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   surf_model_t ws_model = surf_resource_model(host, SURF_WKS_LEVEL);
 140   return ws_model->extension.workstation.get_properties(host);
 141 }
 142
 143 double SIMIX_pre_host_get_speed(smx_simcall_t simcall, smx_host_t host){
 144   return SIMIX_host_get_speed(host);
 145 }
 146 double SIMIX_host_get_speed(smx_host_t host){
 147   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 148
 149   surf_model_t ws_model = surf_resource_model(host, SURF_WKS_LEVEL);
 150   return ws_model->extension.workstation.get_speed(host, 1.0);
 151 }
 152
 153 double SIMIX_pre_host_get_available_speed(smx_simcall_t simcall, smx_host_t host){
 154   return SIMIX_host_get_available_speed(host);
 155 }
 156 double SIMIX_host_get_available_speed(smx_host_t host){
 157   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 158
 159   surf_model_t ws_model = surf_resource_model(host, SURF_WKS_LEVEL);
 160   return ws_model->extension.workstation.get_available_speed(host);
 161 }
 162
 163 int SIMIX_pre_host_get_state(smx_simcall_t simcall, smx_host_t host){
 164   return SIMIX_host_get_state(host);
 165 }
 166 int SIMIX_host_get_state(smx_host_t host){
 167   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 168
 169   surf_model_t ws_model = surf_resource_model(host, SURF_WKS_LEVEL);
 170   return ws_model->extension.workstation.get_state(host);
 171 }
 172
 173 void* SIMIX_pre_host_self_get_data(smx_simcall_t simcall){
 174   return SIMIX_host_self_get_data();
 175 }
 176 void* SIMIX_host_self_get_data(void)
 177 {
 178   smx_host_t self = SIMIX_host_self();
 179   return SIMIX_host_get_data(self);
 180 }
 181
 182 void SIMIX_host_self_set_data(void *data)
 183 {
 184   smx_host_t self = SIMIX_host_self();
 185   SIMIX_host_set_data(self, data);
 186 }
 187
 188 void* SIMIX_pre_host_get_data(smx_simcall_t simcall,smx_host_t host){
 189   return SIMIX_host_get_data(host);
 190 }
 191 void* SIMIX_host_get_data(smx_host_t host){
 192   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 193
 194   return SIMIX_host_priv(host)->data;
 195 }
 196 void _SIMIX_host_free_process_arg(void *);
 197 void _SIMIX_host_free_process_arg(void *data)
 198 {
 199   smx_process_arg_t arg = *(void**)data;
 200   xbt_free(arg->name);
 201   xbt_free(arg);
 202 }
 203 /**
 204  * \brief Add a process to the list of the processes that the host will restart when it comes back
 205  * This function add a process to the list of the processes that will be restarted when the host comes
 206  * back. It is expected that this function is called when the host is down.
 207  * The processes will only be restarted once, meaning that you will have to register the process
 208  * again to restart the process again.
 209  */
 210 void SIMIX_host_add_auto_restart_process(smx_host_t host,
 211                                          const char *name,
 212                                          xbt_main_func_t code,
 213                                          void *data,
 214                                          const char *hostname,
 215                                          double kill_time,
 216                                          int argc, char **argv,
 217                                          xbt_dict_t properties,
 218                                          int auto_restart)
 219 {
 220   if (!SIMIX_host_priv(host)->auto_restart_processes) {
 221     SIMIX_host_priv(host)->auto_restart_processes = xbt_dynar_new(sizeof(smx_process_arg_t),_SIMIX_host_free_process_arg);
 222   }
 223   smx_process_arg_t arg = xbt_new(s_smx_process_arg_t,1);
 224   arg->name = xbt_strdup(name);
 225   arg->code = code;
 226   arg->data = data;
 227   arg->hostname = hostname;
 228   arg->kill_time = kill_time;
 229   arg->argc = argc;
 230
 231   arg->argv = xbt_new(char*,argc + 1);
 232
 233   int i;
 234   for (i = 0; i < argc; i++) {
 235     arg->argv[i] = xbt_strdup(argv[i]);
 236   }
 237   arg->argv[argc] = NULL;
 238
 239   arg->properties = properties;
 240   arg->auto_restart = auto_restart;
 241
 242   if( SIMIX_host_get_state(host) == SURF_RESOURCE_OFF
 243       && !xbt_dict_get_or_null(watched_hosts_lib,sg_host_name(host))){
 244     xbt_dict_set(watched_hosts_lib,sg_host_name(host),host,NULL);
 245     XBT_DEBUG("Have push host %s to watched_hosts_lib because state == SURF_RESOURCE_OFF",sg_host_name(host));
