src/simix/smx_host.c

   1 /* Copyright (c) 2007-2015. 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 #include "src/mc/mc_replay.h"
  13
  14 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_host, simix,
  15                                 "SIMIX hosts");
  16
  17 static void SIMIX_execution_finish(smx_synchro_t synchro);
  18
  19 /**
  20  * \brief Internal function to create a SIMIX host.
  21  * \param name name of the host to create
  22  */
  23 void SIMIX_host_create(const char *name) // FIXME: braindead prototype. Take sg_host as parameter
  24 {
  25   sg_host_t host = xbt_lib_get_elm_or_null(host_lib, name);
  26   smx_host_priv_t smx_host = xbt_new0(s_smx_host_priv_t, 1);
  27   s_smx_process_t proc;
  28
  29   /* Host structure */
  30   smx_host->process_list =
  31       xbt_swag_new(xbt_swag_offset(proc, host_proc_hookup));
  32
  33   /* Update global variables */
  34   sg_host_simix_set(host, smx_host);
  35 }
  36
  37 /**
  38  * \brief Start the host if it is off
  39  *
  40  */
  41 void SIMIX_host_on(sg_host_t h)
  42 {
  43   smx_host_priv_t host = sg_host_simix(h);
  44
  45   xbt_assert((host != NULL), "Invalid parameters");
  46
  47   if (surf_host_get_state(surf_host_resource_priv(h))==SURF_RESOURCE_OFF) {
  48     surf_host_set_state(surf_host_resource_priv(h), SURF_RESOURCE_ON);
  49
  50     unsigned int cpt;
  51     smx_process_arg_t arg;
  52     xbt_dynar_foreach(host->boot_processes,cpt,arg) {
  53
  54       char** argv = xbt_new(char*, arg->argc);
  55       for (int i=0; i<arg->argc; i++)
  56         argv[i] = xbt_strdup(arg->argv[i]);
  57
  58       XBT_DEBUG("Booting Process %s(%s) right now", arg->argv[0], arg->hostname);
  59       if (simix_global->create_process_function) {
  60         simix_global->create_process_function(argv[0],
  61                                               arg->code,
  62                                               NULL,
  63                                               arg->hostname,
  64                                               arg->kill_time,
  65                                               arg->argc,
  66                                               argv,
  67                                               arg->properties,
  68                                               arg->auto_restart,
  69                                               NULL);
  70       } else {
  71         simcall_process_create(arg->argv[0],
  72                                arg->code,
  73                                NULL,
  74                                arg->hostname,
  75                                arg->kill_time,
  76                                arg->argc,
  77                                argv,
  78                                arg->properties,
  79                                arg->auto_restart);
  80       }
  81     }
  82   }
  83 }
  84
  85 void simcall_HANDLER_host_off(smx_simcall_t simcall, sg_host_t h)
  86 {
  87  SIMIX_host_off(h, simcall->issuer);
  88 }
  89
  90 /**
  91  * \brief Stop the host if it is on
  92  *
  93  */
  94 void SIMIX_host_off(sg_host_t h, smx_process_t issuer)
  95 {
  96   smx_host_priv_t host = sg_host_simix(h);
  97
  98   xbt_assert((host != NULL), "Invalid parameters");
  99
 100   if (surf_host_get_state(surf_host_resource_priv(h))==SURF_RESOURCE_ON) {
 101         surf_host_set_state(surf_host_resource_priv(h), SURF_RESOURCE_OFF);
 102
 103     /* Clean Simulator data */
 104     if (xbt_swag_size(host->process_list) != 0) {
 105       smx_process_t process = NULL;
 106       xbt_swag_foreach(process, host->process_list) {
 107         SIMIX_process_kill(process, issuer);
 108         XBT_DEBUG("Killing %s on %s by %s", process->name,  sg_host_name(process->host), issuer->name);
 109       }
 110     }
 111   }
 112 }
 113
 114 /**
 115  * \brief Internal function to destroy a SIMIX host.
