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