 246   }
 247   xbt_dynar_push_as(SIMIX_host_priv(host)->auto_restart_processes,smx_process_arg_t,arg);
 248 }
 249 /**
 250  * \brief Restart the list of processes that have been registered to the host
 251  */
 252 void SIMIX_host_restart_processes(smx_host_t host)
 253 {
 254   unsigned int cpt;
 255   smx_process_arg_t arg;
 256   xbt_dynar_foreach(SIMIX_host_priv(host)->auto_restart_processes,cpt,arg) {
 257
 258     smx_process_t process;
 259
 260     XBT_DEBUG("Restarting Process %s(%s) right now", arg->argv[0], arg->hostname);
 261     if (simix_global->create_process_function) {
 262       simix_global->create_process_function(&process,
 263                                             arg->argv[0],
 264                                             arg->code,
 265                                             NULL,
 266                                             arg->hostname,
 267                                             arg->kill_time,
 268                                             arg->argc,
 269                                             arg->argv,
 270                                             arg->properties,
 271                                             arg->auto_restart);
 272     }
 273     else {
 274       simcall_process_create(&process,
 275                                             arg->argv[0],
 276                                             arg->code,
 277                                             NULL,
 278                                             arg->hostname,
 279                                             arg->kill_time,
 280                                             arg->argc,
 281                                             arg->argv,
 282                                             arg->properties,
 283                                             arg->auto_restart);
 284
 285     }
 286   }
 287   xbt_dynar_reset(SIMIX_host_priv(host)->auto_restart_processes);
 288 }
 289
 290 void SIMIX_host_autorestart(smx_host_t host)
 291 {
 292   if(simix_global->autorestart)
 293     simix_global->autorestart(host);
 294   else
 295     xbt_die("No function for simix_global->autorestart");
 296 }
 297
 298 void SIMIX_pre_host_set_data(smx_simcall_t simcall, smx_host_t host, void *data) {
 299   SIMIX_host_set_data(host, data);
 300 }
 301 void SIMIX_host_set_data(smx_host_t host, void *data){
 302   xbt_assert((host != NULL), "Invalid parameters");
 303   xbt_assert((SIMIX_host_priv(host)->data == NULL), "Data already set");
 304
 305   SIMIX_host_priv(host)->data = data;
 306 }
 307
 308 smx_action_t SIMIX_pre_host_execute(smx_simcall_t simcall,const char *name,
 309     smx_host_t host, double computation_amount, double priority){
 310   return SIMIX_host_execute(name, host, computation_amount, priority);
 311 }
 312 smx_action_t SIMIX_host_execute(const char *name,
 313     smx_host_t host, double computation_amount, double priority){
 314
 315   /* alloc structures and initialize */
 316   smx_action_t action = xbt_mallocator_get(simix_global->action_mallocator);
 317   action->type = SIMIX_ACTION_EXECUTE;
 318   action->name = xbt_strdup(name);
 319   action->state = SIMIX_RUNNING;
 320   action->execution.host = host;
 321
 322 #ifdef HAVE_TRACING
 323   action->category = NULL;
 324 #endif
 325
 326   surf_model_t ws_model = surf_resource_model(host, SURF_WKS_LEVEL);
 327   /* set surf's action */
 328   if (!MC_is_active()) {
 329     action->execution.surf_exec = ws_model->extension.workstation.execute(host, computation_amount);
 330     ws_model->action_data_set(action->execution.surf_exec, action);
 331     ws_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
 371   /* FIXME: what happens if host_list contains VMs and PMs. If
 372    * execute_parallel_task() does not change the state of the model, we can mix
 373    * them. */
 374   surf_model_t ws_model = surf_resource_model(host_list[0], SURF_WKS_LEVEL);
 375   for (i = 1; i < host_nb; i++) {
 376     surf_model_t ws_model_tmp = surf_resource_model(host_list[i], SURF_WKS_LEVEL);
 377     if (ws_model_tmp != ws_model) {
 378       XBT_CRITICAL("mixing VMs and PMs is not supported");
 379       DIE_IMPOSSIBLE;
 380     }
 381   }
 382
 383   /* set surf's action */
 384   if (!MC_is_active()) {
 385     action->execution.surf_exec =
 386       ws_model->extension.workstation.
