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 xbt_swag_t SIMIX_pre_host_get_process_list(smx_simcall_t simcall, smx_host_t host){
 162   return SIMIX_host_get_process_list(host);
 163 }
 164
 165 xbt_swag_t SIMIX_host_get_process_list(smx_host_t host){
 166   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 167   smx_host_priv_t host_priv = SIMIX_host_priv(host);
 168
 169   return host_priv->process_list;
 170 }
 171
 172
 173 double SIMIX_pre_host_get_available_speed(smx_simcall_t simcall, smx_host_t host){
 174   return SIMIX_host_get_available_speed(host);
 175 }
 176 double SIMIX_host_get_available_speed(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_available_speed(host);
 181 }
 182
 183 int SIMIX_pre_host_get_state(smx_simcall_t simcall, smx_host_t host){
 184   return SIMIX_host_get_state(host);
 185 }
 186 int SIMIX_host_get_state(smx_host_t host){
 187   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 188
 189   return surf_workstation_model->extension.workstation.
 190       get_state(host);
 191 }
 192
 193 void* SIMIX_pre_host_self_get_data(smx_simcall_t simcall){
 194   return SIMIX_host_self_get_data();
 195 }
 196 void* SIMIX_host_self_get_data(void)
 197 {
 198   smx_host_t self = SIMIX_host_self();
 199   return SIMIX_host_get_data(self);
 200 }
 201
 202 void SIMIX_host_self_set_data(void *data)
 203 {
 204   smx_host_t self = SIMIX_host_self();
 205   SIMIX_host_set_data(self, data);
 206 }
 207
 208 void* SIMIX_pre_host_get_data(smx_simcall_t simcall,smx_host_t host){
 209   return SIMIX_host_get_data(host);
 210 }
 211 void* SIMIX_host_get_data(smx_host_t host){
 212   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 213
 214   return SIMIX_host_priv(host)->data;
 215 }
 216 void _SIMIX_host_free_process_arg(void *);
 217 void _SIMIX_host_free_process_arg(void *data)
 218 {
 219   smx_process_arg_t arg = *(void**)data;
 220   xbt_free(arg->name);
 221   xbt_free(arg);
 222 }
 223 /**
 224  * \brief Add a process to the list of the processes that the host will restart when it comes back
 225  * This function add a process to the list of the processes that will be restarted when the host comes
 226  * back. It is expected that this function is called when the host is down.
 227  * The processes will only be restarted once, meaning that you will have to register the process
 228  * again to restart the process again.
 229  */
 230 void SIMIX_host_add_auto_restart_process(smx_host_t host,
 231                                          const char *name,
 232                                          xbt_main_func_t code,
 233                                          void *data,
 234                                          const char *hostname,
 235                                          double kill_time,
 236                                          int argc, char **argv,
 237                                          xbt_dict_t properties,
 238                                          int auto_restart)
 239 {
 240   if (!SIMIX_host_priv(host)->auto_restart_processes) {
 241     SIMIX_host_priv(host)->auto_restart_processes = xbt_dynar_new(sizeof(smx_process_arg_t),_SIMIX_host_free_process_arg);
 242   }
 243   smx_process_arg_t arg = xbt_new(s_smx_process_arg_t,1);
 244   arg->name = xbt_strdup(name);
 245   arg->code = code;
 246   arg->data = data;
 247   arg->hostname = hostname;
 248   arg->kill_time = kill_time;
 249   arg->argc = argc;
 250
 251   arg->argv = xbt_new(char*,argc + 1);
 252
 253   int i;
 254   for (i = 0; i < argc; i++) {
 255     arg->argv[i] = xbt_strdup(argv[i]);
 256   }
 257   arg->argv[argc] = NULL;
 258
 259   arg->properties = properties;
 260   arg->auto_restart = auto_restart;
 261
 262   if( SIMIX_host_get_state(host) == SURF_RESOURCE_OFF
 263       && !xbt_dict_get_or_null(watched_hosts_lib,sg_host_name(host))){
 264     xbt_dict_set(watched_hosts_lib,sg_host_name(host),host,NULL);
 265     XBT_DEBUG("Have push host %s to watched_hosts_lib because state == SURF_RESOURCE_OFF",sg_host_name(host));
 266   }
 267   xbt_dynar_push_as(SIMIX_host_priv(host)->auto_restart_processes,smx_process_arg_t,arg);
 268 }
 269 /**
 270  * \brief Restart the list of processes that have been registered to the host
 271  */
 272 void SIMIX_host_restart_processes(smx_host_t host)
 273 {
 274   unsigned int cpt;
 275   smx_process_arg_t arg;
 276   xbt_dynar_foreach(SIMIX_host_priv(host)->auto_restart_processes,cpt,arg) {
 277
 278     smx_process_t process;
 279
 280     XBT_DEBUG("Restarting Process %s(%s) right now", arg->argv[0], arg->hostname);
 281     if (simix_global->create_process_function) {
