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
  98 smx_host_t SIMIX_host_get_by_name(const char *name)
  99 {
 100   xbt_assert(((simix_global != NULL)
 101                && (host_lib != NULL)),
 102               "Environment not set yet");
 103
 104   return xbt_lib_get_or_null(host_lib, name, SIMIX_HOST_LEVEL);
 105 }
 106
 107 smx_host_t SIMIX_host_self(void)
 108 {
 109   smx_process_t process = SIMIX_process_self();
 110   return (process == NULL) ? NULL : SIMIX_process_get_host(process);
 111 }
 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_host_get_name(smx_host_t host)
 125 {
 126   xbt_assert((host != NULL), "Invalid parameters");
 127
 128   return host->name;
 129 }
 130
 131 xbt_dict_t SIMIX_host_get_properties(smx_host_t host)
 132 {
 133   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 134
 135   return surf_workstation_model->extension.workstation.get_properties(host->host);
 136 }
 137
 138 double SIMIX_host_get_speed(smx_host_t host)
 139 {
 140   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 141
 142   return surf_workstation_model->extension.workstation.
 143       get_speed(host->host, 1.0);
 144 }
 145
 146 double SIMIX_host_get_available_speed(smx_host_t host)
 147 {
 148   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 149
 150   return surf_workstation_model->extension.workstation.
 151       get_available_speed(host->host);
 152 }
 153
 154 int SIMIX_host_get_state(smx_host_t host)
 155 {
 156   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 157
 158   return surf_workstation_model->extension.workstation.
 159       get_state(host->host);
 160 }
 161
 162 void* SIMIX_host_self_get_data(void)
 163 {
 164   return SIMIX_host_get_data(SIMIX_host_self());
 165 }
 166
 167 void SIMIX_host_self_set_data(void *data)
 168 {
 169   SIMIX_host_set_data(SIMIX_host_self(), data);
 170 }
 171
 172 void* SIMIX_host_get_data(smx_host_t host)
 173 {
 174   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 175
 176   return host->data;
 177 }
 178 void _SIMIX_host_free_process_arg(void *);
 179 void _SIMIX_host_free_process_arg(void *data)
 180 {
 181   smx_process_arg_t arg = *(void**)data;
 182   xbt_free(arg->name);
 183   xbt_free(arg);
 184 }
 185 /**
 186  * \brief Add a process to the list of the processes that the host will restart when it comes back
 187  * This function add a process to the list of the processes that will be restarted when the host comes
 188  * back. It is expected that this function is called when the host is down.
 189  * The processes will only be restarted once, meaning that you will have to register the process
 190  * again to restart the process again.
 191  */
 192 void SIMIX_host_add_auto_restart_process(smx_host_t host,
 193                                          const char *name,
 194                                          xbt_main_func_t code,
 195                                          void *data,
 196                                          const char *hostname,
 197                                          double kill_time,
 198                                          int argc, char **argv,
 199                                          xbt_dict_t properties,
 200                                          int auto_restart)
 201 {
 202   if (!host->auto_restart_processes) {
 203     host->auto_restart_processes = xbt_dynar_new(sizeof(smx_process_arg_t),_SIMIX_host_free_process_arg);
 204   }
 205   smx_process_arg_t arg = xbt_new(s_smx_process_arg_t,1);
 206   arg->name = xbt_strdup(name);
 207   arg->code = code;
 208   arg->data = data;
 209   arg->hostname = hostname;
 210   arg->kill_time = kill_time;
 211   arg->argc = argc;
 212
 213   arg->argv = xbt_new(char*,argc + 1);
 214
 215   int i;
 216   for (i = 0; i < argc; i++) {
 217     arg->argv[i] = xbt_strdup(argv[i]);
 218   }
 219   arg->argv[argc] = NULL;
 220
 221   arg->properties = properties;
 222   arg->auto_restart = auto_restart;
 223
 224   if( SIMIX_host_get_state(host) == SURF_RESOURCE_OFF
 225       && !xbt_dict_get_or_null(watched_hosts_lib,host->name)){
 226     xbt_dict_set(watched_hosts_lib,host->name,host,NULL);
 227     XBT_DEBUG("Have push host %s to watched_hosts_lib because state == SURF_RESOURCE_OFF",host->name);
 228   }
 229   xbt_dynar_push_as(host->auto_restart_processes,smx_process_arg_t,arg);
 230 }
 231 /**
 232  * \brief Restart the list of processes that have been registered to the host
 233  */
 234 void SIMIX_host_restart_processes(smx_host_t host)
 235 {
 236   unsigned int cpt;
 237   smx_process_arg_t arg;
 238   xbt_dynar_foreach(host->auto_restart_processes,cpt,arg) {
 239
 240     smx_process_t process;
 241
 242     XBT_DEBUG("Restarting Process %s(%s) right now", arg->argv[0], arg->hostname);
 243     if (simix_global->create_process_function) {
 244       simix_global->create_process_function(&process,
