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_get_by_name(const char *name){
 152   return xbt_lib_get_elm_or_null(host_lib, name);
 153 }
 154
 155 sg_host_t SIMIX_host_self(void)
 156 {
 157   smx_process_t process = SIMIX_process_self();
 158   return (process == NULL) ? NULL : SIMIX_process_get_host(process);
 159 }
 160
 161 /* needs to be public and without simcall because it is called
 162    by exceptions and logging events */
 163 const char* SIMIX_host_self_get_name(void)
 164 {
 165   sg_host_t host = SIMIX_host_self();
 166   if (host == NULL || SIMIX_process_self() == simix_global->maestro_process)
 167     return "";
 168
 169   return SIMIX_host_get_name(host);
 170 }
 171
 172 xbt_dict_t SIMIX_host_get_properties(sg_host_t host){
 173   return surf_host_get_properties(surf_host_resource_priv(host));
 174 }
 175
 176 double SIMIX_host_get_speed(sg_host_t host){
 177   return surf_host_get_speed(host, 1.0);
 178 }
 179
 180 int SIMIX_host_get_core(sg_host_t host){
 181   return surf_host_get_core(host);
 182 }
 183
 184 xbt_swag_t SIMIX_host_get_process_list(sg_host_t host){
 185   smx_host_priv_t host_priv = sg_host_simix(host);
 186
 187   return host_priv->process_list;
 188 }
 189
 190
 191 double SIMIX_host_get_available_speed(sg_host_t host){
 192   return surf_host_get_available_speed(host);
 193 }
 194
 195 double SIMIX_host_get_current_power_peak(sg_host_t host) {
 196           return surf_host_get_current_power_peak(host);
 197 }
 198
 199 double SIMIX_host_get_power_peak_at(sg_host_t host, int pstate_index) {
 200           return surf_host_get_power_peak_at(host, pstate_index);
 201 }
 202
 203 int SIMIX_host_get_nb_pstates(sg_host_t host) {
 204           return surf_host_get_nb_pstates(host);
 205 }
 206
 207
 208 void SIMIX_host_set_pstate(sg_host_t host, int pstate_index) {
 209           surf_host_set_pstate(host, pstate_index);
 210 }
 211 int SIMIX_host_get_pstate(sg_host_t host) {
 212           return surf_host_get_pstate(host);
 213 }
 214
 215 double SIMIX_host_get_consumed_energy(sg_host_t host) {
 216           return surf_host_get_consumed_energy(host);
 217 }
 218 double SIMIX_host_get_wattmin_at(sg_host_t host,int pstate) {
 219           return surf_host_get_wattmin_at(host,pstate);
 220 }
 221 double SIMIX_host_get_wattmax_at(sg_host_t host,int pstate) {
 222           return surf_host_get_wattmax_at(host,pstate);
 223 }
 224
 225 int SIMIX_host_get_state(sg_host_t host){
 226   return surf_host_get_state(surf_host_resource_priv(host));
 227 }
 228
 229 void _SIMIX_host_free_process_arg(void *data)
 230 {
 231   smx_process_arg_t arg = *(void**)data;
 232   int i;
 233   for (i = 0; i < arg->argc; i++)
 234     xbt_free(arg->argv[i]);
 235   xbt_free(arg->argv);
 236   xbt_free(arg->name);
 237   xbt_free(arg);
 238 }
 239 /**
 240  * \brief Add a process to the list of the processes that the host will restart when it comes back
 241  * This function add a process to the list of the processes that will be restarted when the host comes
 242  * back. It is expected that this function is called when the host is down.
 243  * The processes will only be restarted once, meaning that you will have to register the process
 244  * again to restart the process again.