 282       simix_global->create_process_function(&process,
 283                                             arg->argv[0],
 284                                             arg->code,
 285                                             NULL,
 286                                             arg->hostname,
 287                                             arg->kill_time,
 288                                             arg->argc,
 289                                             arg->argv,
 290                                             arg->properties,
 291                                             arg->auto_restart);
 292     }
 293     else {
 294       simcall_process_create(&process,
 295                                             arg->argv[0],
 296                                             arg->code,
 297                                             NULL,
 298                                             arg->hostname,
 299                                             arg->kill_time,
 300                                             arg->argc,
 301                                             arg->argv,
 302                                             arg->properties,
 303                                             arg->auto_restart);
 304
 305     }
 306   }
 307   xbt_dynar_reset(SIMIX_host_priv(host)->auto_restart_processes);
 308 }
 309
 310 void SIMIX_host_autorestart(smx_host_t host)
 311 {
 312   if(simix_global->autorestart)
 313     simix_global->autorestart(host);
 314   else
 315     xbt_die("No function for simix_global->autorestart");
 316 }
 317
 318 void SIMIX_pre_host_set_data(smx_simcall_t simcall, smx_host_t host, void *data) {
 319   SIMIX_host_set_data(host, data);
 320 }
 321 void SIMIX_host_set_data(smx_host_t host, void *data){
 322   xbt_assert((host != NULL), "Invalid parameters");
 323   xbt_assert((SIMIX_host_priv(host)->data == NULL), "Data already set");
 324
 325   SIMIX_host_priv(host)->data = data;
 326 }
 327
 328 smx_action_t SIMIX_pre_host_execute(smx_simcall_t simcall,const char *name,
 329     smx_host_t host, double computation_amount, double priority){
 330   return SIMIX_host_execute(name, host, computation_amount, priority);
 331 }
 332 smx_action_t SIMIX_host_execute(const char *name,
 333     smx_host_t host, double computation_amount, double priority){
 334
 335   /* alloc structures and initialize */
 336   smx_action_t action = xbt_mallocator_get(simix_global->action_mallocator);
 337   action->type = SIMIX_ACTION_EXECUTE;
 338   action->name = xbt_strdup(name);
 339   action->state = SIMIX_RUNNING;
 340   action->execution.host = host;
 341
 342 #ifdef HAVE_TRACING
 343   action->category = NULL;
 344 #endif
 345
 346   /* set surf's action */
 347   if (!MC_is_active()) {
 348     action->execution.surf_exec =
 349       surf_workstation_model->extension.workstation.execute(host,
 350     computation_amount);
 351     surf_workstation_model->action_data_set(action->execution.surf_exec, action);
 352     surf_workstation_model->set_priority(action->execution.surf_exec, priority);
 353   }
 354
 355   XBT_DEBUG("Create execute action %p", action);
 356
 357   return action;
 358 }
 359
 360 smx_action_t SIMIX_pre_host_parallel_execute(smx_simcall_t simcall, const char *name,
 361     int host_nb, smx_host_t *host_list,
 362     double *computation_amount, double *communication_amount,
 363     double amount, double rate){
 364   return SIMIX_host_parallel_execute(name, host_nb, host_list, computation_amount,
 365                                      communication_amount, amount, rate);
 366 }
 367 smx_action_t SIMIX_host_parallel_execute(const char *name,
 368     int host_nb, smx_host_t *host_list,
 369     double *computation_amount, double *communication_amount,
 370     double amount, double rate){
 371
 372   void **workstation_list = NULL;
 373   int i;
 374
 375   /* alloc structures and initialize */
 376   smx_action_t action = xbt_mallocator_get(simix_global->action_mallocator);
 377   action->type = SIMIX_ACTION_PARALLEL_EXECUTE;
 378   action->name = xbt_strdup(name);
 379   action->state = SIMIX_RUNNING;
 380   action->execution.host = NULL; /* FIXME: do we need the list of hosts? */
 381
 382 #ifdef HAVE_TRACING
 383   action->category = NULL;
 384 #endif
 385
 386   /* set surf's action */
 387   workstation_list = xbt_new0(void *, host_nb);
 388   for (i = 0; i < host_nb; i++)
 389     workstation_list[i] = host_list[i];
 390
 391   /* set surf's action */
 392   if (!MC_is_active()) {
 393     action->execution.surf_exec =
 394       surf_workstation_model->extension.workstation.