 387       execute_parallel_task(host_nb, workstation_list, computation_amount,
 388                       communication_amount, rate);
 389
 390     ws_model->action_data_set(action->execution.surf_exec, action);
 391   }
 392   XBT_DEBUG("Create parallel execute action %p", action);
 393
 394   return action;
 395 }
 396
 397 static surf_model_t get_ws_model_from_action(smx_action_t action)
 398 {
 399   xbt_assert(action->type == SIMIX_ACTION_EXECUTE);
 400   smx_host_t host = action->execution.host;
 401   surf_model_t model = surf_resource_model(host, SURF_WKS_LEVEL);
 402
 403   xbt_assert((model == surf_workstation_model) || (model == surf_vm_workstation_model));
 404
 405   return model;
 406 }
 407
 408 void SIMIX_pre_host_execution_destroy(smx_simcall_t simcall, smx_action_t action){
 409   SIMIX_host_execution_destroy(action);
 410 }
 411 void SIMIX_host_execution_destroy(smx_action_t action){
 412   XBT_DEBUG("Destroy action %p", action);
 413
 414   surf_model_t ws_model = get_ws_model_from_action(action);
 415
 416   if (action->execution.surf_exec) {
 417     ws_model->action_unref(action->execution.surf_exec);
 418     action->execution.surf_exec = NULL;
 419   }
 420   xbt_free(action->name);
 421   xbt_mallocator_release(simix_global->action_mallocator, action);
 422 }
 423
 424 void SIMIX_pre_host_execution_cancel(smx_simcall_t simcall, smx_action_t action){
 425   SIMIX_host_execution_cancel(action);
 426 }
 427 void SIMIX_host_execution_cancel(smx_action_t action){
 428   XBT_DEBUG("Cancel action %p", action);
 429
 430   surf_model_t ws_model = get_ws_model_from_action(action);
 431
 432   if (action->execution.surf_exec)
 433     ws_model->action_cancel(action->execution.surf_exec);
 434 }
 435
 436 double SIMIX_pre_host_execution_get_remains(smx_simcall_t simcall, smx_action_t action){
 437   return SIMIX_host_execution_get_remains(action);
 438 }
 439 double SIMIX_host_execution_get_remains(smx_action_t action){
 440   double result = 0.0;
 441   surf_model_t ws_model = get_ws_model_from_action(action);
 442
 443   if (action->state == SIMIX_RUNNING)
 444     result = ws_model->get_remains(action->execution.surf_exec);
 445
 446   return result;
 447 }
 448
 449 e_smx_state_t SIMIX_pre_host_execution_get_state(smx_simcall_t simcall, smx_action_t action){
 450   return SIMIX_host_execution_get_state(action);
 451 }
 452 e_smx_state_t SIMIX_host_execution_get_state(smx_action_t action){
 453   return action->state;
 454 }
 455
 456 void SIMIX_pre_host_execution_set_priority(smx_simcall_t simcall, smx_action_t action,
 457                                         double priority){
 458   return SIMIX_host_execution_set_priority(action, priority);
 459 }
 460 void SIMIX_host_execution_set_priority(smx_action_t action, double priority){
 461   surf_model_t ws_model = get_ws_model_from_action(action);
 462
 463   if(action->execution.surf_exec)
 464     ws_model->set_priority(action->execution.surf_exec, priority);
 465 }
 466
 467 void SIMIX_pre_host_execution_wait(smx_simcall_t simcall, smx_action_t action){
 468
 469   XBT_DEBUG("Wait for execution of action %p, state %d", action, (int)action->state);
 470
 471   /* Associate this simcall to the action */
 472   xbt_fifo_push(action->simcalls, simcall);
 473   simcall->issuer->waiting_action = action;
 474
 475   /* set surf's action */
 476   if (MC_is_active()) {
 477     action->state = SIMIX_DONE;
 478     SIMIX_execution_finish(action);
 479     return;
 480   }
 481
 482   /* If the action is already finished then perform the error handling */
 483   if (action->state != SIMIX_RUNNING)
 484     SIMIX_execution_finish(action);
 485 }
 486
 487 void SIMIX_host_execution_suspend(smx_action_t action)
 488 {
 489   surf_model_t ws_model = get_ws_model_from_action(action);
 490
 491   if(action->execution.surf_exec)