 245  */
 246 void SIMIX_host_add_auto_restart_process(sg_host_t host,
 247                                          const char *name,
 248                                          xbt_main_func_t code,
 249                                          void *data,
 250                                          const char *hostname,
 251                                          double kill_time,
 252                                          int argc, char **argv,
 253                                          xbt_dict_t properties,
 254                                          int auto_restart)
 255 {
 256   if (!sg_host_simix(host)->auto_restart_processes) {
 257     sg_host_simix(host)->auto_restart_processes = xbt_dynar_new(sizeof(smx_process_arg_t),_SIMIX_host_free_process_arg);
 258   }
 259   smx_process_arg_t arg = xbt_new(s_smx_process_arg_t,1);
 260   arg->name = xbt_strdup(name);
 261   arg->code = code;
 262   arg->data = data;
 263   arg->hostname = hostname;
 264   arg->kill_time = kill_time;
 265   arg->argc = argc;
 266
 267   arg->argv = xbt_new(char*,argc + 1);
 268
 269   int i;
 270   for (i = 0; i < argc; i++) {
 271     arg->argv[i] = xbt_strdup(argv[i]);
 272   }
 273   arg->argv[argc] = NULL;
 274
 275   arg->properties = properties;
 276   arg->auto_restart = auto_restart;
 277
 278   if( SIMIX_host_get_state(host) == SURF_RESOURCE_OFF
 279       && !xbt_dict_get_or_null(watched_hosts_lib,sg_host_name(host))){
 280     xbt_dict_set(watched_hosts_lib,sg_host_name(host),host,NULL);
 281     XBT_DEBUG("Have pushed host %s to watched_hosts_lib because state == SURF_RESOURCE_OFF",sg_host_name(host));
 282   }
 283   xbt_dynar_push_as(sg_host_simix(host)->auto_restart_processes,smx_process_arg_t,arg);
 284 }
 285 /**
 286  * \brief Restart the list of processes that have been registered to the host
 287  */
 288 void SIMIX_host_restart_processes(sg_host_t host)
 289 {
 290   unsigned int cpt;
 291   smx_process_arg_t arg;
 292   xbt_dynar_t process_list = sg_host_simix(host)->auto_restart_processes;
 293   if (!process_list)
 294     return;
 295
 296   xbt_dynar_foreach (process_list, cpt, arg) {
 297
 298     smx_process_t process;
 299
 300     XBT_DEBUG("Restarting Process %s(%s) right now", arg->argv[0], arg->hostname);
 301     if (simix_global->create_process_function) {
 302       simix_global->create_process_function(&process,
 303                                             arg->argv[0],
 304                                             arg->code,
 305                                             NULL,
 306                                             arg->hostname,
 307                                             arg->kill_time,
 308                                             arg->argc,
 309                                             arg->argv,
 310                                             arg->properties,
 311                                             arg->auto_restart,
 312                                             NULL);
 313     } else {
 314       simcall_process_create(&process,
 315                              arg->argv[0],
 316                              arg->code,
 317                              NULL,
 318                              arg->hostname,
 319                              arg->kill_time,
 320                              arg->argc,
 321                              arg->argv,
 322                              arg->properties,
 323                              arg->auto_restart);
 324
 325     }
 326     /* arg->argv is used by the process created above.  Hide it to
 327      * _SIMIX_host_free_process_arg() which is called by xbt_dynar_reset()
