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