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 int SIMIX_pre_host_get_core(smx_simcall_t simcall, smx_host_t host){
 152   return SIMIX_host_get_core(host);
 153 }
 154 int SIMIX_host_get_core(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_core(host);
 159 }
 160
 161
 162
 163 double SIMIX_pre_host_get_available_speed(smx_simcall_t simcall, smx_host_t host){
 164   return SIMIX_host_get_available_speed(host);
 165 }
 166 double SIMIX_host_get_available_speed(smx_host_t host){
 167   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 168
 169   return surf_workstation_model->extension.workstation.
 170       get_available_speed(host);
 171 }
 172
 173 int SIMIX_pre_host_get_state(smx_simcall_t simcall, smx_host_t host){
 174   return SIMIX_host_get_state(host);
 175 }
 176 int SIMIX_host_get_state(smx_host_t host){
 177   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 178
 179   return surf_workstation_model->extension.workstation.
 180       get_state(host);
 181 }
 182
 183 void* SIMIX_pre_host_self_get_data(smx_simcall_t simcall){
 184   return SIMIX_host_self_get_data();
 185 }
 186 void* SIMIX_host_self_get_data(void)
 187 {
 188   smx_host_t self = SIMIX_host_self();
 189   return SIMIX_host_get_data(self);
 190 }
 191
 192 void SIMIX_host_self_set_data(void *data)
 193 {
 194   smx_host_t self = SIMIX_host_self();
 195   SIMIX_host_set_data(self, data);
 196 }
 197
 198 void* SIMIX_pre_host_get_data(smx_simcall_t simcall,smx_host_t host){
 199   return SIMIX_host_get_data(host);
 200 }
 201 void* SIMIX_host_get_data(smx_host_t host){
 202   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 203
 204   return SIMIX_host_priv(host)->data;
 205 }
 206 void _SIMIX_host_free_process_arg(void *);
 207 void _SIMIX_host_free_process_arg(void *data)
 208 {
 209   smx_process_arg_t arg = *(void**)data;
 210   xbt_free(arg->name);
 211   xbt_free(arg);
 212 }
 213 /**
 214  * \brief Add a process to the list of the processes that the host will restart when it comes back
 215  * This function add a process to the list of the processes that will be restarted when the host comes
 216  * back. It is expected that this function is called when the host is down.
 217  * The processes will only be restarted once, meaning that you will have to register the process
 218  * again to restart the process again.
 219  */
 220 void SIMIX_host_add_auto_restart_process(smx_host_t host,
 221                                          const char *name,
 222                                          xbt_main_func_t code,
 223                                          void *data,
 224                                          const char *hostname,
 225                                          double kill_time,
 226                                          int argc, char **argv,
 227                                          xbt_dict_t properties,
 228                                          int auto_restart)
 229 {
 230   if (!SIMIX_host_priv(host)->auto_restart_processes) {
 231     SIMIX_host_priv(host)->auto_restart_processes = xbt_dynar_new(sizeof(smx_process_arg_t),_SIMIX_host_free_process_arg);
 232   }
 233   smx_process_arg_t arg = xbt_new(s_smx_process_arg_t,1);
 234   arg->name = xbt_strdup(name);
 235   arg->code = code;
 236   arg->data = data;
 237   arg->hostname = hostname;
 238   arg->kill_time = kill_time;
 239   arg->argc = argc;
 240
 241   arg->argv = xbt_new(char*,argc + 1);
 242
 243   int i;
 244   for (i = 0; i < argc; i++) {
 245     arg->argv[i] = xbt_strdup(argv[i]);
 246   }
 247   arg->argv[argc] = NULL;
 248
 249   arg->properties = properties;
 250   arg->auto_restart = auto_restart;
 251
 252   if( SIMIX_host_get_state(host) == SURF_RESOURCE_OFF
 253       && !xbt_dict_get_or_null(watched_hosts_lib,sg_host_name(host))){
 254     xbt_dict_set(watched_hosts_lib,sg_host_name(host),host,NULL);
 255     XBT_DEBUG("Have push host %s to watched_hosts_lib because state == SURF_RESOURCE_OFF",sg_host_name(host));
 256   }
 257   xbt_dynar_push_as(SIMIX_host_priv(host)->auto_restart_processes,smx_process_arg_t,arg);
 258 }
 259 /**
 260  * \brief Restart the list of processes that have been registered to the host
 261  */
 262 void SIMIX_host_restart_processes(smx_host_t host)
 263 {
 264   unsigned int cpt;
 265   smx_process_arg_t arg;