 116  *
 117  * \param h the host to destroy (a sg_host_t)
 118  */
 119 void SIMIX_host_destroy(void *h)
 120 {
 121   smx_host_priv_t host = (smx_host_priv_t) h;
 122
 123   xbt_assert((host != NULL), "Invalid parameters");
 124
 125   /* Clean Simulator data */
 126   if (xbt_swag_size(host->process_list) != 0) {
 127     char *msg = xbt_strdup("Shutting down host, but it's not empty:");
 128     char *tmp;
 129     smx_process_t process = NULL;
 130
 131     xbt_swag_foreach(process, host->process_list) {
 132       tmp = bprintf("%s\n\t%s", msg, process->name);
 133       free(msg);
 134       msg = tmp;
 135     }
 136     SIMIX_display_process_status();
 137     THROWF(arg_error, 0, "%s", msg);
 138   }
 139   xbt_dynar_free(&host->auto_restart_processes);
 140   xbt_dynar_free(&host->boot_processes);
 141   xbt_swag_free(host->process_list);
 142
 143   /* Clean host structure */
 144   free(host);
 145   return;
 146 }
 147
 148 sg_host_t SIMIX_host_self(void)
 149 {
 150   smx_process_t process = SIMIX_process_self();
 151   return (process == NULL) ? NULL : SIMIX_process_get_host(process);
 152 }
 153
 154 /* needs to be public and without simcall because it is called
 155    by exceptions and logging events */
 156 const char* SIMIX_host_self_get_name(void)
 157 {
 158   sg_host_t host = SIMIX_host_self();
 159   if (host == NULL || SIMIX_process_self() == simix_global->maestro_process)
 160     return "";
 161
 162   return SIMIX_host_get_name(host);
 163 }
 164
 165 xbt_dict_t SIMIX_host_get_properties(sg_host_t host){
 166   return surf_host_get_properties(surf_host_resource_priv(host));
 167 }
 168
 169 double SIMIX_host_get_speed(sg_host_t host){
 170   return surf_host_get_speed(host, 1.0);
 171 }
 172
 173 int SIMIX_host_get_core(sg_host_t host){
 174   return surf_host_get_core(host);
 175 }
 176
 177 xbt_swag_t SIMIX_host_get_process_list(sg_host_t host){
 178   smx_host_priv_t host_priv = sg_host_simix(host);
 179
 180   return host_priv->process_list;
 181 }
 182
 183
 184 double SIMIX_host_get_available_speed(sg_host_t host){
 185   return surf_host_get_available_speed(host);
 186 }
 187
 188 double SIMIX_host_get_current_power_peak(sg_host_t host) {
 189           return surf_host_get_current_power_peak(host);
 190 }
 191
 192 double SIMIX_host_get_power_peak_at(sg_host_t host, int pstate_index) {
 193           return surf_host_get_power_peak_at(host, pstate_index);
 194 }
 195
 196 int SIMIX_host_get_nb_pstates(sg_host_t host) {
 197           return surf_host_get_nb_pstates(host);
 198 }
 199
 200
 201 void SIMIX_host_set_pstate(sg_host_t host, int pstate_index) {
 202           surf_host_set_pstate(host, pstate_index);
 203 }
 204 int SIMIX_host_get_pstate(sg_host_t host) {
 205           return surf_host_get_pstate(host);
 206 }
 207
 208 double SIMIX_host_get_consumed_energy(sg_host_t host) {
 209           return surf_host_get_consumed_energy(host);
 210 }
 211 double SIMIX_host_get_wattmin_at(sg_host_t host,int pstate) {
 212           return surf_host_get_wattmin_at(host,pstate);
 213 }
 214 double SIMIX_host_get_wattmax_at(sg_host_t host,int pstate) {
 215           return surf_host_get_wattmax_at(host,pstate);
 216 }
 217
 218 int SIMIX_host_get_state(sg_host_t host){
 219   return surf_host_get_state(surf_host_resource_priv(host));
 220 }
 221
 222 void _SIMIX_host_free_process_arg(void *data)
 223 {
 224   smx_process_arg_t arg = *(void**)data;
 225   int i;
 226   for (i = 0; i < arg->argc; i++)
 227     xbt_free(arg->argv[i]);
 228   xbt_free(arg->argv);
 229   xbt_free(arg->name);
 230   xbt_free(arg);
 231 }
 232 /**
 233  * \brief Add a process to the list of the processes that the host will restart when it comes back
 234  * This function add a process to the list of the processes that will be restarted when the host comes
 235  * back. It is expected that this function is called when the host is down.
