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 xbt_dict_t watched_hosts_lib;
  17
  18 static void SIMIX_execution_finish(smx_action_t action);
  19
  20 /**
  21  * \brief Internal function to create a SIMIX host.
  22  * \param name name of the host to create
  23  * \param workstation the SURF workstation to encapsulate
  24  * \param data some user data (may be NULL)
  25  */
  26 smx_host_t SIMIX_host_create(const char *name,
  27                                void *workstation, void *data)
  28 {
  29   smx_host_t smx_host = xbt_new0(s_smx_host_t, 1);
  30   s_smx_process_t proc;
  31
  32   /* Host structure */
  33   smx_host->name = xbt_strdup(name);
  34   smx_host->data = data;
  35   smx_host->host = workstation;
  36   smx_host->process_list =
  37       xbt_swag_new(xbt_swag_offset(proc, host_proc_hookup));
  38
  39   /* Update global variables */
  40   xbt_lib_set(host_lib,smx_host->name,SIMIX_HOST_LEVEL,smx_host);
  41
  42   return smx_host;
  43 }
  44
  45 /**
  46  * \brief Internal function to destroy a SIMIX host.
  47  *
  48  * \param h the host to destroy (a smx_host_t)
  49  */
  50 void SIMIX_host_destroy(void *h)
  51 {
  52   smx_host_t host = (smx_host_t) h;
  53
  54   xbt_assert((host != NULL), "Invalid parameters");
  55
  56   /* Clean Simulator data */
  57   if (xbt_swag_size(host->process_list) != 0) {
  58     char *msg =
  59         bprintf("Shutting down host %s, but it's not empty:", host->name);
  60     char *tmp;
  61     smx_process_t process = NULL;
  62
  63     xbt_swag_foreach(process, host->process_list) {
  64       tmp = bprintf("%s\n\t%s", msg, process->name);
  65       free(msg);
  66       msg = tmp;
  67     }
  68     SIMIX_display_process_status();
  69     THROWF(arg_error, 0, "%s", msg);
  70   }
  71   xbt_dynar_free(&host->auto_restart_processes);
  72   xbt_swag_free(host->process_list);
  73
  74   /* Clean host structure */
  75   free(host->name);
  76   free(host);
  77
  78   return;
  79 }
  80
  81 /**
  82  * \brief Returns a dict of all hosts.
  83  *
  84  * \return List of all hosts (as a #xbt_dict_t)
  85  */
  86 xbt_dict_t SIMIX_host_get_dict(void)
  87 {
  88   xbt_dict_t host_dict = xbt_dict_new_homogeneous(NULL);
  89   xbt_lib_cursor_t cursor = NULL;
  90   char *name = NULL;
  91   void **host = NULL;
  92
  93   xbt_lib_foreach(host_lib, cursor, name, host){
  94     if(host[SIMIX_HOST_LEVEL])
  95             xbt_dict_set(host_dict,name,host[SIMIX_HOST_LEVEL], NULL);
  96   }
  97   return host_dict;
  98 }
  99
 100 smx_host_t SIMIX_host_get_by_name(const char *name)
 101 {
 102   xbt_assert(((simix_global != NULL)
 103                && (host_lib != NULL)),
 104               "Environment not set yet");
 105
 106   return xbt_lib_get_or_null(host_lib, name, SIMIX_HOST_LEVEL);
 107 }
 108
 109 smx_host_t SIMIX_host_self(void)
 110 {
 111   smx_process_t process = SIMIX_process_self();
 112   return (process == NULL) ? NULL : SIMIX_process_get_host(process);
 113 }
 114
 115 /* needs to be public and without simcall because it is called
 116    by exceptions and logging events */
 117 const char* SIMIX_host_self_get_name(void)
 118 {
 119   smx_host_t host = SIMIX_host_self();
 120   if (host == NULL || SIMIX_process_self() == simix_global->maestro_process)
 121     return "";
 122
 123   return SIMIX_host_get_name(host);
 124 }
 125
 126 const char* SIMIX_host_get_name(smx_host_t host)
 127 {
 128   xbt_assert((host != NULL), "Invalid parameters");
 129
 130   return host->name;
 131 }
 132
 133 xbt_dict_t SIMIX_host_get_properties(smx_host_t host)
 134 {
 135   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 136
 137   return surf_workstation_model->extension.workstation.get_properties(host->host);
 138 }
 139
 140 double SIMIX_host_get_speed(smx_host_t host)
 141 {
 142   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 143
 144   return surf_workstation_model->extension.workstation.