 266   xbt_dynar_foreach(SIMIX_host_priv(host)->auto_restart_processes,cpt,arg) {
 267
 268     smx_process_t process;
 269
 270     XBT_DEBUG("Restarting Process %s(%s) right now", arg->argv[0], arg->hostname);
 271     if (simix_global->create_process_function) {
 272       simix_global->create_process_function(&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     else {
 284       simcall_process_create(&process,
 285                                             arg->argv[0],
 286                                             arg->code,
 287                                             NULL,
 288                                             arg->hostname,
 289                                             arg->kill_time,
 290                                             arg->argc,
 291                                             arg->argv,
 292                                             arg->properties,
 293                                             arg->auto_restart);
 294
 295     }
 296   }
 297   xbt_dynar_reset(SIMIX_host_priv(host)->auto_restart_processes);
 298 }
 299
 300 void SIMIX_host_autorestart(smx_host_t host)
 301 {
 302   if(simix_global->autorestart)
 303     simix_global->autorestart(host);
 304   else
 305     xbt_die("No function for simix_global->autorestart");
 306 }
 307
 308 void SIMIX_pre_host_set_data(smx_simcall_t simcall, smx_host_t host, void *data) {
 309   SIMIX_host_set_data(host, data);
 310 }
 311 void SIMIX_host_set_data(smx_host_t host, void *data){
 312   xbt_assert((host != NULL), "Invalid parameters");
 313   xbt_assert((SIMIX_host_priv(host)->data == NULL), "Data already set");
 314
 315   SIMIX_host_priv(host)->data = data;
 316 }
 317
 318 smx_action_t SIMIX_pre_host_execute(smx_simcall_t simcall,const char *name,
 319     smx_host_t host, double computation_amount, double priority){
 320   return SIMIX_host_execute(name, host, computation_amount, priority);
 321 }
 322 smx_action_t SIMIX_host_execute(const char *name,
 323     smx_host_t host, double computation_amount, double priority){
 324
 325   /* alloc structures and initialize */
 326   smx_action_t action = xbt_mallocator_get(simix_global->action_mallocator);
 327   action->type = SIMIX_ACTION_EXECUTE;
 328   action->name = xbt_strdup(name);
 329   action->state = SIMIX_RUNNING;
 330   action->execution.host = host;
 331
 332 #ifdef HAVE_TRACING
 333   action->category = NULL;
 334 #endif
 335
 336   /* set surf's action */
 337   if (!MC_is_active()) {
 338     action->execution.surf_exec =
 339       surf_workstation_model->extension.workstation.execute(host,
 340     computation_amount);
 341     surf_workstation_model->action_data_set(action->execution.surf_exec, action);
 342     surf_workstation_model->set_priority(action->execution.surf_exec, priority);
 343   }
 344
 345   XBT_DEBUG("Create execute action %p", action);
 346
 347   return action;
 348 }
 349
 350 smx_action_t SIMIX_pre_host_parallel_execute(smx_simcall_t simcall, const char *name,
 351     int host_nb, smx_host_t *host_list,
 352     double *computation_amount, double *communication_amount,
 353     double amount, double rate){
 354   return SIMIX_host_parallel_execute(name, host_nb, host_list, computation_amount,
 355                                      communication_amount, amount, rate);
 356 }
 357 smx_action_t SIMIX_host_parallel_execute(const char *name,
 358     int host_nb, smx_host_t *host_list,
 359     double *computation_amount, double *communication_amount,
 360     double amount, double rate){
 361
 362   void **workstation_list = NULL;
 363   int i;
 364
 365   /* alloc structures and initialize */
 366   smx_action_t action = xbt_mallocator_get(simix_global->action_mallocator);
 367   action->type = SIMIX_ACTION_PARALLEL_EXECUTE;
 368   action->name = xbt_strdup(name);
 369   action->state = SIMIX_RUNNING;
 370   action->execution.host = NULL; /* FIXME: do we need the list of hosts? */
 371
 372 #ifdef HAVE_TRACING
 373   action->category = NULL;
 374 #endif
 375
 376   /* set surf's action */
 377   workstation_list = xbt_new0(void *, host_nb);
 378   for (i = 0; i < host_nb; i++)
 379     workstation_list[i] = host_list[i];
 380
 381   /* set surf's action */
 382   if (!MC_is_active()) {
 383     action->execution.surf_exec =
 384       surf_workstation_model->extension.workstation.