 245                                             arg->argv[0],
 246                                             arg->code,
 247                                             NULL,
 248                                             arg->hostname,
 249                                             arg->kill_time,
 250                                             arg->argc,
 251                                             arg->argv,
 252                                             arg->properties,
 253                                             arg->auto_restart);
 254     }
 255     else {
 256       simcall_process_create(&process,
 257                                             arg->argv[0],
 258                                             arg->code,
 259                                             NULL,
 260                                             arg->hostname,
 261                                             arg->kill_time,
 262                                             arg->argc,
 263                                             arg->argv,
 264                                             arg->properties,
 265                                             arg->auto_restart);
 266
 267     }
 268   }
 269   xbt_dynar_reset(host->auto_restart_processes);
 270 }
 271
 272 void SIMIX_host_autorestart(smx_host_t host)
 273 {
 274   if(simix_global->autorestart)
 275     simix_global->autorestart(host);
 276   else
 277     xbt_die("No function for simix_global->autorestart");
 278 }
 279
 280 void SIMIX_host_set_data(smx_host_t host, void *data)
 281 {
 282   xbt_assert((host != NULL), "Invalid parameters");
 283   xbt_assert((host->data == NULL), "Data already set");
 284
 285   host->data = data;
 286 }
 287
 288 smx_action_t SIMIX_host_execute(const char *name, smx_host_t host,
 289                                 double computation_amount,
 290                                 double priority)
 291 {
 292   /* alloc structures and initialize */
 293   smx_action_t action = xbt_mallocator_get(simix_global->action_mallocator);
 294   action->type = SIMIX_ACTION_EXECUTE;
 295   action->name = xbt_strdup(name);
 296   action->state = SIMIX_RUNNING;
 297   action->execution.host = host;
 298
 299 #ifdef HAVE_TRACING
 300   action->category = NULL;
 301 #endif
 302
 303   /* set surf's action */
 304   if (!MC_IS_ENABLED) {
 305     action->execution.surf_exec =
 306       surf_workstation_model->extension.workstation.execute(host->host,
 307     computation_amount);
 308     surf_workstation_model->action_data_set(action->execution.surf_exec, action);
 309     surf_workstation_model->set_priority(action->execution.surf_exec, priority);
 310   }
 311
 312   XBT_DEBUG("Create execute action %p", action);
 313
 314   return action;
 315 }
 316
 317 smx_action_t SIMIX_host_parallel_execute( const char *name,
 318     int host_nb, smx_host_t *host_list,
 319     double *computation_amount, double *communication_amount,
 320     double amount, double rate)
 321 {
 322   void **workstation_list = NULL;
 323   int i;
 324
 325   /* alloc structures and initialize */
 326   smx_action_t action = xbt_mallocator_get(simix_global->action_mallocator);
 327   action->type = SIMIX_ACTION_PARALLEL_EXECUTE;
 328   action->name = xbt_strdup(name);
 329   action->state = SIMIX_RUNNING;
 330   action->execution.host = NULL; /* FIXME: do we need the list of hosts? */
 331
 332 #ifdef HAVE_TRACING
 333   action->category = NULL;
 334 #endif
 335
 336   /* set surf's action */
 337   workstation_list = xbt_new0(void *, host_nb);
 338   for (i = 0; i < host_nb; i++)
 339     workstation_list[i] = host_list[i]->host;
 340
 341   /* set surf's action */
 342   if (!MC_IS_ENABLED) {
 343     action->execution.surf_exec =
 344       surf_workstation_model->extension.workstation.
 345       execute_parallel_task(host_nb, workstation_list, computation_amount,
 346                       communication_amount, rate);
 347
 348     surf_workstation_model->action_data_set(action->execution.surf_exec, action);
 349   }
 350   XBT_DEBUG("Create parallel execute action %p", action);
 351
 352   return action;
 353 }
 354
 355 void SIMIX_host_execution_destroy(smx_action_t action)
 356 {
 357   int destroyed=0;
 358   XBT_DEBUG("Destroy action %p", action);
 359
 360
 361   if (action->execution.surf_exec) {
 362     destroyed = surf_workstation_model->action_unref(action->execution.surf_exec);
 363     action->execution.surf_exec = NULL;
 364   }
 365
 366   if (destroyed) {
 367     xbt_free(action->name);
 368     xbt_mallocator_release(simix_global->action_mallocator, action);
 369   }
 370 }
 371
 372 void SIMIX_host_execution_cancel(smx_action_t action)
 373 {
 374   XBT_DEBUG("Cancel action %p", action);
 375
 376   if (action->execution.surf_exec)
 377     surf_workstation_model->action_cancel(action->execution.surf_exec);
 378 }
 379
 380 double SIMIX_host_execution_get_remains(smx_action_t action)
