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
 170 int SIMIX_host_get_core(sg_host_t host){
 171   return surf_host_get_core(host);
 172 }
 173
 174 xbt_swag_t SIMIX_host_get_process_list(sg_host_t host){
 175   smx_host_priv_t host_priv = sg_host_simix(host);
 176
 177   return host_priv->process_list;
 178 }
 179
 180
 181 double SIMIX_host_get_current_power_peak(sg_host_t host) {
 182           return surf_host_get_current_power_peak(host);
 183 }
 184
 185 double SIMIX_host_get_power_peak_at(sg_host_t host, int pstate_index) {
 186           return surf_host_get_power_peak_at(host, pstate_index);
 187 }
 188
 189 int SIMIX_host_get_nb_pstates(sg_host_t host) {
 190           return surf_host_get_nb_pstates(host);
 191 }
 192
 193
 194 void SIMIX_host_set_pstate(sg_host_t host, int pstate_index) {
 195           surf_host_set_pstate(host, pstate_index);
 196 }
 197 int SIMIX_host_get_pstate(sg_host_t host) {
 198           return surf_host_get_pstate(host);
 199 }
 200
 201 double SIMIX_host_get_consumed_energy(sg_host_t host) {
 202           return surf_host_get_consumed_energy(host);
 203 }
 204 double SIMIX_host_get_wattmin_at(sg_host_t host,int pstate) {
 205           return surf_host_get_wattmin_at(host,pstate);
 206 }
 207 double SIMIX_host_get_wattmax_at(sg_host_t host,int pstate) {
 208           return surf_host_get_wattmax_at(host,pstate);
 209 }
 210
 211 int SIMIX_host_get_state(sg_host_t host){
 212   return surf_host_get_state(surf_host_resource_priv(host));
 213 }
 214
 215 void _SIMIX_host_free_process_arg(void *data)
 216 {
 217   smx_process_arg_t arg = *(void**)data;
 218   int i;
 219   for (i = 0; i < arg->argc; i++)
 220     xbt_free(arg->argv[i]);
 221   xbt_free(arg->argv);
 222   xbt_free(arg->name);
 223   xbt_free(arg);
 224 }
 225 /**
 226  * \brief Add a process to the list of the processes that the host will restart when it comes back
 227  * This function add a process to the list of the processes that will be restarted when the host comes
 228  * back. It is expected that this function is called when the host is down.
 229  * The processes will only be restarted once, meaning that you will have to register the process
 230  * again to restart the process again.
 231  */
 232 void SIMIX_host_add_auto_restart_process(sg_host_t host,
 233                                          const char *name,
 234                                          xbt_main_func_t code,
 235                                          void *data,
 236                                          const char *hostname,
 237                                          double kill_time,
 238                                          int argc, char **argv,
 239                                          xbt_dict_t properties,
 240                                          int auto_restart)
 241 {
 242   if (!sg_host_simix(host)->auto_restart_processes) {
 243     sg_host_simix(host)->auto_restart_processes = xbt_dynar_new(sizeof(smx_process_arg_t),_SIMIX_host_free_process_arg);
 244   }
 245   smx_process_arg_t arg = xbt_new(s_smx_process_arg_t,1);
 246   arg->name = xbt_strdup(name);
 247   arg->code = code;
 248   arg->data = data;
 249   arg->hostname = hostname;
 250   arg->kill_time = kill_time;
 251   arg->argc = argc;
 252
 253   arg->argv = xbt_new(char*,argc + 1);
 254
 255   int i;
 256   for (i = 0; i < argc; i++) {
 257     arg->argv[i] = xbt_strdup(argv[i]);
 258   }
 259   arg->argv[argc] = NULL;
 260
 261   arg->properties = properties;
 262   arg->auto_restart = auto_restart;
 263
 264   if( SIMIX_host_get_state(host) == SURF_RESOURCE_OFF
 265       && !xbt_dict_get_or_null(watched_hosts_lib,sg_host_name(host))){
 266     xbt_dict_set(watched_hosts_lib,sg_host_name(host),host,NULL);
 267     XBT_DEBUG("Have pushed host %s to watched_hosts_lib because state == SURF_RESOURCE_OFF",sg_host_name(host));
 268   }
 269   xbt_dynar_push_as(sg_host_simix(host)->auto_restart_processes,smx_process_arg_t,arg);
 270 }
 271 /**
 272  * \brief Restart the list of processes that have been registered to the host
 273  */
 274 void SIMIX_host_restart_processes(sg_host_t host)
 275 {
 276   unsigned int cpt;
 277   smx_process_arg_t arg;
 278   xbt_dynar_t process_list = sg_host_simix(host)->auto_restart_processes;