 145       get_speed(host->host, 1.0);
 146 }
 147
 148 double SIMIX_host_get_available_speed(smx_host_t host)
 149 {
 150   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 151
 152   return surf_workstation_model->extension.workstation.
 153       get_available_speed(host->host);
 154 }
 155
 156 int SIMIX_host_get_state(smx_host_t host)
 157 {
 158   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 159
 160   return surf_workstation_model->extension.workstation.
 161       get_state(host->host);
 162 }
 163
 164 void* SIMIX_host_self_get_data(void)
 165 {
 166   return SIMIX_host_get_data(SIMIX_host_self());
 167 }
 168
 169 void SIMIX_host_self_set_data(void *data)
 170 {
 171   SIMIX_host_set_data(SIMIX_host_self(), data);
 172 }
 173
 174 void* SIMIX_host_get_data(smx_host_t host)
 175 {
 176   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 177
 178   return host->data;
 179 }
 180 void _SIMIX_host_free_process_arg(void *);
 181 void _SIMIX_host_free_process_arg(void *data)
 182 {
 183   smx_process_arg_t arg = *(void**)data;
 184   int i;
 185   xbt_free(arg->name);
 186   for (i = 0; i < arg->argc; i++) {
 187     xbt_free(arg->argv[i]);
 188   }
 189   xbt_free(arg->argv);
 190   xbt_free(arg);
 191 }
 192 /**
 193  * \brief Add a process to the list of the processes that the host will restart when it comes back
 194  * This function add a process to the list of the processes that will be restarted when the host comes
 195  * back. It is expected that this function is called when the host is down.
 196  * The processes will only be restarted once, meaning that you will have to register the process
 197  * again to restart the process again.
 198  */
 199 void SIMIX_host_add_auto_restart_process(smx_host_t host,
 200                                          const char *name,
 201                                          xbt_main_func_t code,
 202                                          void *data,
 203                                          const char *hostname,
 204                                          double kill_time,
 205                                          int argc, char **argv,
 206                                          xbt_dict_t properties,
 207                                          int auto_restart)
 208 {
 209   if (!host->auto_restart_processes) {
 210     host->auto_restart_processes = xbt_dynar_new(sizeof(smx_process_arg_t),_SIMIX_host_free_process_arg);
 211   }
 212   smx_process_arg_t arg = xbt_new(s_smx_process_arg_t,1);
 213
 214   arg->name = xbt_strdup(name);
 215   arg->code = code;
 216   arg->data = data;
 217   arg->hostname = hostname;
 218   arg->kill_time = kill_time;
 219   arg->argc = argc;
 220   arg->argv = xbt_new(char*,argc + 1);
 221
 222   int i;
 223   for (i = 0; i < argc; i++) {
 224     arg->argv[i] = xbt_strdup(argv[i]);
 225   }
 226
 227   arg->properties = properties;
 228   arg->auto_restart = auto_restart;
 229
 230   if( SIMIX_host_get_state(host) == SURF_RESOURCE_OFF
 231       && !xbt_dict_get_or_null(watched_hosts_lib,host->name)){
 232     xbt_dict_set(watched_hosts_lib,host->name,host,NULL);
 233     XBT_DEBUG("Have push host %s to watched_hosts_lib because state == SURF_RESOURCE_OFF",host->name);
 234   }
 235   xbt_dynar_push_as(host->auto_restart_processes,smx_process_arg_t,arg);
 236 }
 237 /**
 238  * \brief Restart the list of processes that have been registered to the host
 239  */
 240 void SIMIX_host_restart_processes(smx_host_t host)
 241 {
 242   unsigned int cpt;
 243   smx_process_arg_t arg;
 244   xbt_dynar_foreach(host->auto_restart_processes,cpt,arg) {
 245
 246     smx_process_t process;
 247
 248     XBT_DEBUG("Restarting Process %s(%s) right now", arg->argv[0], arg->hostname);
 249     if (simix_global->create_process_function) {
 250       simix_global->create_process_function(&process,
 251                                             arg->argv[0],
 252                                             arg->code,
 253                                             NULL,
 254                                             arg->hostname,
 255                                             arg->kill_time,
 256                                             arg->argc,
 257                                             arg->argv,
 258                                             arg->properties,
