src/simix/smx_host.c

   1 /* Copyright (c) 2007, 2008, 2009, 2010. 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   smx_process_arg_t arg = *(void**)data;
 183   int i;
 184   xbt_free(arg->name);
 185   for (i = 0; i < arg->argc; i++) {
 186     xbt_free(arg->argv[i]);
 187   }
 188   xbt_free(arg->argv);
 189   xbt_free(arg);
 190 }
 191 void SIMIX_host_add_auto_restart_process(smx_host_t host,
 192                                          const char *name,
 193                                          xbt_main_func_t code,
 194                                          void *data,
 195                                          const char *hostname,
 196                                          double kill_time,
 197                                          int argc, char **argv,
 198                                          xbt_dict_t properties,
 199                                          int auto_restart) {
 200   if (!host->auto_restart_processes) {
 201     host->auto_restart_processes = xbt_dynar_new(sizeof(smx_process_arg_t),_SIMIX_host_free_process_arg);
 202   }
 203   smx_process_arg_t arg = xbt_new(s_smx_process_arg_t,1);
 204
 205   arg->name = xbt_strdup(name);
 206   arg->code = code;
 207   arg->data = data;
 208   arg->hostname = hostname;
 209   arg->kill_time = kill_time;
 210   arg->argc = argc;
 211   arg->argv = xbt_new(char*,argc + 1);
 212
 213   int i;
 214   for (i = 0; i < argc; i++) {
 215     arg->argv[i] = xbt_strdup(argv[i]);
 216   }
 217
 218   arg->properties = properties;
 219   arg->auto_restart = auto_restart;
 220
 221   if( SIMIX_host_get_state(host) == SURF_RESOURCE_OFF
 222       && !xbt_dict_get_or_null(watched_hosts_lib,host->name)){
 223     xbt_dict_set(watched_hosts_lib,host->name,host,NULL);
 224     XBT_DEBUG("Have push host %s to watched_hosts_lib because state == SURF_RESOURCE_OFF",host->name);
 225   }
 226   xbt_dynar_push_as(host->auto_restart_processes,smx_process_arg_t,arg);
 227 }
 228
 229
 230 void SIMIX_host_set_data(smx_host_t host, void *data)
 231 {
 232   xbt_assert((host != NULL), "Invalid parameters");
 233   xbt_assert((host->data == NULL), "Data already set");
 234
 235   host->data = data;
 236 }
 237
 238 smx_action_t SIMIX_host_execute(const char *name, smx_host_t host,
 239                                 double computation_amount,
 240                                 double priority)
 241 {
 242   /* alloc structures and initialize */
 243   smx_action_t action = xbt_mallocator_get(simix_global->action_mallocator);
 244   action->type = SIMIX_ACTION_EXECUTE;
 245   action->name = xbt_strdup(name);
 246   action->state = SIMIX_RUNNING;
 247   action->execution.host = host;
 248
 249 #ifdef HAVE_TRACING
 250   action->category = NULL;
 251 #endif
 252
 253   /* set surf's action */
 254   if (!MC_IS_ENABLED) {
 255     action->execution.surf_exec =
 256       surf_workstation_model->extension.workstation.execute(host->host,
 257     computation_amount);
 258     surf_workstation_model->action_data_set(action->execution.surf_exec, action);
 259     surf_workstation_model->set_priority(action->execution.surf_exec, priority);
 260   }
 261
 262   XBT_DEBUG("Create execute action %p", action);
 263
 264   return action;
 265 }
 266
 267 smx_action_t SIMIX_host_parallel_execute( const char *name,
 268     int host_nb, smx_host_t *host_list,
 269     double *computation_amount, double *communication_amount,
 270     double amount, double rate)
 271 {
 272   void **workstation_list = NULL;
 273   int i;
 274
 275   /* alloc structures and initialize */
 276   smx_action_t action = xbt_mallocator_get(simix_global->action_mallocator);
 277   action->type = SIMIX_ACTION_PARALLEL_EXECUTE;
 278   action->name = xbt_strdup(name);
 279   action->state = SIMIX_RUNNING;
 280   action->execution.host = NULL; /* FIXME: do we need the list of hosts? */
 281
 282 #ifdef HAVE_TRACING
 283   action->category = NULL;
 284 #endif
 285
 286   /* set surf's action */
 287   workstation_list = xbt_new0(void *, host_nb);
 288   for (i = 0; i < host_nb; i++)
 289     workstation_list[i] = host_list[i]->host;
 290
 291   /* set surf's action */
 292   if (!MC_IS_ENABLED) {
 293     action->execution.surf_exec =
 294       surf_workstation_model->extension.workstation.
