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_t smx_host = xbt_new0(s_smx_host_t, 1);
  28   s_smx_process_t proc;
  29
  30   /* Host structure */
  31   smx_host->name = xbt_strdup(name);
  32   smx_host->data = data;
  33   smx_host->host = workstation;
  34   smx_host->process_list =
  35       xbt_swag_new(xbt_swag_offset(proc, host_proc_hookup));
  36
  37   /* Update global variables */
  38   xbt_lib_set(host_lib,smx_host->name,SIMIX_HOST_LEVEL,smx_host);
  39
  40   return smx_host;
  41 }
  42
  43 /**
  44  * \brief Internal function to destroy a SIMIX host.
  45  *
  46  * \param h the host to destroy (a smx_host_t)
  47  */
  48 void SIMIX_host_destroy(void *h)
  49 {
  50   smx_host_t host = (smx_host_t) h;
  51
  52   xbt_assert((host != NULL), "Invalid parameters");
  53
  54   /* Clean Simulator data */
  55   if (xbt_swag_size(host->process_list) != 0) {
  56     char *msg =
  57         bprintf("Shutting down host %s, but it's not empty:", host->name);
  58     char *tmp;
  59     smx_process_t process = NULL;
  60
  61     xbt_swag_foreach(process, host->process_list) {
  62       tmp = bprintf("%s\n\t%s", msg, process->name);
  63       free(msg);
  64       msg = tmp;
  65     }
  66     SIMIX_display_process_status();
  67     THROWF(arg_error, 0, "%s", msg);
  68   }
  69   xbt_dynar_free(&host->auto_restart_processes);
  70   xbt_swag_free(host->process_list);
  71
  72   /* Clean host structure */
  73   free(host->name);
  74   free(host);
  75
  76   return;
  77 }
  78
  79 ///**
  80 // * \brief Returns a dict of all hosts.
  81 // *
  82 // * \return List of all hosts (as a #xbt_dict_t)
  83 // */
  84 //xbt_dict_t SIMIX_host_get_dict(void)
  85 //{
  86 //  xbt_dict_t host_dict = xbt_dict_new_homogeneous(NULL);
  87 //  xbt_lib_cursor_t cursor = NULL;
  88 //  char *name = NULL;
  89 //  void **host = NULL;
  90 //
  91 //  xbt_lib_foreach(host_lib, cursor, name, host){
  92 //    if(host[SIMIX_HOST_LEVEL])
  93 //            xbt_dict_set(host_dict,name,host[SIMIX_HOST_LEVEL], NULL);
  94 //  }
  95 //  return host_dict;
  96 //}
  97 smx_host_t SIMIX_pre_host_get_by_name(smx_simcall_t simcall, const char *name){
  98    return SIMIX_host_get_by_name(name);
  99 }
 100 smx_host_t SIMIX_host_get_by_name(const char *name){
 101   xbt_assert(((simix_global != NULL)
 102                && (host_lib != NULL)),
 103               "Environment not set yet");
 104
 105   return xbt_lib_get_or_null(host_lib, name, SIMIX_HOST_LEVEL);
 106 }
 107
 108 smx_host_t SIMIX_host_self(void)
 109 {
 110   smx_process_t process = SIMIX_process_self();
 111   return (process == NULL) ? NULL : SIMIX_process_get_host(process);
 112 }
 113
 114 const char* SIMIX_pre_host_self_get_name(smx_simcall_t simcall){
 115    return SIMIX_host_self_get_name();
 116 }
 117 /* needs to be public and without simcall because it is called
 118    by exceptions and logging events */
 119 const char* SIMIX_host_self_get_name(void)
 120 {
 121   smx_host_t host = SIMIX_host_self();
 122   if (host == NULL || SIMIX_process_self() == simix_global->maestro_process)
 123     return "";
 124
 125   return SIMIX_host_get_name(host);
 126 }
 127
 128 const char* SIMIX_pre_host_get_name(smx_simcall_t simcall, smx_host_t host){
 129    return SIMIX_host_get_name(host);
 130 }
 131 const char* SIMIX_host_get_name(smx_host_t host){
 132   xbt_assert((host != NULL), "Invalid parameters");
 133
 134   return host->name;
 135 }
 136
 137 xbt_dict_t SIMIX_pre_host_get_properties(smx_simcall_t simcall, smx_host_t host){
 138   return SIMIX_host_get_properties(host);
 139 }
 140 xbt_dict_t SIMIX_host_get_properties(smx_host_t host){
 141   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 142
 143   return surf_workstation_model->extension.workstation.get_properties(host->host);
 144 }
 145
 146 double SIMIX_pre_host_get_speed(smx_simcall_t simcall, smx_host_t host){
 147   return SIMIX_host_get_speed(host);
 148 }
 149 double SIMIX_host_get_speed(smx_host_t host){
 150   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 151
 152   return surf_workstation_model->extension.workstation.