 259                                             arg->auto_restart);
 260     }
 261     else {
 262       simcall_process_create(&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     }
 274   }
 275   xbt_dynar_reset(host->auto_restart_processes);
 276 }
 277
 278 void SIMIX_host_set_data(smx_host_t host, void *data)
 279 {
 280   xbt_assert((host != NULL), "Invalid parameters");
 281   xbt_assert((host->data == NULL), "Data already set");
 282
 283   host->data = data;
 284 }
 285
 286 smx_action_t SIMIX_host_execute(const char *name, smx_host_t host,
 287                                 double computation_amount,
 288                                 double priority)
 289 {
 290   /* alloc structures and initialize */
 291   smx_action_t action = xbt_mallocator_get(simix_global->action_mallocator);
 292   action->type = SIMIX_ACTION_EXECUTE;
 293   action->name = xbt_strdup(name);
 294   action->state = SIMIX_RUNNING;
 295   action->execution.host = host;
 296
 297 #ifdef HAVE_TRACING
 298   action->category = NULL;
 299 #endif
 300
 301   /* set surf's action */
 302   if (!MC_IS_ENABLED) {
 303     action->execution.surf_exec =
 304       surf_workstation_model->extension.workstation.execute(host->host,
 305     computation_amount);
 306     surf_workstation_model->action_data_set(action->execution.surf_exec, action);
 307     surf_workstation_model->set_priority(action->execution.surf_exec, priority);
 308   }
 309
 310   XBT_DEBUG("Create execute action %p", action);
 311
 312   return action;
 313 }
 314
 315 smx_action_t SIMIX_host_parallel_execute( const char *name,
 316     int host_nb, smx_host_t *host_list,
 317     double *computation_amount, double *communication_amount,
 318     double amount, double rate)
 319 {
 320   void **workstation_list = NULL;
 321   int i;
 322
 323   /* alloc structures and initialize */
 324   smx_action_t action = xbt_mallocator_get(simix_global->action_mallocator);
 325   action->type = SIMIX_ACTION_PARALLEL_EXECUTE;
 326   action->name = xbt_strdup(name);
 327   action->state = SIMIX_RUNNING;
 328   action->execution.host = NULL; /* FIXME: do we need the list of hosts? */
 329
 330 #ifdef HAVE_TRACING
 331   action->category = NULL;
 332 #endif
 333
 334   /* set surf's action */
 335   workstation_list = xbt_new0(void *, host_nb);
 336   for (i = 0; i < host_nb; i++)
 337     workstation_list[i] = host_list[i]->host;
 338
 339   /* set surf's action */
 340   if (!MC_IS_ENABLED) {
 341     action->execution.surf_exec =
 342       surf_workstation_model->extension.workstation.
 343       execute_parallel_task(host_nb, workstation_list, computation_amount,
 344                       communication_amount, rate);
 345
 346     surf_workstation_model->action_data_set(action->execution.surf_exec, action);
 347   }
 348   XBT_DEBUG("Create parallel execute action %p", action);
 349
 350   return action;
 351 }
 352
 353 void SIMIX_host_execution_destroy(smx_action_t action)
 354 {
 355   int destroyed=0;
 356   XBT_DEBUG("Destroy action %p", action);
 357
 358
 359   if (action->execution.surf_exec) {
 360     destroyed = surf_workstation_model->action_unref(action->execution.surf_exec);
 361     action->execution.surf_exec = NULL;
 362   }
 363
 364   if (destroyed) {
 365     xbt_free(action->name);
 366     xbt_mallocator_release(simix_global->action_mallocator, action);
 367   }
 368 }
 369
 370 void SIMIX_host_execution_cancel(smx_action_t action)
 371 {
 372   XBT_DEBUG("Cancel action %p", action);
 373
 374   if (action->execution.surf_exec)
 375     surf_workstation_model->action_cancel(action->execution.surf_exec);
 376 }
 377
 378 double SIMIX_host_execution_get_remains(smx_action_t action)
 379 {
 380   double result = 0.0;
 381
 382   if (action->state == SIMIX_RUNNING)
 383     result = surf_workstation_model->get_remains(action->execution.surf_exec);
 384
 385   return result;
 386 }
 387
 388 e_smx_state_t SIMIX_host_execution_get_state(smx_action_t action)
 389 {
 390   return action->state;
 391 }
 392
 393 void SIMIX_host_execution_set_priority(smx_action_t action, double priority)
 394 {
 395   if(action->execution.surf_exec)
 396     surf_workstation_model->set_priority(action->execution.surf_exec, priority);
 397 }
 398
 399 void SIMIX_pre_host_execution_wait(smx_simcall_t simcall)