 295       execute_parallel_task(host_nb, workstation_list, computation_amount,
 296                       communication_amount, rate);
 297
 298     surf_workstation_model->action_data_set(action->execution.surf_exec, action);
 299   }
 300   XBT_DEBUG("Create parallel execute action %p", action);
 301
 302   return action;
 303 }
 304
 305 void SIMIX_host_execution_destroy(smx_action_t action)
 306 {
 307   int destroyed=0;
 308   XBT_DEBUG("Destroy action %p", action);
 309
 310
 311   if (action->execution.surf_exec) {
 312     destroyed = surf_workstation_model->action_unref(action->execution.surf_exec);
 313     action->execution.surf_exec = NULL;
 314   }
 315
 316   if (destroyed) {
 317     xbt_free(action->name);
 318     xbt_mallocator_release(simix_global->action_mallocator, action);
 319   }
 320 }
 321
 322 void SIMIX_host_execution_cancel(smx_action_t action)
 323 {
 324   XBT_DEBUG("Cancel action %p", action);
 325
 326   if (action->execution.surf_exec)
 327     surf_workstation_model->action_cancel(action->execution.surf_exec);
 328 }
 329
 330 double SIMIX_host_execution_get_remains(smx_action_t action)
 331 {
 332   double result = 0.0;
 333
 334   if (action->state == SIMIX_RUNNING)
 335     result = surf_workstation_model->get_remains(action->execution.surf_exec);
 336
 337   return result;
 338 }
 339
 340 e_smx_state_t SIMIX_host_execution_get_state(smx_action_t action)
 341 {
 342   return action->state;
 343 }
 344
 345 void SIMIX_host_execution_set_priority(smx_action_t action, double priority)
 346 {
 347   if(action->execution.surf_exec)
 348     surf_workstation_model->set_priority(action->execution.surf_exec, priority);
 349 }
 350
 351 void SIMIX_pre_host_execution_wait(smx_simcall_t simcall)
 352 {
 353   smx_action_t action = simcall->host_execution_wait.execution;
 354
 355   XBT_DEBUG("Wait for execution of action %p, state %d", action, (int)action->state);
 356
 357   /* Associate this simcall to the action */
 358   xbt_fifo_push(action->simcalls, simcall);
 359   simcall->issuer->waiting_action = action;
 360
 361   /* set surf's action */
 362   if (MC_IS_ENABLED) {
 363     action->state = SIMIX_DONE;
 364     SIMIX_execution_finish(action);
 365     return;
 366   }
 367
 368   /* If the action is already finished then perform the error handling */
 369   if (action->state != SIMIX_RUNNING)
 370     SIMIX_execution_finish(action);
 371 }
 372
 373 void SIMIX_host_execution_suspend(smx_action_t action)
 374 {
 375   if(action->execution.surf_exec)
 376     surf_workstation_model->suspend(action->execution.surf_exec);
 377 }
 378
 379 void SIMIX_host_execution_resume(smx_action_t action)
 380 {
 381   if(action->execution.surf_exec)
 382     surf_workstation_model->resume(action->execution.surf_exec);
 383 }
 384
 385 void SIMIX_execution_finish(smx_action_t action)
 386 {
 387   xbt_fifo_item_t item;
 388   smx_simcall_t simcall;
 389
 390   xbt_fifo_foreach(action->simcalls, item, simcall, smx_simcall_t) {
 391
 392     switch (action->state) {
 393
 394       case SIMIX_DONE:
 395         /* do nothing, action done */
 396   XBT_DEBUG("SIMIX_execution_finished: execution successful");
 397         break;
 398
 399       case SIMIX_FAILED:
 400         XBT_DEBUG("SIMIX_execution_finished: host '%s' failed", simcall->issuer->smx_host->name);
 401         simcall->issuer->context->iwannadie = 1;
 402         //SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
 403         break;
 404
 405       case SIMIX_CANCELED:
 406         XBT_DEBUG("SIMIX_execution_finished: execution canceled");
 407         SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Canceled");
 408         break;
 409
 410       default:
 411         xbt_die("Internal error in SIMIX_execution_finish: unexpected action state %d",
 412             (int)action->state);
 413     }
 414     /* check if the host is down */
 415     if (surf_workstation_model->extension.
 416         workstation.get_state(simcall->issuer->smx_host->host) != SURF_RESOURCE_ON) {
 417       simcall->issuer->context->iwannadie = 1;
 418     }
 419
 420     simcall->issuer->waiting_action =    NULL;
 421     simcall->host_execution_wait.result = action->state;
 422     SIMIX_simcall_answer(simcall);
 423   }
 424
 425   /* We no longer need it */
 426   SIMIX_host_execution_destroy(action);
 427 }
 428
 429 void SIMIX_post_host_execute(smx_action_t action)
 430 {
 431   if (action->type == SIMIX_ACTION_EXECUTE && /* FIMXE: handle resource failure
 432                                                * for parallel tasks too */
 433       surf_workstation_model->extension.workstation.get_state(action->execution.host->host) == SURF_RESOURCE_OFF) {
 434     /* If the host running the action failed, notice it so that the asking
 435      * process can be killed if it runs on that host itself */
 436     action->state = SIMIX_FAILED;
 437   } else if (surf_workstation_model->action_state_get(action->execution.surf_exec) == SURF_ACTION_FAILED) {
 438     /* If the host running the action didn't fail, then the action was
 439      * canceled */
 440     action->state = SIMIX_CANCELED;
 441   } else {
 442     action->state = SIMIX_DONE;
 443   }
 444
 445   if (action->execution.surf_exec) {
 446     surf_workstation_model->action_unref(action->execution.surf_exec);
 447     action->execution.surf_exec = NULL;
 448   }
 449
 450   /* If there are simcalls associated with the action, then answer them */
 451   if (xbt_fifo_size(action->simcalls)) {
 452     SIMIX_execution_finish(action);
 453   }
 454 }
 455
 456
 457 #ifdef HAVE_TRACING
 458 void SIMIX_set_category(smx_action_t action, const char *category)
 459 {
 460   if (action->state != SIMIX_RUNNING) return;
 461   if (action->type == SIMIX_ACTION_EXECUTE){
 462     surf_workstation_model->set_category(action->execution.surf_exec, category);
 463   }else if (action->type == SIMIX_ACTION_COMMUNICATE){
 464     surf_workstation_model->set_category(action->comm.surf_comm, category);
 465   }
 466 }
 467 #endif
 468