 381 {
 382   double result = 0.0;
 383
 384   if (action->state == SIMIX_RUNNING)
 385     result = surf_workstation_model->get_remains(action->execution.surf_exec);
 386
 387   return result;
 388 }
 389
 390 e_smx_state_t SIMIX_host_execution_get_state(smx_action_t action)
 391 {
 392   return action->state;
 393 }
 394
 395 void SIMIX_host_execution_set_priority(smx_action_t action, double priority)
 396 {
 397   if(action->execution.surf_exec)
 398     surf_workstation_model->set_priority(action->execution.surf_exec, priority);
 399 }
 400
 401 void SIMIX_pre_host_execution_wait(smx_simcall_t simcall)
 402 {
 403   smx_action_t action = simcall->host_execution_wait.execution;
 404
 405   XBT_DEBUG("Wait for execution of action %p, state %d", action, (int)action->state);
 406
 407   /* Associate this simcall to the action */
 408   xbt_fifo_push(action->simcalls, simcall);
 409   simcall->issuer->waiting_action = action;
 410
 411   /* set surf's action */
 412   if (MC_IS_ENABLED) {
 413     action->state = SIMIX_DONE;
 414     SIMIX_execution_finish(action);
 415     return;
 416   }
 417
 418   /* If the action is already finished then perform the error handling */
 419   if (action->state != SIMIX_RUNNING)
 420     SIMIX_execution_finish(action);
 421 }
 422
 423 void SIMIX_host_execution_suspend(smx_action_t action)
 424 {
 425   if(action->execution.surf_exec)
 426     surf_workstation_model->suspend(action->execution.surf_exec);
 427 }
 428
 429 void SIMIX_host_execution_resume(smx_action_t action)
 430 {
 431   if(action->execution.surf_exec)
 432     surf_workstation_model->resume(action->execution.surf_exec);
 433 }
 434
 435 void SIMIX_execution_finish(smx_action_t action)
 436 {
 437   xbt_fifo_item_t item;
 438   smx_simcall_t simcall;
 439
 440   xbt_fifo_foreach(action->simcalls, item, simcall, smx_simcall_t) {
 441
 442     switch (action->state) {
 443
 444       case SIMIX_DONE:
 445         /* do nothing, action done */
 446   XBT_DEBUG("SIMIX_execution_finished: execution successful");
 447         break;
 448
 449       case SIMIX_FAILED:
 450         XBT_DEBUG("SIMIX_execution_finished: host '%s' failed", simcall->issuer->smx_host->name);
 451         simcall->issuer->context->iwannadie = 1;
 452         //SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
 453         break;
 454
 455       case SIMIX_CANCELED:
 456         XBT_DEBUG("SIMIX_execution_finished: execution canceled");
 457         SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Canceled");
 458         break;
 459
 460       default:
 461         xbt_die("Internal error in SIMIX_execution_finish: unexpected action state %d",
 462             (int)action->state);
 463     }
 464     /* check if the host is down */
 465     if (surf_workstation_model->extension.
 466         workstation.get_state(simcall->issuer->smx_host->host) != SURF_RESOURCE_ON) {
 467       simcall->issuer->context->iwannadie = 1;
 468     }
 469
 470     simcall->issuer->waiting_action =    NULL;
 471     simcall->host_execution_wait.result = action->state;
 472     SIMIX_simcall_answer(simcall);
 473   }
 474
 475   /* We no longer need it */
 476   SIMIX_host_execution_destroy(action);
 477 }
 478
 479 void SIMIX_post_host_execute(smx_action_t action)
 480 {
 481   if (action->type == SIMIX_ACTION_EXECUTE && /* FIMXE: handle resource failure
 482                                                * for parallel tasks too */
 483       surf_workstation_model->extension.workstation.get_state(action->execution.host->host) == SURF_RESOURCE_OFF) {
 484     /* If the host running the action failed, notice it so that the asking
 485      * process can be killed if it runs on that host itself */
 486     action->state = SIMIX_FAILED;
 487   } else if (surf_workstation_model->action_state_get(action->execution.surf_exec) == SURF_ACTION_FAILED) {
 488     /* If the host running the action didn't fail, then the action was
 489      * canceled */
 490     action->state = SIMIX_CANCELED;
 491   } else {
 492     action->state = SIMIX_DONE;
 493   }
 494
 495   if (action->execution.surf_exec) {
 496     surf_workstation_model->action_unref(action->execution.surf_exec);
 497     action->execution.surf_exec = NULL;
 498   }
 499
 500   /* If there are simcalls associated with the action, then answer them */
 501   if (xbt_fifo_size(action->simcalls)) {
 502     SIMIX_execution_finish(action);
 503   }
 504 }
 505
 506
 507 #ifdef HAVE_TRACING
 508 void SIMIX_set_category(smx_action_t action, const char *category)
 509 {
 510   if (action->state != SIMIX_RUNNING) return;
 511   if (action->type == SIMIX_ACTION_EXECUTE){
 512     surf_workstation_model->set_category(action->execution.surf_exec, category);
 513   }else if (action->type == SIMIX_ACTION_COMMUNICATE){
 514     surf_workstation_model->set_category(action->comm.surf_comm, category);
 515   }
 516 }
 517 #endif
 518