src/simix/smx_host.c

   1 /* Copyright (c) 2007-2014. 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 "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       smx_process_t process;
  54
  55       char** argv = xbt_new(char*, arg->argc);
  56       for (int i=0; i<arg->argc; i++)
  57         argv[i] = xbt_strdup(arg->argv[i]);
  58
  59       XBT_DEBUG("Booting Process %s(%s) right now", arg->argv[0], arg->hostname);
  60       if (simix_global->create_process_function) {
  61         simix_global->create_process_function(&process,
  62                                               argv[0],
  63                                               arg->code,
  64                                               NULL,
  65                                               arg->hostname,
  66                                               arg->kill_time,
  67                                               arg->argc,
  68                                               argv,
  69                                               arg->properties,
  70                                               arg->auto_restart,
  71                                               NULL);
  72       } else {
  73         simcall_process_create(&process,
  74                                arg->argv[0],
  75                                arg->code,
  76                                NULL,
  77                                arg->hostname,
  78                                arg->kill_time,
  79                                arg->argc,
  80                                argv,
  81                                arg->properties,
  82                                arg->auto_restart);
  83       }
  84     }
  85   }
  86 }
  87
  88 void simcall_HANDLER_host_off(smx_simcall_t simcall, sg_host_t h)
  89 {
  90  SIMIX_host_off(h, simcall->issuer);
  91 }
  92
  93 /**
  94  * \brief Stop the host if it is on
  95  *
  96  */
  97 void SIMIX_host_off(sg_host_t h, smx_process_t issuer)
  98 {
  99   smx_host_priv_t host = sg_host_simix(h);
 100
 101   xbt_assert((host != NULL), "Invalid parameters");
 102
 103   if (surf_host_get_state(surf_host_resource_priv(h))==SURF_RESOURCE_ON) {
 104         surf_host_set_state(surf_host_resource_priv(h), SURF_RESOURCE_OFF);
 105
 106     /* Clean Simulator data */
 107     if (xbt_swag_size(host->process_list) != 0) {
 108       smx_process_t process = NULL;
 109       xbt_swag_foreach(process, host->process_list) {
 110         SIMIX_process_kill(process, issuer);
 111         XBT_DEBUG("Killing %s on %s by %s", process->name,  sg_host_name(process->host), issuer->name);
 112       }
 113     }
 114   }
 115 }
 116
 117 /**
 118  * \brief Internal function to destroy a SIMIX host.
 119  *
 120  * \param h the host to destroy (a sg_host_t)
 121  */
 122 void SIMIX_host_destroy(void *h)
 123 {
 124   smx_host_priv_t host = (smx_host_priv_t) h;
 125
 126   xbt_assert((host != NULL), "Invalid parameters");
 127
 128   /* Clean Simulator data */
 129   if (xbt_swag_size(host->process_list) != 0) {
 130     char *msg = xbt_strdup("Shutting down host, but it's not empty:");
 131     char *tmp;
 132     smx_process_t process = NULL;
 133
 134     xbt_swag_foreach(process, host->process_list) {
 135       tmp = bprintf("%s\n\t%s", msg, process->name);
 136       free(msg);
 137       msg = tmp;
 138     }
 139     SIMIX_display_process_status();
 140     THROWF(arg_error, 0, "%s", msg);
 141   }
 142   xbt_dynar_free(&host->auto_restart_processes);
 143   xbt_dynar_free(&host->boot_processes);
 144   xbt_swag_free(host->process_list);
 145
 146   /* Clean host structure */
 147   free(host);
 148   return;
 149 }
 150
 151 sg_host_t SIMIX_host_self(void)
 152 {
 153   smx_process_t process = SIMIX_process_self();
 154   return (process == NULL) ? NULL : SIMIX_process_get_host(process);
 155 }
 156
 157 /* needs to be public and without simcall because it is called
 158    by exceptions and logging events */
 159 const char* SIMIX_host_self_get_name(void)
 160 {
 161   sg_host_t host = SIMIX_host_self();
 162   if (host == NULL || SIMIX_process_self() == simix_global->maestro_process)
 163     return "";
 164
 165   return SIMIX_host_get_name(host);
 166 }
 167
 168 xbt_dict_t SIMIX_host_get_properties(sg_host_t host){