 279   if (!process_list)
 280     return;
 281
 282   xbt_dynar_foreach (process_list, cpt, arg) {
 283
 284     XBT_DEBUG("Restarting Process %s(%s) right now", arg->argv[0], arg->hostname);
 285     if (simix_global->create_process_function) {
 286       simix_global->create_process_function(arg->argv[0],
 287                                             arg->code,
 288                                             NULL,
 289                                             arg->hostname,
 290                                             arg->kill_time,
 291                                             arg->argc,
 292                                             arg->argv,
 293                                             arg->properties,
 294                                             arg->auto_restart,
 295                                             NULL);
 296     } else {
 297       simcall_process_create(arg->argv[0],
 298                              arg->code,
 299                              NULL,
 300                              arg->hostname,
 301                              arg->kill_time,
 302                              arg->argc,
 303                              arg->argv,
 304                              arg->properties,
 305                              arg->auto_restart);
 306
 307     }
 308     /* arg->argv is used by the process created above.  Hide it to
 309      * _SIMIX_host_free_process_arg() which is called by xbt_dynar_reset()
 310      * below. */
 311     arg->argc = 0;
 312     arg->argv = NULL;
 313   }
 314   xbt_dynar_reset(process_list);
 315 }
 316
 317 void SIMIX_host_autorestart(sg_host_t host)
 318 {
 319   if(simix_global->autorestart)
 320     simix_global->autorestart(host);
 321   else
 322     xbt_die("No function for simix_global->autorestart");
 323 }
 324 smx_synchro_t simcall_HANDLER_process_execute(smx_simcall_t simcall,
 325                 const char* name, double flops_amount, double priority, double bound, unsigned long affinity_mask) {
 326         return SIMIX_process_execute(simcall->issuer, name,flops_amount,priority,bound,affinity_mask);
 327 }
 328 smx_synchro_t SIMIX_process_execute(smx_process_t issuer, const char *name,
 329      double flops_amount, double priority, double bound, unsigned long affinity_mask){
 330
 331   /* alloc structures and initialize */
 332   smx_synchro_t synchro = xbt_mallocator_get(simix_global->synchro_mallocator);
 333   synchro->type = SIMIX_SYNC_EXECUTE;
 334   synchro->name = xbt_strdup(name);
 335   synchro->state = SIMIX_RUNNING;
 336   synchro->execution.host = issuer->host;
 337   synchro->category = NULL;
 338
 339   /* set surf's action */
 340   if (!MC_is_active() && !MC_record_replay_is_active()) {
 341
 342     synchro->execution.surf_exec = surf_host_execute(issuer->host, flops_amount);
 343     surf_action_set_data(synchro->execution.surf_exec, synchro);
 344     surf_action_set_priority(synchro->execution.surf_exec, priority);
 345
 346     /* Note (hypervisor): for multicore, the bound value being passed to the
 347      * surf layer should not be zero (i.e., unlimited). It should be the
 348      * capacity of a CPU core. */
 349     if (bound == 0)
 350       surf_cpu_action_set_bound(synchro->execution.surf_exec, sg_host_get_speed(issuer->host));
 351     else
 352       surf_cpu_action_set_bound(synchro->execution.surf_exec, bound);
 353
 354     if (affinity_mask != 0) {
 355       /* just a double check to confirm that this host is the host where this task is running. */
 356       xbt_assert(synchro->execution.host == issuer->host);
 357       surf_cpu_action_set_affinity(synchro->execution.surf_exec, issuer->host, affinity_mask);
 358     }
 359   }
 360
 361   XBT_DEBUG("Create execute synchro %p: %s", synchro, synchro->name);
 362
 363   return synchro;
 364 }
 365
 366 smx_synchro_t SIMIX_process_parallel_execute(const char *name,
 367     int host_nb, sg_host_t *host_list,
 368     double *flops_amount, double *bytes_amount,
 369     double amount, double rate){
 370
 371   sg_host_t*host_list_cpy = NULL;
 372   int i;
 373
 374   /* alloc structures and initialize */
 375   smx_synchro_t synchro = xbt_mallocator_get(simix_global->synchro_mallocator);
 376   synchro->type = SIMIX_SYNC_PARALLEL_EXECUTE;
