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