 169   return surf_host_get_properties(surf_host_resource_priv(host));
 170 }
 171
 172 double SIMIX_host_get_speed(sg_host_t host){
 173   return surf_host_get_speed(host, 1.0);
 174 }
 175
 176 int SIMIX_host_get_core(sg_host_t host){
 177   return surf_host_get_core(host);
 178 }
 179
 180 xbt_swag_t SIMIX_host_get_process_list(sg_host_t host){
 181   smx_host_priv_t host_priv = sg_host_simix(host);
 182
 183   return host_priv->process_list;
 184 }
 185
 186
 187 double SIMIX_host_get_available_speed(sg_host_t host){
 188   return surf_host_get_available_speed(host);
 189 }
 190
 191 double SIMIX_host_get_current_power_peak(sg_host_t host) {
 192           return surf_host_get_current_power_peak(host);
 193 }
 194
 195 double SIMIX_host_get_power_peak_at(sg_host_t host, int pstate_index) {
 196           return surf_host_get_power_peak_at(host, pstate_index);
 197 }
 198
 199 int SIMIX_host_get_nb_pstates(sg_host_t host) {
 200           return surf_host_get_nb_pstates(host);
 201 }
 202
 203
 204 void SIMIX_host_set_pstate(sg_host_t host, int pstate_index) {
 205           surf_host_set_pstate(host, pstate_index);
 206 }
 207 int SIMIX_host_get_pstate(sg_host_t host) {
 208           return surf_host_get_pstate(host);
 209 }
 210
 211 double SIMIX_host_get_consumed_energy(sg_host_t host) {
 212           return surf_host_get_consumed_energy(host);
 213 }
 214 double SIMIX_host_get_wattmin_at(sg_host_t host,int pstate) {
 215           return surf_host_get_wattmin_at(host,pstate);
 216 }
 217 double SIMIX_host_get_wattmax_at(sg_host_t host,int pstate) {
 218           return surf_host_get_wattmax_at(host,pstate);
 219 }
 220
 221 int SIMIX_host_get_state(sg_host_t host){
 222   return surf_host_get_state(surf_host_resource_priv(host));
 223 }
 224
 225 void _SIMIX_host_free_process_arg(void *data)
 226 {
 227   smx_process_arg_t arg = *(void**)data;
 228   int i;
 229   for (i = 0; i < arg->argc; i++)
 230     xbt_free(arg->argv[i]);
 231   xbt_free(arg->argv);
 232   xbt_free(arg->name);
 233   xbt_free(arg);
 234 }
 235 /**
 236  * \brief Add a process to the list of the processes that the host will restart when it comes back
 237  * This function add a process to the list of the processes that will be restarted when the host comes
 238  * back. It is expected that this function is called when the host is down.
 239  * The processes will only be restarted once, meaning that you will have to register the process
 240  * again to restart the process again.
 241  */
 242 void SIMIX_host_add_auto_restart_process(sg_host_t host,
 243                                          const char *name,
 244                                          xbt_main_func_t code,
 245                                          void *data,
 246                                          const char *hostname,
 247                                          double kill_time,
 248                                          int argc, char **argv,
 249                                          xbt_dict_t properties,
 250                                          int auto_restart)
 251 {
 252   if (!sg_host_simix(host)->auto_restart_processes) {
 253     sg_host_simix(host)->auto_restart_processes = xbt_dynar_new(sizeof(smx_process_arg_t),_SIMIX_host_free_process_arg);
 254   }
 255   smx_process_arg_t arg = xbt_new(s_smx_process_arg_t,1);
 256   arg->name = xbt_strdup(name);
 257   arg->code = code;
 258   arg->data = data;
 259   arg->hostname = hostname;
 260   arg->kill_time = kill_time;
 261   arg->argc = argc;
 262
 263   arg->argv = xbt_new(char*,argc + 1);
 264
 265   int i;
 266   for (i = 0; i < argc; i++) {
 267     arg->argv[i] = xbt_strdup(argv[i]);
 268   }
 269   arg->argv[argc] = NULL;
 270
 271   arg->properties = properties;
 272   arg->auto_restart = auto_restart;
 273
 274   if( SIMIX_host_get_state(host) == SURF_RESOURCE_OFF
 275       && !xbt_dict_get_or_null(watched_hosts_lib,sg_host_name(host))){
 276     xbt_dict_set(watched_hosts_lib,sg_host_name(host),host,NULL);
 277     XBT_DEBUG("Have pushed host %s to watched_hosts_lib because state == SURF_RESOURCE_OFF",sg_host_name(host));
 278   }
 279   xbt_dynar_push_as(sg_host_simix(host)->auto_restart_processes,smx_process_arg_t,arg);
 280 }
 281 /**