 377   synchro->name = xbt_strdup(name);
 378   synchro->state = SIMIX_RUNNING;
 379   synchro->execution.host = NULL; /* FIXME: do we need the list of hosts? */
 380   synchro->category = NULL;
 381
 382   /* set surf's synchro */
 383   host_list_cpy = xbt_new0(sg_host_t, host_nb);
 384   for (i = 0; i < host_nb; i++)
 385     host_list_cpy[i] = host_list[i];
 386
 387
 388   /* FIXME: what happens if host_list contains VMs and PMs. If
 389    * execute_parallel_task() does not change the state of the model, we can mix
 390    * them. */
 391   surf_model_t ws_model = surf_resource_model(host_list[0], SURF_HOST_LEVEL);
 392   for (i = 1; i < host_nb; i++) {
 393     surf_model_t ws_model_tmp = surf_resource_model(host_list[i], SURF_HOST_LEVEL);
 394     if (ws_model_tmp != ws_model) {
 395       XBT_CRITICAL("mixing VMs and PMs is not supported");
 396       DIE_IMPOSSIBLE;
 397     }
 398   }
 399
 400   /* set surf's synchro */
 401   if (!MC_is_active() && !MC_record_replay_is_active()) {
 402     synchro->execution.surf_exec =
 403       surf_host_model_execute_parallel_task(surf_host_model,
 404                   host_nb, host_list_cpy, flops_amount, bytes_amount, rate);
 405
 406     surf_action_set_data(synchro->execution.surf_exec, synchro);
 407   }
 408   XBT_DEBUG("Create parallel execute synchro %p", synchro);
 409
 410   return synchro;
 411 }
 412
 413 void SIMIX_process_execution_destroy(smx_synchro_t synchro){
 414   XBT_DEBUG("Destroy synchro %p", synchro);
 415
 416   if (synchro->execution.surf_exec) {
 417     surf_action_unref(synchro->execution.surf_exec);
 418     synchro->execution.surf_exec = NULL;
 419   }
 420   xbt_free(synchro->name);
 421   xbt_mallocator_release(simix_global->synchro_mallocator, synchro);
 422 }
 423
 424 void SIMIX_process_execution_cancel(smx_synchro_t synchro){
 425   XBT_DEBUG("Cancel synchro %p", synchro);
 426
 427   if (synchro->execution.surf_exec)
 428     surf_action_cancel(synchro->execution.surf_exec);
 429 }
 430
 431 double SIMIX_process_execution_get_remains(smx_synchro_t synchro){
 432   double result = 0.0;
 433
 434   if (synchro->state == SIMIX_RUNNING)
 435     result = surf_action_get_remains(synchro->execution.surf_exec);
 436
 437   return result;
 438 }
 439
 440 e_smx_state_t SIMIX_process_execution_get_state(smx_synchro_t synchro){
 441   return synchro->state;
 442 }
 443
 444 void SIMIX_process_execution_set_priority(smx_synchro_t synchro, double priority){
 445
 446   if(synchro->execution.surf_exec)
 447         surf_action_set_priority(synchro->execution.surf_exec, priority);
 448 }
 449
 450 void SIMIX_process_execution_set_bound(smx_synchro_t synchro, double bound){
 451
 452   if(synchro->execution.surf_exec)
 453         surf_cpu_action_set_bound(synchro->execution.surf_exec, bound);
 454 }
 455
 456 void SIMIX_process_execution_set_affinity(smx_synchro_t synchro, sg_host_t host, unsigned long mask){
 457   xbt_assert(synchro->type == SIMIX_SYNC_EXECUTE);
 458
 459   if (synchro->execution.surf_exec) {
 460     /* just a double check to confirm that this host is the host where this task is running. */
 461     xbt_assert(synchro->execution.host == host);
 462     surf_cpu_action_set_affinity(synchro->execution.surf_exec, host, mask);
 463   }
 464 }
 465
 466 void simcall_HANDLER_process_execution_wait(smx_simcall_t simcall, smx_synchro_t synchro){
 467
 468   XBT_DEBUG("Wait for execution of synchro %p, state %d", synchro, (int)synchro->state);
 469
 470   /* Associate this simcall to the synchro */
 471   xbt_fifo_push(synchro->simcalls, simcall);
 472   simcall->issuer->waiting_synchro = synchro;
 473
 474   /* set surf's synchro */
 475   if (MC_is_active() || MC_record_replay_is_active()) {
 476     synchro->state = SIMIX_DONE;
 477     SIMIX_execution_finish(synchro);
 478     return;
 479   }
 480
 481   /* If the synchro is already finished then perform the error handling */
 482   if (synchro->state != SIMIX_RUNNING)
 483     SIMIX_execution_finish(synchro);
 484 }
 485
 486 void SIMIX_host_execution_suspend(smx_synchro_t synchro)
 487 {
 488   if(synchro->execution.surf_exec)
 489     surf_action_suspend(synchro->execution.surf_exec);
 490 }
 491
 492 void SIMIX_host_execution_resume(smx_synchro_t synchro)
