src/simix/smx_host.cpp

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