 385       execute_parallel_task(host_nb, workstation_list, computation_amount,
 386                       communication_amount, rate);
 387
 388     surf_workstation_model->action_data_set(action->execution.surf_exec, action);
 389   }
 390   XBT_DEBUG("Create parallel execute action %p", action);
 391
 392   return action;
 393 }
 394
 395 void SIMIX_pre_host_execution_destroy(smx_simcall_t simcall, smx_action_t action){
 396   SIMIX_host_execution_destroy(action);
 397 }
 398 void SIMIX_host_execution_destroy(smx_action_t action){
 399   XBT_DEBUG("Destroy action %p", action);
 400
 401   if (action->execution.surf_exec) {
 402     surf_workstation_model->action_unref(action->execution.surf_exec);
 403     action->execution.surf_exec = NULL;
 404   }
 405   xbt_free(action->name);
 406   xbt_mallocator_release(simix_global->action_mallocator, action);
 407 }
 408
 409 void SIMIX_pre_host_execution_cancel(smx_simcall_t simcall, smx_action_t action){
 410   SIMIX_host_execution_cancel(action);
 411 }
 412 void SIMIX_host_execution_cancel(smx_action_t action){
 413   XBT_DEBUG("Cancel action %p", action);
 414
 415   if (action->execution.surf_exec)
 416     surf_workstation_model->action_cancel(action->execution.surf_exec);
 417 }
 418
 419 double SIMIX_pre_host_execution_get_remains(smx_simcall_t simcall, smx_action_t action){
 420   return SIMIX_host_execution_get_remains(action);
 421 }
 422 double SIMIX_host_execution_get_remains(smx_action_t action){
 423   double result = 0.0;
 424
 425   if (action->state == SIMIX_RUNNING)
 426     result = surf_workstation_model->get_remains(action->execution.surf_exec);
 427
 428   return result;
 429 }
 430
 431 e_smx_state_t SIMIX_pre_host_execution_get_state(smx_simcall_t simcall, smx_action_t action){
 432   return SIMIX_host_execution_get_state(action);
 433 }
 434 e_smx_state_t SIMIX_host_execution_get_state(smx_action_t action){
 435   return action->state;
 436 }
 437
 438 void SIMIX_pre_host_execution_set_priority(smx_simcall_t simcall, smx_action_t action,
 439                                         double priority){
 440   return SIMIX_host_execution_set_priority(action, priority);
 441 }
 442 void SIMIX_host_execution_set_priority(smx_action_t action, double priority){
 443   if(action->execution.surf_exec)
 444     surf_workstation_model->set_priority(action->execution.surf_exec, priority);
 445 }
 446
 447 void SIMIX_pre_host_execution_wait(smx_simcall_t simcall, smx_action_t action){
 448
 449   XBT_DEBUG("Wait for execution of action %p, state %d", action, (int)action->state);
 450
 451   /* Associate this simcall to the action */
 452   xbt_fifo_push(action->simcalls, simcall);
 453   simcall->issuer->waiting_action = action;
 454
 455   /* set surf's action */
 456   if (MC_is_active()) {
 457     action->state = SIMIX_DONE;
 458     SIMIX_execution_finish(action);
 459     return;
 460   }
 461
 462   /* If the action is already finished then perform the error handling */
 463   if (action->state != SIMIX_RUNNING)
 464     SIMIX_execution_finish(action);
 465 }
 466
 467 void SIMIX_host_execution_suspend(smx_action_t action)
 468 {
 469   if(action->execution.surf_exec)
 470     surf_workstation_model->suspend(action->execution.surf_exec);
 471 }
 472
 473 void SIMIX_host_execution_resume(smx_action_t action)
 474 {
 475   if(action->execution.surf_exec)
 476     surf_workstation_model->resume(action->execution.surf_exec);
 477 }