 493 {
 494   if(synchro->execution.surf_exec)
 495     surf_action_resume(synchro->execution.surf_exec);
 496 }
 497
 498 void SIMIX_execution_finish(smx_synchro_t synchro)
 499 {
 500   xbt_fifo_item_t item;
 501   smx_simcall_t simcall;
 502
 503   xbt_fifo_foreach(synchro->simcalls, item, simcall, smx_simcall_t) {
 504
 505     switch (synchro->state) {
 506
 507       case SIMIX_DONE:
 508         /* do nothing, synchro done */
 509         XBT_DEBUG("SIMIX_execution_finished: execution successful");
 510         break;
 511
 512       case SIMIX_FAILED:
 513         XBT_DEBUG("SIMIX_execution_finished: host '%s' failed", sg_host_name(simcall->issuer->host));
 514         simcall->issuer->context->iwannadie = 1;
 515         SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
 516         break;
 517
 518       case SIMIX_CANCELED:
 519         XBT_DEBUG("SIMIX_execution_finished: execution canceled");
 520         SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Canceled");
 521         break;
 522
 523       default:
 524         xbt_die("Internal error in SIMIX_execution_finish: unexpected synchro state %d",
 525             (int)synchro->state);
 526     }
 527     /* check if the host is down */
 528     if (surf_host_get_state(surf_host_resource_priv(simcall->issuer->host)) != SURF_RESOURCE_ON) {
 529       simcall->issuer->context->iwannadie = 1;
 530     }
 531
 532     simcall->issuer->waiting_synchro =    NULL;
 533     simcall_process_execution_wait__set__result(simcall, synchro->state);
 534     SIMIX_simcall_answer(simcall);
 535   }
 536
 537   /* We no longer need it */
 538   SIMIX_process_execution_destroy(synchro);
 539 }
 540
 541
 542 void SIMIX_post_host_execute(smx_synchro_t synchro)
 543 {
 544   if (synchro->type == SIMIX_SYNC_EXECUTE && /* FIMXE: handle resource failure
 545                                                * for parallel tasks too */
 546       surf_host_get_state(surf_host_resource_priv(synchro->execution.host)) == SURF_RESOURCE_OFF) {
 547     /* If the host running the synchro failed, notice it so that the asking
 548      * process can be killed if it runs on that host itself */
 549     synchro->state = SIMIX_FAILED;
 550   } else if (surf_action_get_state(synchro->execution.surf_exec) == SURF_ACTION_FAILED) {
 551     /* If the host running the synchro didn't fail, then the synchro was
 552      * canceled */
 553     synchro->state = SIMIX_CANCELED;
 554   } else {
 555     synchro->state = SIMIX_DONE;
 556   }
 557
 558   if (synchro->execution.surf_exec) {
 559     surf_action_unref(synchro->execution.surf_exec);
 560     synchro->execution.surf_exec = NULL;
 561   }
 562
 563   /* If there are simcalls associated with the synchro, then answer them */
 564   if (xbt_fifo_size(synchro->simcalls)) {
 565     SIMIX_execution_finish(synchro);
 566   }
 567 }
 568
 569
 570 void SIMIX_set_category(smx_synchro_t synchro, const char *category)
 571 {
 572   if (synchro->state != SIMIX_RUNNING) return;
 573   if (synchro->type == SIMIX_SYNC_EXECUTE){
 574     surf_action_set_category(synchro->execution.surf_exec, category);
 575   }else if (synchro->type == SIMIX_SYNC_COMMUNICATE){
 576     surf_action_set_category(synchro->comm.surf_comm, category);
 577   }
 578 }
 579
 580 /**
 581  * \brief Function to get the parameters of the given the SIMIX host.
 582  *
 583  * \param host the host to get_phys_host (a sg_host_t)
 584  * \param param the parameter object space to be overwritten (a ws_params_t)
 585  */
 586 void SIMIX_host_get_params(sg_host_t ind_vm, vm_params_t params)
 587 {
 588   /* jump to ws_get_params(). */
 589   surf_host_get_params(ind_vm, params);
 590 }
 591
 592 void SIMIX_host_set_params(sg_host_t ind_vm, vm_params_t params)
 593 {
 594   /* jump to ws_set_params(). */
 595   surf_host_set_params(ind_vm, params);
 596 }
 597
 598 xbt_dict_t SIMIX_host_get_mounted_storage_list(sg_host_t host){
 599   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 600
 601   return surf_host_get_mounted_storage_list(host);
 602 }
 603
 604 xbt_dynar_t SIMIX_host_get_attached_storage_list(sg_host_t host){
 605   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 606
 607   return surf_host_get_attached_storage_list(host);
 608 }