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 #include "src/surf/host_interface.hpp"
  14
  15 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_host, simix,
  16                                 "SIMIX hosts");
  17
  18 static void SIMIX_execution_finish(smx_synchro_t synchro);
  19
  20 /**
  21  * \brief Internal function to create a SIMIX host.
  22  * \param name name of the host to create
  23  */
  24 void SIMIX_host_create(sg_host_t host) // FIXME: braindead prototype. Take sg_host as parameter
  25 {
  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 (h->isOff()) {
  48     simgrid::surf::Host* surf_host = h->extension<simgrid::surf::Host>();
  49     surf_host->turnOn();
  50
  51     unsigned int cpt;
  52     smx_process_arg_t arg;
  53     xbt_dynar_foreach(host->boot_processes,cpt,arg) {
  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(argv[0],
  62                                               arg->code,
  63                                               NULL,
  64                                               arg->hostname,
  65                                               arg->kill_time,
  66                                               arg->argc,
  67                                               argv,
  68                                               arg->properties,
  69                                               arg->auto_restart,
  70                                               NULL);
  71       } else {
  72         simcall_process_create(arg->argv[0],
  73                                arg->code,
  74                                NULL,
  75                                arg->hostname,
  76                                arg->kill_time,
  77                                arg->argc,
  78                                argv,
  79                                arg->properties,
  80                                arg->auto_restart);
  81       }
  82     }
  83   }
  84 }
  85
  86 void simcall_HANDLER_host_off(smx_simcall_t simcall, sg_host_t h)
  87 {
  88  SIMIX_host_off(h, simcall->issuer);
  89 }
  90
  91 /**
  92  * \brief Stop the host if it is on
  93  *
  94  */
  95 void SIMIX_host_off(sg_host_t h, smx_process_t issuer)
  96 {
  97   smx_host_priv_t host = sg_host_simix(h);
  98
  99   xbt_assert((host != NULL), "Invalid parameters");
 100
 101   if (h->isOn()) {
 102     simgrid::surf::Host* surf_host = h->extension<simgrid::surf::Host>();
 103     surf_host->turnOff();
 104
 105     /* Clean Simulator data */
 106     if (xbt_swag_size(host->process_list) != 0) {
 107       smx_process_t process = NULL;
 108       xbt_swag_foreach(process, host->process_list) {
 109         SIMIX_process_kill(process, issuer);
 110         XBT_DEBUG("Killing %s on %s by %s", process->name,  sg_host_get_name(process->host), issuer->name);
 111       }
 112     }
 113   } else {
 114     XBT_INFO("Host %s is already off",h->getName().c_str());
 115   }
 116 }
 117
 118 /**
 119  * \brief Internal function to destroy a SIMIX host.
 120  *
 121  * \param h the host to destroy (a sg_host_t)
 122  */
 123 void SIMIX_host_destroy(void *h)
 124 {
 125   smx_host_priv_t host = (smx_host_priv_t) h;
 126
 127   xbt_assert((host != NULL), "Invalid parameters");
 128
 129   /* Clean Simulator data */
 130   if (xbt_swag_size(host->process_list) != 0) {
 131     char *msg = xbt_strdup("Shutting down host, but it's not empty:");
 132     char *tmp;
 133     smx_process_t process = NULL;
 134
 135     xbt_swag_foreach(process, host->process_list) {
 136       tmp = bprintf("%s\n\t%s", msg, process->name);
 137       free(msg);
 138       msg = tmp;
 139     }
 140     SIMIX_display_process_status();
 141     THROWF(arg_error, 0, "%s", msg);
 142   }
 143   xbt_dynar_free(&host->auto_restart_processes);