 236  * The processes will only be restarted once, meaning that you will have to register the process
 237  * again to restart the process again.
 238  */
 239 void SIMIX_host_add_auto_restart_process(sg_host_t host,
 240                                          const char *name,
 241                                          xbt_main_func_t code,
 242                                          void *data,
 243                                          const char *hostname,
 244                                          double kill_time,
 245                                          int argc, char **argv,
 246                                          xbt_dict_t properties,
 247                                          int auto_restart)
 248 {
 249   if (!sg_host_simix(host)->auto_restart_processes) {
 250     sg_host_simix(host)->auto_restart_processes = xbt_dynar_new(sizeof(smx_process_arg_t),_SIMIX_host_free_process_arg);
 251   }
 252   smx_process_arg_t arg = xbt_new(s_smx_process_arg_t,1);
 253   arg->name = xbt_strdup(name);
 254   arg->code = code;
 255   arg->data = data;
 256   arg->hostname = hostname;
 257   arg->kill_time = kill_time;
 258   arg->argc = argc;
 259
 260   arg->argv = xbt_new(char*,argc + 1);
 261
 262   int i;
 263   for (i = 0; i < argc; i++) {
 264     arg->argv[i] = xbt_strdup(argv[i]);
 265   }
 266   arg->argv[argc] = NULL;
 267
 268   arg->properties = properties;
 269   arg->auto_restart = auto_restart;
 270
 271   if( SIMIX_host_get_state(host) == SURF_RESOURCE_OFF
 272       && !xbt_dict_get_or_null(watched_hosts_lib,sg_host_name(host))){
 273     xbt_dict_set(watched_hosts_lib,sg_host_name(host),host,NULL);
 274     XBT_DEBUG("Have pushed host %s to watched_hosts_lib because state == SURF_RESOURCE_OFF",sg_host_name(host));
 275   }
 276   xbt_dynar_push_as(sg_host_simix(host)->auto_restart_processes,smx_process_arg_t,arg);
 277 }
 278 /**
 279  * \brief Restart the list of processes that have been registered to the host
 280  */
 281 void SIMIX_host_restart_processes(sg_host_t host)
 282 {
 283   unsigned int cpt;
 284   smx_process_arg_t arg;
 285   xbt_dynar_t process_list = sg_host_simix(host)->auto_restart_processes;
 286   if (!process_list)
 287     return;
 288
 289   xbt_dynar_foreach (process_list, cpt, arg) {
 290
 291     XBT_DEBUG("Restarting Process %s(%s) right now", arg->argv[0], arg->hostname);
 292     if (simix_global->create_process_function) {
 293       simix_global->create_process_function(arg->argv[0],
 294                                             arg->code,
 295                                             NULL,
 296                                             arg->hostname,
 297                                             arg->kill_time,
 298                                             arg->argc,
 299                                             arg->argv,
 300                                             arg->properties,
 301                                             arg->auto_restart,
 302                                             NULL);
 303     } else {
 304       simcall_process_create(arg->argv[0],
 305                              arg->code,
 306                              NULL,
 307                              arg->hostname,
 308                              arg->kill_time,
 309                              arg->argc,
 310                              arg->argv,
 311                              arg->properties,
 312                              arg->auto_restart);
 313
 314     }
 315     /* arg->argv is used by the process created above.  Hide it to
 316      * _SIMIX_host_free_process_arg() which is called by xbt_dynar_reset()
 317      * below. */
 318     arg->argc = 0;
 319     arg->argv = NULL;
 320   }