 153       get_speed(host->host, 1.0);
 154 }
 155
 156 double SIMIX_pre_host_get_available_speed(smx_simcall_t simcall, smx_host_t host){
 157   return SIMIX_host_get_available_speed(host);
 158 }
 159 double SIMIX_host_get_available_speed(smx_host_t host){
 160   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 161
 162   return surf_workstation_model->extension.workstation.
 163       get_available_speed(host->host);
 164 }
 165
 166 int SIMIX_pre_host_get_state(smx_simcall_t simcall, smx_host_t host){
 167   return SIMIX_host_get_state(host);
 168 }
 169 int SIMIX_host_get_state(smx_host_t host){
 170   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 171
 172   return surf_workstation_model->extension.workstation.
 173       get_state(host->host);
 174 }
 175
 176 void* SIMIX_pre_host_self_get_data(smx_simcall_t simcall){
 177   return SIMIX_host_self_get_data();
 178 }
 179 void* SIMIX_host_self_get_data(void)
 180 {
 181   smx_host_t self = SIMIX_host_self();
 182   return SIMIX_host_get_data(self);
 183 }
 184
 185 void SIMIX_host_self_set_data(void *data)
 186 {
 187   smx_host_t self = SIMIX_host_self();
 188   SIMIX_host_set_data(self, data);
 189 }
 190
 191 void* SIMIX_pre_host_get_data(smx_simcall_t simcall,smx_host_t host){
 192   return SIMIX_host_get_data(host);
 193 }
 194 void* SIMIX_host_get_data(smx_host_t host){
 195   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 196
 197   return host->data;
 198 }
 199 void _SIMIX_host_free_process_arg(void *);
 200 void _SIMIX_host_free_process_arg(void *data)
 201 {
 202   smx_process_arg_t arg = *(void**)data;
 203   xbt_free(arg->name);
 204   xbt_free(arg);
 205 }
 206 /**
 207  * \brief Add a process to the list of the processes that the host will restart when it comes back
 208  * This function add a process to the list of the processes that will be restarted when the host comes
 209  * back. It is expected that this function is called when the host is down.
 210  * The processes will only be restarted once, meaning that you will have to register the process
 211  * again to restart the process again.
 212  */
 213 void SIMIX_host_add_auto_restart_process(smx_host_t host,
 214                                          const char *name,
 215                                          xbt_main_func_t code,
 216                                          void *data,
 217                                          const char *hostname,
 218                                          double kill_time,
 219                                          int argc, char **argv,
 220                                          xbt_dict_t properties,
 221                                          int auto_restart)
 222 {
 223   if (!host->auto_restart_processes) {
 224     host->auto_restart_processes = xbt_dynar_new(sizeof(smx_process_arg_t),_SIMIX_host_free_process_arg);
 225   }
 226   smx_process_arg_t arg = xbt_new(s_smx_process_arg_t,1);
 227   arg->name = xbt_strdup(name);
 228   arg->code = code;
 229   arg->data = data;
 230   arg->hostname = hostname;
 231   arg->kill_time = kill_time;
 232   arg->argc = argc;
 233
 234   arg->argv = xbt_new(char*,argc + 1);
 235
 236   int i;
 237   for (i = 0; i < argc; i++) {
 238     arg->argv[i] = xbt_strdup(argv[i]);
 239   }
 240   arg->argv[argc] = NULL;
 241
 242   arg->properties = properties;
 243   arg->auto_restart = auto_restart;
 244
 245   if( SIMIX_host_get_state(host) == SURF_RESOURCE_OFF
 246       && !xbt_dict_get_or_null(watched_hosts_lib,host->name)){
 247     xbt_dict_set(watched_hosts_lib,host->name,host,NULL);
 248     XBT_DEBUG("Have push host %s to watched_hosts_lib because state == SURF_RESOURCE_OFF",host->name);
 249   }
 250   xbt_dynar_push_as(host->auto_restart_processes,smx_process_arg_t,arg);
 251 }
 252 /**
 253  * \brief Restart the list of processes that have been registered to the host
 254  */
 255 void SIMIX_host_restart_processes(smx_host_t host)
 256 {
 257   unsigned int cpt;
 258   smx_process_arg_t arg;
 259   xbt_dynar_foreach(host->auto_restart_processes,cpt,arg) {
 260
 261     smx_process_t process;
 262
 263     XBT_DEBUG("Restarting Process %s(%s) right now", arg->argv[0], arg->hostname);
 264     if (simix_global->create_process_function) {
 265       simix_global->create_process_function(&process,
 266                                             arg->argv[0],
 267                                             arg->code,
 268                                             NULL,
 269                                             arg->hostname,
 270                                             arg->kill_time,
 271                                             arg->argc,
 272                                             arg->argv,
 273                                             arg->properties,
 274                                             arg->auto_restart);
 275     }
 276     else {
 277       simcall_process_create(&process,
 278                                             arg->argv[0],
 279                                             arg->code,