 492     ws_model->suspend(action->execution.surf_exec);
 493 }
 494
 495 void SIMIX_host_execution_resume(smx_action_t action)
 496 {
 497   surf_model_t ws_model = get_ws_model_from_action(action);
 498
 499   if(action->execution.surf_exec)
 500     ws_model->resume(action->execution.surf_exec);
 501 }
 502
 503 void SIMIX_execution_finish(smx_action_t action)
 504 {
 505   xbt_fifo_item_t item;
 506   smx_simcall_t simcall;
 507   surf_model_t ws_model = get_ws_model_from_action(action);
 508
 509   xbt_fifo_foreach(action->simcalls, item, simcall, smx_simcall_t) {
 510
 511     switch (action->state) {
 512
 513       case SIMIX_DONE:
 514         /* do nothing, action done */
 515   XBT_DEBUG("SIMIX_execution_finished: execution successful");
 516         break;
 517
 518       case SIMIX_FAILED:
 519         XBT_DEBUG("SIMIX_execution_finished: host '%s' failed", sg_host_name(simcall->issuer->smx_host));
 520         simcall->issuer->context->iwannadie = 1;
 521         //SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
 522         break;
 523
 524       case SIMIX_CANCELED:
 525         XBT_DEBUG("SIMIX_execution_finished: execution canceled");
 526         SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Canceled");
 527         break;
 528
 529       default:
 530         xbt_die("Internal error in SIMIX_execution_finish: unexpected action state %d",
 531             (int)action->state);
 532     }
 533     /* check if the host is down */
 534     if (ws_model->extension.workstation.get_state(simcall->issuer->smx_host) != SURF_RESOURCE_ON) {
 535       simcall->issuer->context->iwannadie = 1;
 536     }
 537
 538     simcall->issuer->waiting_action =    NULL;
 539     simcall_host_execution_wait__set__result(simcall, action->state);
 540     SIMIX_simcall_answer(simcall);
 541   }
 542
 543   /* We no longer need it */
 544   SIMIX_host_execution_destroy(action);
 545 }
 546
 547
 548 void SIMIX_post_host_execute(smx_action_t action)
 549 {
 550   if (action->type == SIMIX_ACTION_EXECUTE && /* FIMXE: handle resource failure
 551                                                * for parallel tasks too */
 552       surf_workstation_model->extension.workstation.get_state(action->execution.host) == SURF_RESOURCE_OFF) {
 553     /* If the host running the action failed, notice it so that the asking
 554      * process can be killed if it runs on that host itself */
 555     action->state = SIMIX_FAILED;
 556   } else if (surf_workstation_model->action_state_get(action->execution.surf_exec) == SURF_ACTION_FAILED) {
 557     /* If the host running the action didn't fail, then the action was
 558      * canceled */
 559     action->state = SIMIX_CANCELED;
 560   } else {
 561     action->state = SIMIX_DONE;
 562   }
 563
 564   if (action->execution.surf_exec) {
 565     surf_workstation_model->action_unref(action->execution.surf_exec);
 566     action->execution.surf_exec = NULL;
 567   }
 568
 569   /* If there are simcalls associated with the action, then answer them */
 570   if (xbt_fifo_size(action->simcalls)) {
 571     SIMIX_execution_finish(action);
 572   }
 573 }
 574
 575
 576 #ifdef HAVE_TRACING
 577 void SIMIX_pre_set_category(smx_simcall_t simcall, smx_action_t action,
 578                             const char *category){
 579   SIMIX_set_category(action, category);
 580 }
 581 void SIMIX_set_category(smx_action_t action, const char *category)
 582 {
 583   surf_model_t ws_model = get_ws_model_from_action(action);
 584
 585   if (action->state != SIMIX_RUNNING) return;
 586   if (action->type == SIMIX_ACTION_EXECUTE){
 587     ws_model->set_category(action->execution.surf_exec, category);
 588   }else if (action->type == SIMIX_ACTION_COMMUNICATE){
 589     ws_model->set_category(action->comm.surf_comm, category);
 590   }
 591 }
 592 #endif
 593