 321   xbt_dynar_reset(process_list);
 322 }
 323
 324 void SIMIX_host_autorestart(sg_host_t host)
 325 {
 326   if(simix_global->autorestart)
 327     simix_global->autorestart(host);
 328   else
 329     xbt_die("No function for simix_global->autorestart");
 330 }
 331 smx_synchro_t simcall_HANDLER_process_execute(smx_simcall_t simcall,
 332                 const char* name, double flops_amount, double priority, double bound, unsigned long affinity_mask) {
 333         return SIMIX_process_execute(simcall->issuer, name,flops_amount,priority,bound,affinity_mask);
 334 }
 335 smx_synchro_t SIMIX_process_execute(smx_process_t issuer, const char *name,
 336      double flops_amount, double priority, double bound, unsigned long affinity_mask){
 337
 338   /* alloc structures and initialize */
 339   smx_synchro_t synchro = xbt_mallocator_get(simix_global->synchro_mallocator);
 340   synchro->type = SIMIX_SYNC_EXECUTE;
 341   synchro->name = xbt_strdup(name);
 342   synchro->state = SIMIX_RUNNING;
 343   synchro->execution.host = issuer->host;
 344   synchro->category = NULL;
 345
 346   /* set surf's action */
 347   if (!MC_is_active() && !MC_record_replay_is_active()) {
 348
 349     synchro->execution.surf_exec = surf_host_execute(issuer->host, flops_amount);
 350     surf_action_set_data(synchro->execution.surf_exec, synchro);
 351     surf_action_set_priority(synchro->execution.surf_exec, priority);
 352
 353     /* Note (hypervisor): for multicore, the bound value being passed to the
 354      * surf layer should not be zero (i.e., unlimited). It should be the
 355      * capacity of a CPU core. */
 356     if (bound == 0)
 357       surf_cpu_action_set_bound(synchro->execution.surf_exec, SIMIX_host_get_speed(issuer->host));
 358     else
 359       surf_cpu_action_set_bound(synchro->execution.surf_exec, bound);
 360
 361     if (affinity_mask != 0) {
 362       /* just a double check to confirm that this host is the host where this task is running. */
 363       xbt_assert(synchro->execution.host == issuer->host);
 364       surf_cpu_action_set_affinity(synchro->execution.surf_exec, issuer->host, affinity_mask);
 365     }
 366   }
 367
 368   XBT_DEBUG("Create execute synchro %p: %s", synchro, synchro->name);
 369
 370   return synchro;
 371 }
 372
 373 smx_synchro_t SIMIX_process_parallel_execute(const char *name,
 374     int host_nb, sg_host_t *host_list,
 375     double *flops_amount, double *bytes_amount,
 376     double amount, double rate){
 377
 378   sg_host_t*host_list_cpy = NULL;
 379   int i;
 380
 381   /* alloc structures and initialize */
 382   smx_synchro_t synchro = xbt_mallocator_get(simix_global->synchro_mallocator);
 383   synchro->type = SIMIX_SYNC_PARALLEL_EXECUTE;
 384   synchro->name = xbt_strdup(name);
 385   synchro->state = SIMIX_RUNNING;
 386   synchro->execution.host = NULL; /* FIXME: do we need the list of hosts? */
 387   synchro->category = NULL;
 388
 389   /* set surf's synchro */
 390   host_list_cpy = xbt_new0(sg_host_t, host_nb);
 391   for (i = 0; i < host_nb; i++)
 392     host_list_cpy[i] = host_list[i];
 393
 394
 395   /* FIXME: what happens if host_list contains VMs and PMs. If
 396    * execute_parallel_task() does not change the state of the model, we can mix
 397    * them. */
 398   surf_model_t ws_model = surf_resource_model(host_list[0], SURF_HOST_LEVEL);
 399   for (i = 1; i < host_nb; i++) {
 400     surf_model_t ws_model_tmp = surf_resource_model(host_list[i], SURF_HOST_LEVEL);
 401     if (ws_model_tmp != ws_model) {
 402       XBT_CRITICAL("mixing VMs and PMs is not supported");
 403       DIE_IMPOSSIBLE;
 404     }
 405   }
 406
 407   /* set surf's synchro */
 408   if (!MC_is_active() && !MC_record_replay_is_active()) {