 400 {
 401   smx_action_t action = simcall->host_execution_wait.execution;
 402
 403   XBT_DEBUG("Wait for execution of action %p, state %d", action, (int)action->state);
 404
 405   /* Associate this simcall to the action */
 406   xbt_fifo_push(action->simcalls, simcall);
 407   simcall->issuer->waiting_action = action;
 408
 409   /* set surf's action */
 410   if (MC_IS_ENABLED) {
 411     action->state = SIMIX_DONE;
 412     SIMIX_execution_finish(action);
 413     return;
 414   }
 415
 416   /* If the action is already finished then perform the error handling */
 417   if (action->state != SIMIX_RUNNING)
 418     SIMIX_execution_finish(action);
 419 }
 420
 421 void SIMIX_host_execution_suspend(smx_action_t action)
 422 {
 423   if(action->execution.surf_exec)
 424     surf_workstation_model->suspend(action->execution.surf_exec);
 425 }
 426
 427 void SIMIX_host_execution_resume(smx_action_t action)
 428 {
 429   if(action->execution.surf_exec)
 430     surf_workstation_model->resume(action->execution.surf_exec);
 431 }
 432
 433 void SIMIX_execution_finish(smx_action_t action)
 434 {
 435   xbt_fifo_item_t item;
 436   smx_simcall_t simcall;
 437
 438   xbt_fifo_foreach(action->simcalls, item, simcall, smx_simcall_t) {
 439
 440     switch (action->state) {
 441
 442       case SIMIX_DONE:
 443         /* do nothing, action done */
 444   XBT_DEBUG("SIMIX_execution_finished: execution successful");
 445         break;
 446
 447       case SIMIX_FAILED:
 448         XBT_DEBUG("SIMIX_execution_finished: host '%s' failed", simcall->issuer->smx_host->name);
 449         simcall->issuer->context->iwannadie = 1;
 450         //SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
 451         break;
 452
 453       case SIMIX_CANCELED:
 454         XBT_DEBUG("SIMIX_execution_finished: execution canceled");
 455         SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Canceled");
 456         break;
 457
 458       default:
 459         xbt_die("Internal error in SIMIX_execution_finish: unexpected action state %d",
 460             (int)action->state);
 461     }
 462     /* check if the host is down */
 463     if (surf_workstation_model->extension.
 464         workstation.get_state(simcall->issuer->smx_host->host) != SURF_RESOURCE_ON) {
 465       simcall->issuer->context->iwannadie = 1;
 466     }
 467
 468     simcall->issuer->waiting_action =    NULL;
 469     simcall->host_execution_wait.result = action->state;
 470     SIMIX_simcall_answer(simcall);
 471   }
 472
 473   /* We no longer need it */
 474   SIMIX_host_execution_destroy(action);
 475 }
 476
 477 void SIMIX_post_host_execute(smx_action_t action)
 478 {
 479   if (action->type == SIMIX_ACTION_EXECUTE && /* FIMXE: handle resource failure
 480                                                * for parallel tasks too */
 481       surf_workstation_model->extension.workstation.get_state(action->execution.host->host) == SURF_RESOURCE_OFF) {
 482     /* If the host running the action failed, notice it so that the asking
 483      * process can be killed if it runs on that host itself */
 484     action->state = SIMIX_FAILED;
 485   } else if (surf_workstation_model->action_state_get(action->execution.surf_exec) == SURF_ACTION_FAILED) {
 486     /* If the host running the action didn't fail, then the action was
 487      * canceled */
 488     action->state = SIMIX_CANCELED;
 489   } else {
 490     action->state = SIMIX_DONE;
 491   }
 492
 493   if (action->execution.surf_exec) {
 494     surf_workstation_model->action_unref(action->execution.surf_exec);
 495     action->execution.surf_exec = NULL;
 496   }
 497
 498   /* If there are simcalls associated with the action, then answer them */
 499   if (xbt_fifo_size(action->simcalls)) {
 500     SIMIX_execution_finish(action);
 501   }
 502 }
 503
 504
 505 #ifdef HAVE_TRACING
 506 void SIMIX_set_category(smx_action_t action, const char *category)
 507 {
 508   if (action->state != SIMIX_RUNNING) return;
 509   if (action->type == SIMIX_ACTION_EXECUTE){
 510     surf_workstation_model->set_category(action->execution.surf_exec, category);
 511   }else if (action->type == SIMIX_ACTION_COMMUNICATE){
 512     surf_workstation_model->set_category(action->comm.surf_comm, category);
 513   }
 514 }
 515 #endif
 516