 328      * below. */
 329     arg->argc = 0;
 330     arg->argv = NULL;
 331   }
 332   xbt_dynar_reset(process_list);
 333 }
 334
 335 void SIMIX_host_autorestart(sg_host_t host)
 336 {
 337   if(simix_global->autorestart)
 338     simix_global->autorestart(host);
 339   else
 340     xbt_die("No function for simix_global->autorestart");
 341 }
 342
 343 smx_synchro_t SIMIX_host_execute(const char *name,
 344     sg_host_t host, double flops_amount, double priority, double bound, unsigned long affinity_mask){
 345
 346   /* alloc structures and initialize */
 347   smx_synchro_t synchro = xbt_mallocator_get(simix_global->synchro_mallocator);
 348   synchro->type = SIMIX_SYNC_EXECUTE;
 349   synchro->name = xbt_strdup(name);
 350   synchro->state = SIMIX_RUNNING;
 351   synchro->execution.host = host;
 352   synchro->category = NULL;
 353
 354   /* set surf's action */
 355   if (!MC_is_active() && !MC_record_replay_is_active()) {
 356
 357     synchro->execution.surf_exec = surf_host_execute(host, flops_amount);
 358     surf_action_set_data(synchro->execution.surf_exec, synchro);
 359     surf_action_set_priority(synchro->execution.surf_exec, priority);
 360
 361     /* Note (hypervisor): for multicore, the bound value being passed to the
 362      * surf layer should not be zero (i.e., unlimited). It should be the
 363      * capacity of a CPU core. */
 364     if (bound == 0)
 365       surf_cpu_action_set_bound(synchro->execution.surf_exec, SIMIX_host_get_speed(host));
 366     else
 367       surf_cpu_action_set_bound(synchro->execution.surf_exec, bound);
 368
 369     if (affinity_mask != 0) {
 370       /* just a double check to confirm that this host is the host where this task is running. */
 371       xbt_assert(synchro->execution.host == host);
 372       surf_cpu_action_set_affinity(synchro->execution.surf_exec, host, affinity_mask);
 373     }
 374   }
 375
 376   XBT_DEBUG("Create execute synchro %p: %s", synchro, synchro->name);
 377
 378   return synchro;
 379 }
 380
 381 smx_synchro_t SIMIX_host_parallel_execute(const char *name,
 382     int host_nb, sg_host_t *host_list,
 383     double *flops_amount, double *bytes_amount,
 384     double amount, double rate){
 385
 386   void **surf_host_list = NULL;
 387   int i;
 388
 389   /* alloc structures and initialize */
 390   smx_synchro_t synchro = xbt_mallocator_get(simix_global->synchro_mallocator);
 391   synchro->type = SIMIX_SYNC_PARALLEL_EXECUTE;
 392   synchro->name = xbt_strdup(name);
 393   synchro->state = SIMIX_RUNNING;
 394   synchro->execution.host = NULL; /* FIXME: do we need the list of hosts? */
 395   synchro->category = NULL;
 396
 397   /* set surf's synchro */
 398   surf_host_list = xbt_new0(void *, host_nb);
 399   for (i = 0; i < host_nb; i++)
 400     surf_host_list[i] = surf_host_resource_priv(host_list[i]);
 401
 402
 403   /* FIXME: what happens if host_list contains VMs and PMs. If
 404    * execute_parallel_task() does not change the state of the model, we can mix
 405    * them. */
 406   surf_model_t ws_model = surf_resource_model(host_list[0], SURF_HOST_LEVEL);
 407   for (i = 1; i < host_nb; i++) {
 408     surf_model_t ws_model_tmp = surf_resource_model(host_list[i], SURF_HOST_LEVEL);
 409     if (ws_model_tmp != ws_model) {
 410       XBT_CRITICAL("mixing VMs and PMs is not supported");
 411       DIE_IMPOSSIBLE;
 412     }
 413   }
 414
 415   /* set surf's synchro */
 416   if (!MC_is_active() && !MC_record_replay_is_active()) {
 417     synchro->execution.surf_exec =
 418       surf_host_model_execute_parallel_task((surf_host_model_t)surf_host_model,
 419                   host_nb, surf_host_list, flops_amount, bytes_amount, rate);