 395       execute_parallel_task(host_nb, workstation_list, computation_amount,
 396                       communication_amount, rate);
 397
 398     surf_workstation_model->action_data_set(action->execution.surf_exec, action);
 399   }
 400   XBT_DEBUG("Create parallel execute action %p", action);
 401
 402   return action;
 403 }
 404
 405 void SIMIX_pre_host_execution_destroy(smx_simcall_t simcall, smx_action_t action){
 406   SIMIX_host_execution_destroy(action);
 407 }
 408 void SIMIX_host_execution_destroy(smx_action_t action){
 409   XBT_DEBUG("Destroy action %p", action);
 410
 411   if (action->execution.surf_exec) {
 412     surf_workstation_model->action_unref(action->execution.surf_exec);
 413     action->execution.surf_exec = NULL;
 414   }
 415   xbt_free(action->name);
 416   xbt_mallocator_release(simix_global->action_mallocator, action);
 417 }
 418
 419 void SIMIX_pre_host_execution_cancel(smx_simcall_t simcall, smx_action_t action){
 420   SIMIX_host_execution_cancel(action);
 421 }
 422 void SIMIX_host_execution_cancel(smx_action_t action){
 423   XBT_DEBUG("Cancel action %p", action);
 424
 425   if (action->execution.surf_exec)
 426     surf_workstation_model->action_cancel(action->execution.surf_exec);
 427 }
 428
 429 double SIMIX_pre_host_execution_get_remains(smx_simcall_t simcall, smx_action_t action){
 430   return SIMIX_host_execution_get_remains(action);
 431 }
 432 double SIMIX_host_execution_get_remains(smx_action_t action){
 433   double result = 0.0;
 434
 435   if (action->state == SIMIX_RUNNING)
 436     result = surf_workstation_model->get_remains(action->execution.surf_exec);
 437
 438   return result;
 439 }
 440
 441 e_smx_state_t SIMIX_pre_host_execution_get_state(smx_simcall_t simcall, smx_action_t action){
 442   return SIMIX_host_execution_get_state(action);
 443 }
 444 e_smx_state_t SIMIX_host_execution_get_state(smx_action_t action){
 445   return action->state;
 446 }
 447
 448 void SIMIX_pre_host_execution_set_priority(smx_simcall_t simcall, smx_action_t action,
 449                                         double priority){
 450   return SIMIX_host_execution_set_priority(action, priority);
 451 }
 452 void SIMIX_host_execution_set_priority(smx_action_t action, double priority){
 453   if(action->execution.surf_exec)
 454     surf_workstation_model->set_priority(action->execution.surf_exec, priority);
 455 }
 456
 457 void SIMIX_pre_host_execution_wait(smx_simcall_t simcall, smx_action_t action){
 458
 459   XBT_DEBUG("Wait for execution of action %p, state %d", action, (int)action->state);
 460
 461   /* Associate this simcall to the action */
 462   xbt_fifo_push(action->simcalls, simcall);
 463   simcall->issuer->waiting_action = action;
 464
 465   /* set surf's action */
 466   if (MC_is_active()) {
 467     action->state = SIMIX_DONE;
 468     SIMIX_execution_finish(action);
 469     return;
 470   }
 471
 472   /* If the action is already finished then perform the error handling */
 473   if (action->state != SIMIX_RUNNING)
 474     SIMIX_execution_finish(action);
 475 }
 476
 477 void SIMIX_host_execution_suspend(smx_action_t action)
 478 {
 479   if(action->execution.surf_exec)
 480     surf_workstation_model->suspend(action->execution.surf_exec);
 481 }
 482
 483 void SIMIX_host_execution_resume(smx_action_t action)
 484 {
 485   if(action->execution.surf_exec)
 486     surf_workstation_model->resume(action->execution.surf_exec);
 487 }