 478
 479 void SIMIX_execution_finish(smx_action_t action)
 480 {
 481   xbt_fifo_item_t item;
 482   smx_simcall_t simcall;
 483
 484   xbt_fifo_foreach(action->simcalls, item, simcall, smx_simcall_t) {
 485
 486     switch (action->state) {
 487
 488       case SIMIX_DONE:
 489         /* do nothing, action done */
 490   XBT_DEBUG("SIMIX_execution_finished: execution successful");
 491         break;
 492
 493       case SIMIX_FAILED:
 494         XBT_DEBUG("SIMIX_execution_finished: host '%s' failed", sg_host_name(simcall->issuer->smx_host));
 495         simcall->issuer->context->iwannadie = 1;
 496         //SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
 497         break;
 498
 499       case SIMIX_CANCELED:
 500         XBT_DEBUG("SIMIX_execution_finished: execution canceled");
 501         SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Canceled");
 502         break;
 503
 504       default:
 505         xbt_die("Internal error in SIMIX_execution_finish: unexpected action state %d",
 506             (int)action->state);
 507     }
 508     /* check if the host is down */
 509     if (surf_workstation_model->extension.
 510         workstation.get_state(simcall->issuer->smx_host) != SURF_RESOURCE_ON) {
 511       simcall->issuer->context->iwannadie = 1;
 512     }
 513
 514     simcall->issuer->waiting_action =    NULL;
 515     simcall_host_execution_wait__set__result(simcall, action->state);
 516     SIMIX_simcall_answer(simcall);
 517   }
 518
 519   /* We no longer need it */
 520   SIMIX_host_execution_destroy(action);
 521 }
 522
 523 void SIMIX_post_host_execute(smx_action_t action)
 524 {
 525   if (action->type == SIMIX_ACTION_EXECUTE && /* FIMXE: handle resource failure
 526                                                * for parallel tasks too */
 527       surf_workstation_model->extension.workstation.get_state(action->execution.host) == SURF_RESOURCE_OFF) {
 528     /* If the host running the action failed, notice it so that the asking
 529      * process can be killed if it runs on that host itself */
 530     action->state = SIMIX_FAILED;
 531   } else if (surf_workstation_model->action_state_get(action->execution.surf_exec) == SURF_ACTION_FAILED) {
 532     /* If the host running the action didn't fail, then the action was
 533      * canceled */
 534     action->state = SIMIX_CANCELED;
 535   } else {
 536     action->state = SIMIX_DONE;
 537   }
 538
 539   if (action->execution.surf_exec) {
 540     surf_workstation_model->action_unref(action->execution.surf_exec);
 541     action->execution.surf_exec = NULL;
 542   }
 543
 544   /* If there are simcalls associated with the action, then answer them */
 545   if (xbt_fifo_size(action->simcalls)) {
 546     SIMIX_execution_finish(action);
 547   }
 548 }
 549
 550
 551 #ifdef HAVE_TRACING
 552 void SIMIX_pre_set_category(smx_simcall_t simcall, smx_action_t action,
 553                             const char *category){
 554   SIMIX_set_category(action, category);
 555 }
 556 void SIMIX_set_category(smx_action_t action, const char *category)
 557 {
 558   if (action->state != SIMIX_RUNNING) return;
 559   if (action->type == SIMIX_ACTION_EXECUTE){
 560     surf_workstation_model->set_category(action->execution.surf_exec, category);
 561   }else if (action->type == SIMIX_ACTION_COMMUNICATE){
 562     surf_workstation_model->set_category(action->comm.surf_comm, category);
 563   }
 564 }
 565 #endif
 566