 420
 421     surf_action_set_data(synchro->execution.surf_exec, synchro);
 422   }
 423   XBT_DEBUG("Create parallel execute synchro %p", synchro);
 424
 425   return synchro;
 426 }
 427
 428 void SIMIX_host_execution_destroy(smx_synchro_t synchro){
 429   XBT_DEBUG("Destroy synchro %p", synchro);
 430
 431   if (synchro->execution.surf_exec) {
 432     surf_action_unref(synchro->execution.surf_exec);
 433     synchro->execution.surf_exec = NULL;
 434   }
 435   xbt_free(synchro->name);
 436   xbt_mallocator_release(simix_global->synchro_mallocator, synchro);
 437 }
 438
 439 void SIMIX_host_execution_cancel(smx_synchro_t synchro){
 440   XBT_DEBUG("Cancel synchro %p", synchro);
 441
 442   if (synchro->execution.surf_exec)
 443     surf_action_cancel(synchro->execution.surf_exec);
 444 }
 445
 446 double SIMIX_host_execution_get_remains(smx_synchro_t synchro){
 447   double result = 0.0;
 448
 449   if (synchro->state == SIMIX_RUNNING)
 450     result = surf_action_get_remains(synchro->execution.surf_exec);
 451
 452   return result;
 453 }
 454
 455 e_smx_state_t SIMIX_host_execution_get_state(smx_synchro_t synchro){
 456   return synchro->state;
 457 }
 458
 459 void SIMIX_host_execution_set_priority(smx_synchro_t synchro, double priority){
 460
 461   if(synchro->execution.surf_exec)
 462         surf_action_set_priority(synchro->execution.surf_exec, priority);
 463 }
 464
 465 void SIMIX_host_execution_set_bound(smx_synchro_t synchro, double bound){
 466
 467   if(synchro->execution.surf_exec)
 468         surf_cpu_action_set_bound(synchro->execution.surf_exec, bound);
 469 }
 470
 471 void SIMIX_host_execution_set_affinity(smx_synchro_t synchro, sg_host_t host, unsigned long mask){
 472   xbt_assert(synchro->type == SIMIX_SYNC_EXECUTE);
 473
 474   if (synchro->execution.surf_exec) {
 475     /* just a double check to confirm that this host is the host where this task is running. */
 476     xbt_assert(synchro->execution.host == host);
 477     surf_cpu_action_set_affinity(synchro->execution.surf_exec, host, mask);
 478   }
 479 }
 480
 481 void simcall_HANDLER_host_execution_wait(smx_simcall_t simcall, smx_synchro_t synchro){
 482
 483   XBT_DEBUG("Wait for execution of synchro %p, state %d", synchro, (int)synchro->state);
 484
 485   /* Associate this simcall to the synchro */
 486   xbt_fifo_push(synchro->simcalls, simcall);
 487   simcall->issuer->waiting_synchro = synchro;
 488
 489   /* set surf's synchro */
 490   if (MC_is_active() || MC_record_replay_is_active()) {
 491     synchro->state = SIMIX_DONE;
 492     SIMIX_execution_finish(synchro);
 493     return;
 494   }
 495
 496   /* If the synchro is already finished then perform the error handling */
 497   if (synchro->state != SIMIX_RUNNING)
 498     SIMIX_execution_finish(synchro);
 499 }
 500
 501 void SIMIX_host_execution_suspend(smx_synchro_t synchro)
 502 {
 503   if(synchro->execution.surf_exec)
 504     surf_action_suspend(synchro->execution.surf_exec);
 505 }
 506
 507 void SIMIX_host_execution_resume(smx_synchro_t synchro)
 508 {
 509   if(synchro->execution.surf_exec)
 510     surf_action_resume(synchro->execution.surf_exec);
 511 }
 512
 513 void SIMIX_execution_finish(smx_synchro_t synchro)
 514 {
 515   xbt_fifo_item_t item;
 516   smx_simcall_t simcall;
 517
 518   xbt_fifo_foreach(synchro->simcalls, item, simcall, smx_simcall_t) {
 519
 520     switch (synchro->state) {
 521
 522       case SIMIX_DONE:
 523         /* do nothing, synchro done */
 524         XBT_DEBUG("SIMIX_execution_finished: execution successful");
 525         break;
 526
 527       case SIMIX_FAILED:
 528         XBT_DEBUG("SIMIX_execution_finished: host '%s' failed", sg_host_name(simcall->issuer->host));
 529         simcall->issuer->context->iwannadie = 1;
 530         SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
 531         break;
 532
 533       case SIMIX_CANCELED:
 534         XBT_DEBUG("SIMIX_execution_finished: execution canceled");
 535         SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Canceled");
 536         break;
 537
 538       default:
 539         xbt_die("Internal error in SIMIX_execution_finish: unexpected synchro state %d",
 540             (int)synchro->state);
 541     }
 542     /* check if the host is down */
 543     if (surf_host_get_state(surf_host_resource_priv(simcall->issuer->host)) != SURF_RESOURCE_ON) {
 544       simcall->issuer->context->iwannadie = 1;
 545     }
 546
 547     simcall->issuer->waiting_synchro =    NULL;
 548     simcall_host_execution_wait__set__result(simcall, synchro->state);
 549     SIMIX_simcall_answer(simcall);
 550   }
 551
 552   /* We no longer need it */
 553   SIMIX_host_execution_destroy(synchro);
 554 }
 555
 556
 557 void SIMIX_post_host_execute(smx_synchro_t synchro)
 558 {
 559   if (synchro->type == SIMIX_SYNC_EXECUTE && /* FIMXE: handle resource failure
 560                                                * for parallel tasks too */
 561       surf_host_get_state(surf_host_resource_priv(synchro->execution.host)) == SURF_RESOURCE_OFF) {
 562     /* If the host running the synchro failed, notice it so that the asking
 563      * process can be killed if it runs on that host itself */
 564     synchro->state = SIMIX_FAILED;
 565   } else if (surf_action_get_state(synchro->execution.surf_exec) == SURF_ACTION_FAILED) {
 566     /* If the host running the synchro didn't fail, then the synchro was
 567      * canceled */
 568     synchro->state = SIMIX_CANCELED;
 569   } else {
 570     synchro->state = SIMIX_DONE;
 571   }
 572
 573   if (synchro->execution.surf_exec) {
 574     surf_action_unref(synchro->execution.surf_exec);
 575     synchro->execution.surf_exec = NULL;
 576   }
 577
 578   /* If there are simcalls associated with the synchro, then answer them */
 579   if (xbt_fifo_size(synchro->simcalls)) {
 580     SIMIX_execution_finish(synchro);
 581   }
 582 }
 583
 584
 585 void SIMIX_set_category(smx_synchro_t synchro, const char *category)
 586 {
 587   if (synchro->state != SIMIX_RUNNING) return;
 588   if (synchro->type == SIMIX_SYNC_EXECUTE){
 589     surf_action_set_category(synchro->execution.surf_exec, category);
 590   }else if (synchro->type == SIMIX_SYNC_COMMUNICATE){
 591     surf_action_set_category(synchro->comm.surf_comm, category);
 592   }
 593 }
 594
 595 /**
 596  * \brief Function to get the parameters of the given the SIMIX host.
 597  *
 598  * \param host the host to get_phys_host (a sg_host_t)
 599  * \param param the parameter object space to be overwritten (a ws_params_t)
 600  */
 601 void SIMIX_host_get_params(sg_host_t ind_vm, ws_params_t params)
 602 {
 603   /* jump to ws_get_params(). */
 604   surf_host_get_params(ind_vm, params);
 605 }
 606
 607 void SIMIX_host_set_params(sg_host_t ind_vm, ws_params_t params)
 608 {
 609   /* jump to ws_set_params(). */
 610   surf_host_set_params(ind_vm, params);
 611 }
 612
 613 xbt_dict_t SIMIX_host_get_mounted_storage_list(sg_host_t host){
 614   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 615
 616   return surf_host_get_mounted_storage_list(host);
 617 }
 618
 619 xbt_dynar_t SIMIX_host_get_attached_storage_list(sg_host_t host){
 620   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 621
 622   return surf_host_get_attached_storage_list(host);
 623 }