 488
 489 void SIMIX_execution_finish(smx_action_t action)
 490 {
 491   xbt_fifo_item_t item;
 492   smx_simcall_t simcall;
 493
 494   xbt_fifo_foreach(action->simcalls, item, simcall, smx_simcall_t) {
 495
 496     switch (action->state) {
 497
 498       case SIMIX_DONE:
 499         /* do nothing, action done */
 500   XBT_DEBUG("SIMIX_execution_finished: execution successful");
 501         break;
 502
 503       case SIMIX_FAILED:
 504         XBT_DEBUG("SIMIX_execution_finished: host '%s' failed", sg_host_name(simcall->issuer->smx_host));
 505         simcall->issuer->context->iwannadie = 1;
 506         //SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
 507         break;
 508
 509       case SIMIX_CANCELED:
 510         XBT_DEBUG("SIMIX_execution_finished: execution canceled");
 511         SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Canceled");
 512         break;
 513
 514       default:
 515         xbt_die("Internal error in SIMIX_execution_finish: unexpected action state %d",
 516             (int)action->state);
 517     }
 518     /* check if the host is down */
 519     if (surf_workstation_model->extension.
 520         workstation.get_state(simcall->issuer->smx_host) != SURF_RESOURCE_ON) {
 521       simcall->issuer->context->iwannadie = 1;
 522     }
 523
 524     simcall->issuer->waiting_action =    NULL;
 525     simcall_host_execution_wait__set__result(simcall, action->state);
 526     SIMIX_simcall_answer(simcall);
 527   }
 528
 529   /* We no longer need it */
 530   SIMIX_host_execution_destroy(action);
 531 }
 532
 533 void SIMIX_post_host_execute(smx_action_t action)
 534 {
 535   if (action->type == SIMIX_ACTION_EXECUTE && /* FIMXE: handle resource failure
 536                                                * for parallel tasks too */
 537       surf_workstation_model->extension.workstation.get_state(action->execution.host) == SURF_RESOURCE_OFF) {
 538     /* If the host running the action failed, notice it so that the asking
 539      * process can be killed if it runs on that host itself */
 540     action->state = SIMIX_FAILED;
 541   } else if (surf_workstation_model->action_state_get(action->execution.surf_exec) == SURF_ACTION_FAILED) {
 542     /* If the host running the action didn't fail, then the action was
 543      * canceled */
 544     action->state = SIMIX_CANCELED;
 545   } else {
 546     action->state = SIMIX_DONE;
 547   }
 548
 549   if (action->execution.surf_exec) {
 550     surf_workstation_model->action_unref(action->execution.surf_exec);
 551     action->execution.surf_exec = NULL;
 552   }
 553
 554   /* If there are simcalls associated with the action, then answer them */
 555   if (xbt_fifo_size(action->simcalls)) {
 556     SIMIX_execution_finish(action);
 557   }
 558 }
 559
 560
 561 #ifdef HAVE_TRACING
 562 void SIMIX_pre_set_category(smx_simcall_t simcall, smx_action_t action,
 563                             const char *category){
 564   SIMIX_set_category(action, category);
 565 }
 566 void SIMIX_set_category(smx_action_t action, const char *category)
 567 {
 568   if (action->state != SIMIX_RUNNING) return;
 569   if (action->type == SIMIX_ACTION_EXECUTE){
 570     surf_workstation_model->set_category(action->execution.surf_exec, category);
 571   }else if (action->type == SIMIX_ACTION_COMMUNICATE){
 572     surf_workstation_model->set_category(action->comm.surf_comm, category);
 573   }
 574 }
 575 #endif
 576