 409     synchro->execution.surf_exec =
 410       surf_host_model_execute_parallel_task(surf_host_model,
 411                   host_nb, host_list_cpy, flops_amount, bytes_amount, rate);
 412
 413     surf_action_set_data(synchro->execution.surf_exec, synchro);
 414   }
 415   XBT_DEBUG("Create parallel execute synchro %p", synchro);
 416
 417   return synchro;
 418 }
 419
 420 void SIMIX_process_execution_destroy(smx_synchro_t synchro){
 421   XBT_DEBUG("Destroy synchro %p", synchro);
 422
 423   if (synchro->execution.surf_exec) {
 424     surf_action_unref(synchro->execution.surf_exec);
 425     synchro->execution.surf_exec = NULL;
 426   }
 427   xbt_free(synchro->name);
 428   xbt_mallocator_release(simix_global->synchro_mallocator, synchro);
 429 }
 430
 431 void SIMIX_process_execution_cancel(smx_synchro_t synchro){
 432   XBT_DEBUG("Cancel synchro %p", synchro);
 433
 434   if (synchro->execution.surf_exec)
 435     surf_action_cancel(synchro->execution.surf_exec);
 436 }
 437
 438 double SIMIX_process_execution_get_remains(smx_synchro_t synchro){
 439   double result = 0.0;
 440
 441   if (synchro->state == SIMIX_RUNNING)
 442     result = surf_action_get_remains(synchro->execution.surf_exec);
 443
 444   return result;
 445 }
 446
 447 e_smx_state_t SIMIX_process_execution_get_state(smx_synchro_t synchro){
 448   return synchro->state;
 449 }
 450
 451 void SIMIX_process_execution_set_priority(smx_synchro_t synchro, double priority){
 452
 453   if(synchro->execution.surf_exec)
 454         surf_action_set_priority(synchro->execution.surf_exec, priority);
 455 }
 456
 457 void SIMIX_process_execution_set_bound(smx_synchro_t synchro, double bound){
 458
 459   if(synchro->execution.surf_exec)
 460         surf_cpu_action_set_bound(synchro->execution.surf_exec, bound);
 461 }
 462
 463 void SIMIX_process_execution_set_affinity(smx_synchro_t synchro, sg_host_t host, unsigned long mask){
 464   xbt_assert(synchro->type == SIMIX_SYNC_EXECUTE);
 465
 466   if (synchro->execution.surf_exec) {
 467     /* just a double check to confirm that this host is the host where this task is running. */
 468     xbt_assert(synchro->execution.host == host);
 469     surf_cpu_action_set_affinity(synchro->execution.surf_exec, host, mask);
 470   }
 471 }
 472
 473 void simcall_HANDLER_process_execution_wait(smx_simcall_t simcall, smx_synchro_t synchro){
 474
 475   XBT_DEBUG("Wait for execution of synchro %p, state %d", synchro, (int)synchro->state);
 476
 477   /* Associate this simcall to the synchro */
 478   xbt_fifo_push(synchro->simcalls, simcall);
 479   simcall->issuer->waiting_synchro = synchro;
 480
 481   /* set surf's synchro */
 482   if (MC_is_active() || MC_record_replay_is_active()) {
 483     synchro->state = SIMIX_DONE;
 484     SIMIX_execution_finish(synchro);
 485     return;
 486   }
 487
 488   /* If the synchro is already finished then perform the error handling */
 489   if (synchro->state != SIMIX_RUNNING)
 490     SIMIX_execution_finish(synchro);
 491 }
 492
 493 void SIMIX_host_execution_suspend(smx_synchro_t synchro)
 494 {
 495   if(synchro->execution.surf_exec)
 496     surf_action_suspend(synchro->execution.surf_exec);
 497 }
 498
 499 void SIMIX_host_execution_resume(smx_synchro_t synchro)
 500 {
 501   if(synchro->execution.surf_exec)
 502     surf_action_resume(synchro->execution.surf_exec);
 503 }
 504
 505 void SIMIX_execution_finish(smx_synchro_t synchro)
 506 {
 507   xbt_fifo_item_t item;
 508   smx_simcall_t simcall;
 509
 510   xbt_fifo_foreach(synchro->simcalls, item, simcall, smx_simcall_t) {
 511
 512     switch (synchro->state) {
 513
 514       case SIMIX_DONE:
 515         /* do nothing, synchro done */
 516         XBT_DEBUG("SIMIX_execution_finished: execution successful");
