIX_host_execution_suspend(smx_action_t action)
 456 {
 457   if(action->execution.surf_exec)
 458     surf_workstation_model->suspend(action->execution.surf_exec);
 459 }
 460
 461 void SIMIX_host_execution_resume(smx_action_t action)
 462 {
 463   if(action->execution.surf_exec)
 464     surf_workstation_model->resume(action->execution.surf_exec);
 465 }
 466
 467 void SIMIX_execution_finish(smx_action_t action)
 468 {
 469   xbt_fifo_item_t item;
 470   smx_simcall_t simcall;
 471
 472   xbt_fifo_foreach(action->simcalls, item, simcall, smx_simcall_t) {
 473
 474     switch (action->state) {
 475
 476       case SIMIX_DONE:
 477         /* do nothing, action done */
 478   XBT_DEBUG("SIMIX_execution_finished: execution successful");
 479         break;
 480
 481       case SIMIX_FAILED:
 482         XBT_DEBUG("SIMIX_execution_finished: host '%s' failed", sg_host_name(simcall->issuer->smx_host));
 483         simcall->issuer->context->iwannadie = 1;
 484         //SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
 485         break;
 486
 487       case SIMIX_CANCELED:
 488         XBT_DEBUG("SIMIX_execution_finished: execution canceled");
 489         SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Canceled");
 490         break;
 491
 492       default:
 493         xbt_die("Internal error in SIMIX_execution_finish: unexpected action state %d",
 494             (int)action->state);
 495     }
 496     /* check if the host is down */
 497     if (surf_workstation_model->extension.
 498         workstation.get_state(simcall->issuer->smx_host) != SURF_RESOURCE_ON) {
 499       simcall->issuer->context->iwannadie = 1;
 500     }
 501
 502     simcall->issuer->waiting_action =    NULL;
 503     simcall_host_execution_wait__set__result(simcall, action->state);
 504     SIMIX_simcall_answer(simcall);
 505   }
 506
 507   /* We no longer need it */
 508   SIMIX_host_execution_destroy(action);
 509 }
 510
 511 void SIMIX_post_host_execute(smx_action_t action)
 512 {
 513   if (action->type == SIMIX_ACTION_EXECUTE && /* FIMXE: handle resource failure
 514                                                * for parallel tasks too */
 515       surf_workstation_model->extension.workstation.get_state(action->execution.host) == SURF_RESOURCE_OFF) {
 516     /* If the host running the action failed, notice it so that the asking
 517      * process can be killed if it runs on that host itself */
 518     action->state = SIMIX_FAILED;
 519   } else if (surf_workstation_model->action_state_get(action->execution.surf_exec) == SURF_ACTION_FAILED) {
 520     /* If the host running the action didn't fail, then the action was
 521      * canceled */
 522     action->state = SIMIX_CANCELED;
 523   } else {
 524     action->state = SIMIX_DONE;
 525   }
 526
 527   if (action->execution.surf_exec) {
 528     surf_workstation_model->action_unref(action->execution.surf_exec);
 529     action->execution.surf_exec = NULL;
 530   }
 531
 532   /* If there are simcalls associated with the action, then answer them */
 533   if (xbt_fifo_size(action->simcalls)) {
 534     SIMIX_execution_finish(action);
 535   }
 536 }
 537
 538
 539 #ifdef HAVE_TRACING
 540 void SIMIX_pre_set_category(smx_simcall_t simcall, smx_action_t action,
 541                             const char *category){
 542   SIMIX_set_category(action, category);
 543 }
 544 void SIMIX_set_category(smx_action_t action, const char *category)
 545 {
 546   if (action->state != SIMIX_RUNNING) return;
 547   if (action->type == SIMIX_ACTION_EXECUTE){
 548     surf_workstation_model->set_category(action->execution.surf_exec, category);
 549   }else if (action->type == SIMIX_ACTION_COMMUNICATE){
 550     surf_workstation_model->set_category(action->comm.surf_comm, category);
 551   }
 552 }
 553 #endif
 554