 282  * \brief Restart the list of processes that have been registered to the host
 283  */
 284 void SIMIX_host_restart_processes(sg_host_t host)
 285 {
 286   unsigned int cpt;
 287   smx_process_arg_t arg;
 288   xbt_dynar_t process_list = sg_host_simix(host)->auto_restart_processes;
 289   if (!process_list)
 290     return;
 291
 292   xbt_dynar_foreach (process_list, cpt, arg) {
 293
 294     smx_process_t process;
 295
 296     XBT_DEBUG("Restarting Process %s(%s) right now", arg->argv[0], arg->hostname);
 297     if (simix_global->create_process_function) {
 298       simix_global->create_process_function(&process,
 299                                             arg->argv[0],
 300                                             arg->code,
 301                                             NULL,
 302                                             arg->hostname,
 303                                             arg->kill_time,
 304                                             arg->argc,
 305                                             arg->argv,
 306                                             arg->properties,
 307                                             arg->auto_restart,
 308                                             NULL);
 309     } else {
 310       simcall_process_create(&process,
 311                              arg->argv[0],
 312                              arg->code,
 313                              NULL,
 314                              arg->hostname,
 315                              arg->kill_time,
 316                              arg->argc,
 317                              arg->argv,
 318                              arg->properties,
 319                              arg->auto_restart);
 320
 321     }
 322     /* arg->argv is used by the process created above.  Hide it to
 323      * _SIMIX_host_free_process_arg() which is called by xbt_dynar_reset()
 324      * below. */
 325     arg->argc = 0;
 326     arg->argv = NULL;
 327   }
 328   xbt_dynar_reset(process_list);
 329 }
 330
 331 void SIMIX_host_autorestart(sg_host_t host)
 332 {
 333   if(simix_global->autorestart)
 334     simix_global->autorestart(host);
 335   else
 336     xbt_die("No function for simix_global->autorestart");
 337 }
 338
 339 smx_synchro_t SIMIX_host_execute(const char *name,
 340     sg_host_t host, double flops_amount, double priority, double bound, unsigned long affinity_mask){
 341
 342   /* alloc structures and initialize */
 343   smx_synchro_t synchro = xbt_mallocator_get(simix_global->synchro_mallocator);
 344   synchro->type = SIMIX_SYNC_EXECUTE;
 345   synchro->name = xbt_strdup(name);
 346   synchro->state = SIMIX_RUNNING;
 347   synchro->execution.host = host;
 348   synchro->category = NULL;
 349
 350   /* set surf's action */
 351   if (!MC_is_active() && !MC_record_replay_is_active()) {
 352
 353     synchro->execution.surf_exec = surf_host_execute(host, flops_amount);
 354     surf_action_set_data(synchro->execution.surf_exec, synchro);
 355     surf_action_set_priority(synchro->execution.surf_exec, priority);
 356
 357     /* Note (hypervisor): for multicore, the bound value being passed to the
 358      * surf layer should not be zero (i.e., unlimited). It should be the
 359      * capacity of a CPU core. */
 360     if (bound == 0)
 361       surf_cpu_action_set_bound(synchro->execution.surf_exec, SIMIX_host_get_speed(host));
 362     else
 363       surf_cpu_action_set_bound(synchro->execution.surf_exec, bound);
 364
 365     if (affinity_mask != 0) {
 366       /* just a double check to confirm that this host is the host where this task is running. */
 367       xbt_assert(synchro->execution.host == host);
 368       surf_cpu_action_set_affinity(synchro->execution.surf_exec, host, affinity_mask);
 369     }
 370   }
 371
 372   XBT_DEBUG("Create execute synchro %p: %s", synchro, synchro->name);
 373
 374   return synchro;
 375 }
 376
 377 smx_synchro_t SIMIX_host_parallel_execute(const char *name,
 378     int host_nb, sg_host_t *host_list,
 379     double *flops_amount, double *bytes_amount,
 380     double amount, double rate){
 381
 382   void **surf_host_list = NULL;
 383   int i;
 384
 385   /* alloc structures and initialize */
 386   smx_synchro_t synchro = xbt_mallocator_get(simix_global->synchro_mallocator);