 280                                             NULL,
 281                                             arg->hostname,
 282                                             arg->kill_time,
 283                                             arg->argc,
 284                                             arg->argv,
 285                                             arg->properties,
 286                                             arg->auto_restart);
 287
 288     }
 289   }
 290   xbt_dynar_reset(host->auto_restart_processes);
 291 }
 292
 293 void SIMIX_host_autorestart(smx_host_t host)
 294 {
 295   if(simix_global->autorestart)
 296     simix_global->autorestart(host);
 297   else
 298     xbt_die("No function for simix_global->autorestart");
 299 }
 300
 301 void SIMIX_pre_host_set_data(smx_simcall_t simcall, smx_host_t host, void *data) {
 302   SIMIX_host_set_data(host, data);
 303 }
 304 void SIMIX_host_set_data(smx_host_t host, void *data){
 305   xbt_assert((host != NULL), "Invalid parameters");
 306   xbt_assert((host->data == NULL), "Data already set");
 307
 308   host->data = data;
 309 }
 310
 311 smx_action_t SIMIX_pre_host_execute(smx_simcall_t simcall,const char *name,
 312     smx_host_t host, double computation_amount, double priority){
 313   return SIMIX_host_execute(name, host, computation_amount, priority);
 314 }
 315 smx_action_t SIMIX_host_execute(const char *name,
 316     smx_host_t host, double computation_amount, double priority){
 317
 318   /* alloc structures and initialize */
 319   smx_action_t action = xbt_mallocator_get(simix_global->action_mallocator);
 320   action->type = SIMIX_ACTION_EXECUTE;
 321   action->name = xbt_strdup(name);
 322   action->state = SIMIX_RUNNING;
 323   action->execution.host = host;
 324
 325 #ifdef HAVE_TRACING
 326   action->category = NULL;
 327 #endif
 328
 329   /* set surf's action */
 330   if (!MC_is_active()) {
 331     action->execution.surf_exec =
 332       surf_workstation_model->extension.workstation.execute(host->host,
 333     computation_amount);
 334     surf_workstation_model->action_data_set(action->execution.surf_exec, action);
 335     surf_workstation_model->set_priority(action->execution.surf_exec, priority);
 336   }
 337
 338   XBT_DEBUG("Create execute action %p", action);
 339
 340   return action;
 341 }
 342
 343 smx_action_t SIMIX_pre_host_parallel_execute(smx_simcall_t simcall, const char *name,
 344     int host_nb, smx_host_t *host_list,
 345     double *computation_amount, double *communication_amount,
 346     double amount, double rate){
 347   return SIMIX_host_parallel_execute(name, host_nb, host_list, computation_amount,
 348                                      communication_amount, amount, rate);
 349 }
 350 smx_action_t SIMIX_host_parallel_execute(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
 355   void **workstation_list = NULL;
 356   int i;
 357
 358   /* alloc structures and initialize */
 359   smx_action_t action = xbt_mallocator_get(simix_global->action_mallocator);
 360   action->type = SIMIX_ACTION_PARALLEL_EXECUTE;
 361   action->name = xbt_strdup(name);
 362   action->state = SIMIX_RUNNING;
 363   action->execution.host = NULL; /* FIXME: do we need the list of hosts? */
 364
 365 #ifdef HAVE_TRACING
 366   action->category = NULL;
 367 #endif
 368
 369   /* set surf's action */
 370   workstation_list = xbt_new0(void *, host_nb);
 371   for (i = 0; i < host_nb; i++)
 372     workstation_list[i] = host_list[i]->host;
 373
 374   /* set surf's action */
 375   if (!MC_is_active()) {
 376     action->execution.surf_exec =
 377       surf_workstation_model->extension.workstation.
 378       execute_parallel_task(host_nb, workstation_list, computation_amount,
 379                       communication_amount, rate);
 380
 381     surf_workstation_model->action_data_set(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_workstation_model->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_workstation_model->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_workstation_model->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_workstation_model->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_workstation_model->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_workstation_model->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", simcall->issuer->smx_host->name);
 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     if (surf_workstation_model->extension.
 503         workstation.get_state(simcall->issuer->smx_host->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_workstation_model->extension.workstation.get_state(action->execution.host->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_workstation_model->action_state_get(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_workstation_model->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_workstation_model->set_category(action->execution.surf_exec, category);
 554   }else if (action->type == SIMIX_ACTION_COMMUNICATE){
 555     surf_workstation_model->set_category(action->comm.surf_comm, category);
 556   }
 557 }
 558 #endif
 559