 517         break;
 518
 519       case SIMIX_FAILED:
 520         XBT_DEBUG("SIMIX_execution_finished: host '%s' failed", sg_host_name(simcall->issuer->host));
 521         simcall->issuer->context->iwannadie = 1;
 522         SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
 523         break;
 524
 525       case SIMIX_CANCELED:
 526         XBT_DEBUG("SIMIX_execution_finished: execution canceled");
 527         SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Canceled");
 528         break;
 529
 530       default:
 531         xbt_die("Internal error in SIMIX_execution_finish: unexpected synchro state %d",
 532             (int)synchro->state);
 533     }
 534     /* check if the host is down */
 535     if (surf_host_get_state(surf_host_resource_priv(simcall->issuer->host)) != SURF_RESOURCE_ON) {
 536       simcall->issuer->context->iwannadie = 1;
 537     }
 538
 539     simcall->issuer->waiting_synchro =    NULL;
 540     simcall_process_execution_wait__set__result(simcall, synchro->state);
 541     SIMIX_simcall_answer(simcall);
 542   }
 543
 544   /* We no longer need it */
 545   SIMIX_process_execution_destroy(synchro);
 546 }
 547
 548
 549 void SIMIX_post_host_execute(smx_synchro_t synchro)
 550 {
 551   if (synchro->type == SIMIX_SYNC_EXECUTE && /* FIMXE: handle resource failure
 552                                                * for parallel tasks too */
 553       surf_host_get_state(surf_host_resource_priv(synchro->execution.host)) == SURF_RESOURCE_OFF) {
 554     /* If the host running the synchro failed, notice it so that the asking
 555      * process can be killed if it runs on that host itself */
 556     synchro->state = SIMIX_FAILED;
 557   } else if (surf_action_get_state(synchro->execution.surf_exec) == SURF_ACTION_FAILED) {
 558     /* If the host running the synchro didn't fail, then the synchro was
 559      * canceled */
 560     synchro->state = SIMIX_CANCELED;
 561   } else {
 562     synchro->state = SIMIX_DONE;
 563   }
 564
 565   if (synchro->execution.surf_exec) {
 566     surf_action_unref(synchro->execution.surf_exec);
 567     synchro->execution.surf_exec = NULL;
 568   }
 569
 570   /* If there are simcalls associated with the synchro, then answer them */
 571   if (xbt_fifo_size(synchro->simcalls)) {
 572     SIMIX_execution_finish(synchro);
 573   }
 574 }
 575
 576
 577 void SIMIX_set_category(smx_synchro_t synchro, const char *category)
 578 {
 579   if (synchro->state != SIMIX_RUNNING) return;
 580   if (synchro->type == SIMIX_SYNC_EXECUTE){
 581     surf_action_set_category(synchro->execution.surf_exec, category);
 582   }else if (synchro->type == SIMIX_SYNC_COMMUNICATE){
 583     surf_action_set_category(synchro->comm.surf_comm, category);
 584   }
 585 }
 586
 587 /**
 588  * \brief Function to get the parameters of the given the SIMIX host.
 589  *
 590  * \param host the host to get_phys_host (a sg_host_t)
 591  * \param param the parameter object space to be overwritten (a ws_params_t)
 592  */
 593 void SIMIX_host_get_params(sg_host_t ind_vm, vm_params_t params)
 594 {
 595   /* jump to ws_get_params(). */
 596   surf_host_get_params(ind_vm, params);
 597 }
 598
 599 void SIMIX_host_set_params(sg_host_t ind_vm, vm_params_t params)
 600 {
 601   /* jump to ws_set_params(). */
 602   surf_host_set_params(ind_vm, params);
 603 }
 604
 605 xbt_dict_t SIMIX_host_get_mounted_storage_list(sg_host_t host){
 606   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 607
 608   return surf_host_get_mounted_storage_list(host);
 609 }
 610
 611 xbt_dynar_t SIMIX_host_get_attached_storage_list(sg_host_t host){
 612   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 613
 614   return surf_host_get_attached_storage_list(host);
 615 }