 387   synchro->type = SIMIX_SYNC_PARALLEL_EXECUTE;
 388   synchro->name = xbt_strdup(name);
 389   synchro->state = SIMIX_RUNNING;
 390   synchro->execution.host = NULL; /* FIXME: do we need the list of hosts? */
 391   synchro->category = NULL;
 392
 393   /* set surf's synchro */
 394   surf_host_list = xbt_new0(void *, host_nb);
 395   for (i = 0; i < host_nb; i++)
 396     surf_host_list[i] = surf_host_resource_priv(host_list[i]);
 397
 398
 399   /* FIXME: what happens if host_list contains VMs and PMs. If
 400    * execute_parallel_task() does not change the state of the model, we can mix
 401    * them. */
 402   surf_model_t ws_model = surf_resource_model(host_list[0], SURF_HOST_LEVEL);
 403   for (i = 1; i < host_nb; i++) {
 404     surf_model_t ws_model_tmp = surf_resource_model(host_list[i], SURF_HOST_LEVEL);
 405     if (ws_model_tmp != ws_model) {
 406       XBT_CRITICAL("mixing VMs and PMs is not supported");
 407       DIE_IMPOSSIBLE;
 408     }
 409   }
 410
 411   /* set surf's synchro */
 412   if (!MC_is_active() && !MC_record_replay_is_active()) {
 413     synchro->execution.surf_exec =
 414       surf_host_model_execute_parallel_task((surf_host_model_t)surf_host_model,
 415                   host_nb, surf_host_list, flops_amount, bytes_amount, rate);
 416
 417     surf_action_set_data(synchro->execution.surf_exec, synchro);
 418   }
 419   XBT_DEBUG("Create parallel execute synchro %p", synchro);
 420
 421   return synchro;
 422 }
 423
 424 void SIMIX_host_execution_destroy(smx_synchro_t synchro){
 425   XBT_DEBUG("Destroy synchro %p", synchro);
 426
 427   if (synchro->execution.surf_exec) {
 428     surf_action_unref(synchro->execution.surf_exec);
 429     synchro->execution.surf_exec = NULL;
 430   }
 431   xbt_free(synchro->name);
 432   xbt_mallocator_release(simix_global->synchro_mallocator, synchro);
 433 }
 434
 435 void SIMIX_host_execution_cancel(smx_synchro_t synchro){
 436   XBT_DEBUG("Cancel synchro %p", synchro);
 437
 438   if (synchro->execution.surf_exec)
 439     surf_action_cancel(synchro->execution.surf_exec);
 440 }
 441
 442 double SIMIX_host_execution_get_remains(smx_synchro_t synchro){
 443   double result = 0.0;
 444
 445   if (synchro->state == SIMIX_RUNNING)
 446     result = surf_action_get_remains(synchro->execution.surf_exec);
 447
 448   return result;
 449 }
 450
 451 e_smx_state_t SIMIX_host_execution_get_state(smx_synchro_t synchro){
 452   return synchro->state;
 453 }
 454
 455 void SIMIX_host_execution_set_priority(smx_synchro_t synchro, double priority){
 456
 457   if(synchro->execution.surf_exec)
 458         surf_action_set_priority(synchro->execution.surf_exec, priority);
 459 }
 460
 461 void SIMIX_host_execution_set_bound(smx_synchro_t synchro, double bound){
 462
 463   if(synchro->execution.surf_exec)
 464         surf_cpu_action_set_bound(synchro->execution.surf_exec, bound);
 465 }
 466
 467 void SIMIX_host_execution_set_affinity(smx_synchro_t synchro, sg_host_t host, unsigned long mask){
 468   xbt_assert(synchro->type == SIMIX_SYNC_EXECUTE);
 469
 470   if (synchro->execution.surf_exec) {
 471     /* just a double check to confirm that this host is the host where this task is running. */
 472     xbt_assert(synchro->execution.host == host);
 473     surf_cpu_action_set_affinity(synchro->execution.surf_exec, host, mask);
 474   }
 475 }
 476
 477 void simcall_HANDLER_host_execution_wait(smx_simcall_t simcall, smx_synchro_t synchro){
 478
 479   XBT_DEBUG("Wait for execution of synchro %p, state %d", synchro, (int)synchro->state);
 480
 481   /* Associate this simcall to the synchro */
 482   xbt_fifo_push(synchro->simcalls, simcall);
 483   simcall->issuer->waiting_synchro = synchro;
 484
 485   /* set surf's synchro */
 486   if (MC_is_active() || MC_record_replay_is_active()) {
 487     synchro->state = SIMIX_DONE;
 488     SIMIX_execution_finish(synchro);
 489     return;
 490   }
 491
 492   /* If the synchro is already finished then perform the error handling */
 493   if (synchro->state != SIMIX_RUNNING)
 494     SIMIX_execution_finish(synchro);
 495 }
 496
 497 void SIMIX_host_execution_suspend(smx_synchro_t synchro)
 498 {
 499   if(synchro->execution.surf_exec)
 500     surf_action_suspend(synchro->execution.surf_exec);
 501 }
 502
 503 void SIMIX_host_execution_resume(smx_synchro_t synchro)
 504 {
 505   if(synchro->execution.surf_exec)
 506     surf_action_resume(synchro->execution.surf_exec);
 507 }
 508
 509 void SIMIX_execution_finish(smx_synchro_t synchro)
 510 {
 511   xbt_fifo_item_t item;
 512   smx_simcall_t simcall;
 513
 514   xbt_fifo_foreach(synchro->simcalls, item, simcall, smx_simcall_t) {
 515
 516     switch (synchro->state) {
 517
 518       case SIMIX_DONE:
 519         /* do nothing, synchro done */
 520         XBT_DEBUG("SIMIX_execution_finished: execution successful");
 521         break;
 522
 523       case SIMIX_FAILED:
 524         XBT_DEBUG("SIMIX_execution_finished: host '%s' failed", sg_host_name(simcall->issuer->host));
 525         simcall->issuer->context->iwannadie = 1;
 526         SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
 527         break;
 528
 529       case SIMIX_CANCELED:
 530         XBT_DEBUG("SIMIX_execution_finished: execution canceled");
 531         SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Canceled");
 532         break;
 533
 534       default:
 535         xbt_die("Internal error in SIMIX_execution_finish: unexpected synchro state %d",
 536             (int)synchro->state);
 537     }
 538     /* check if the host is down */
 539     if (surf_host_get_state(surf_host_resource_priv(simcall->issuer->host)) != SURF_RESOURCE_ON) {
 540       simcall->issuer->context->iwannadie = 1;
 541     }
 542
 543     simcall->issuer->waiting_synchro =    NULL;
 544     simcall_host_execution_wait__set__result(simcall, synchro->state);
 545     SIMIX_simcall_answer(simcall);
 546   }
 547
 548   /* We no longer need it */
 549   SIMIX_host_execution_destroy(synchro);
 550 }
 551
 552
 553 void SIMIX_post_host_execute(smx_synchro_t synchro)
 554 {
 555   if (synchro->type == SIMIX_SYNC_EXECUTE && /* FIMXE: handle resource failure
 556                                                * for parallel tasks too */
 557       surf_host_get_state(surf_host_resource_priv(synchro->execution.host)) == SURF_RESOURCE_OFF) {
 558     /* If the host running the synchro failed, notice it so that the asking
 559      * process can be killed if it runs on that host itself */
 560     synchro->state = SIMIX_FAILED;
 561   } else if (surf_action_get_state(synchro->execution.surf_exec) == SURF_ACTION_FAILED) {
 562     /* If the host running the synchro didn't fail, then the synchro was
 563      * canceled */
 564     synchro->state = SIMIX_CANCELED;
 565   } else {
 566     synchro->state = SIMIX_DONE;
 567   }
 568
 569   if (synchro->execution.surf_exec) {
 570     surf_action_unref(synchro->execution.surf_exec);
 571     synchro->execution.surf_exec = NULL;
 572   }
 573
 574   /* If there are simcalls associated with the synchro, then answer them */
 575   if (xbt_fifo_size(synchro->simcalls)) {
 576     SIMIX_execution_finish(synchro);
 577   }
 578 }
 579
 580
 581 void SIMIX_set_category(smx_synchro_t synchro, const char *category)
 582 {
 583   if (synchro->state != SIMIX_RUNNING) return;
 584   if (synchro->type == SIMIX_SYNC_EXECUTE){
 585     surf_action_set_category(synchro->execution.surf_exec, category);
 586   }else if (synchro->type == SIMIX_SYNC_COMMUNICATE){
 587     surf_action_set_category(synchro->comm.surf_comm, category);
 588   }
 589 }
 590
 591 /**
 592  * \brief Function to get the parameters of the given the SIMIX host.
 593  *
 594  * \param host the host to get_phys_host (a sg_host_t)
 595  * \param param the parameter object space to be overwritten (a ws_params_t)
 596  */
 597 void SIMIX_host_get_params(sg_host_t ind_vm, ws_params_t params)
 598 {
 599   /* jump to ws_get_params(). */
 600   surf_host_get_params(ind_vm, params);
 601 }
 602
 603 void SIMIX_host_set_params(sg_host_t ind_vm, ws_params_t params)
 604 {
 605   /* jump to ws_set_params(). */
 606   surf_host_set_params(ind_vm, params);
 607 }
 608
 609 xbt_dict_t SIMIX_host_get_mounted_storage_list(sg_host_t host){
 610   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 611
 612   return surf_host_get_mounted_storage_list(host);
 613 }
 614
 615 xbt_dynar_t SIMIX_host_get_attached_storage_list(sg_host_t host){
 616   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 617
 618   return surf_host_get_attached_storage_list(host);
 619 }