 144   xbt_dynar_free(&host->boot_processes);
 145   xbt_swag_free(host->process_list);
 146
 147   /* Clean host structure */
 148   free(host);
 149   return;
 150 }
 151
 152 sg_host_t SIMIX_host_self(void)
 153 {
 154   smx_process_t process = SIMIX_process_self();
 155   return (process == NULL) ? NULL : SIMIX_process_get_host(process);
 156 }
 157
 158 /* needs to be public and without simcall because it is called
 159    by exceptions and logging events */
 160 const char* SIMIX_host_self_get_name(void)
 161 {
 162   sg_host_t host = SIMIX_host_self();
 163   if (host == NULL || SIMIX_process_self() == simix_global->maestro_process)
 164     return "";
 165
 166   return SIMIX_host_get_name(host);
 167 }
 168
 169 xbt_dict_t SIMIX_host_get_properties(sg_host_t host){
 170   return sg_host_get_properties(host);
 171 }
 172
 173
 174 xbt_swag_t SIMIX_host_get_process_list(sg_host_t host){
 175   smx_host_priv_t host_priv = sg_host_simix(host);
 176
 177   return host_priv->process_list;
 178 }
 179
 180
 181 double SIMIX_host_get_current_power_peak(sg_host_t host) {
 182           return surf_host_get_current_power_peak(host);
 183 }
 184
 185 double SIMIX_host_get_power_peak_at(sg_host_t host, int pstate_index) {
 186           return surf_host_get_power_peak_at(host, pstate_index);
 187 }
 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     synchro->execution.surf_exec->setData(synchro);
 325     synchro->execution.surf_exec->setPriority(priority);
 326
 327     if (bound != 0)
 328       static_cast<simgrid::surf::CpuAction*>(synchro->execution.surf_exec)
 329         ->setBound(bound);
 330
 331     if (affinity_mask != 0) {
 332       /* just a double check to confirm that this host is the host where this task is running. */
 333       xbt_assert(synchro->execution.host == issuer->host);
 334       static_cast<simgrid::surf::CpuAction*>(synchro->execution.surf_exec)
 335         ->setAffinity(issuer->host->p_cpu, affinity_mask);
 336     }
 337   }
 338
 339   XBT_DEBUG("Create execute synchro %p: %s", synchro, synchro->name);
 340
 341   return synchro;
 342 }
 343
 344 smx_synchro_t SIMIX_process_parallel_execute(const char *name,
 345     int host_nb, sg_host_t *host_list,
 346     double *flops_amount, double *bytes_amount,
 347     double amount, double rate){
 348
 349   sg_host_t*host_list_cpy = NULL;
 350   int i;
 351
 352   /* alloc structures and initialize */
 353   smx_synchro_t synchro = (smx_synchro_t) xbt_mallocator_get(simix_global->synchro_mallocator);
 354   synchro->type = SIMIX_SYNC_PARALLEL_EXECUTE;
 355   synchro->name = xbt_strdup(name);
 356   synchro->state = SIMIX_RUNNING;
 357   synchro->execution.host = NULL; /* FIXME: do we need the list of hosts? */
 358   synchro->category = NULL;
 359
 360   /* set surf's synchro */
 361   host_list_cpy = xbt_new0(sg_host_t, host_nb);
 362   for (i = 0; i < host_nb; i++)
 363     host_list_cpy[i] = host_list[i];
 364
 365
 366   /* FIXME: what happens if host_list contains VMs and PMs. If
 367    * execute_parallel_task() does not change the state of the model, we can mix
 368    * them. */
 369   surf_host_model_t ws_model = surf_host_get_model(host_list[0]);
 370   for (i = 1; i < host_nb; i++) {
 371     surf_host_model_t ws_model_tmp = surf_host_get_model(host_list[0]);
 372     if (ws_model_tmp != ws_model) {
 373       XBT_CRITICAL("mixing VMs and PMs is not supported");
 374       DIE_IMPOSSIBLE;
 375     }
 376   }
 377
 378   /* set surf's synchro */
 379   if (!MC_is_active() && !MC_record_replay_is_active()) {
 380     synchro->execution.surf_exec =
 381       surf_host_model_execute_parallel_task(surf_host_model,
 382                   host_nb, host_list_cpy, flops_amount, bytes_amount, rate);
 383
 384     synchro->execution.surf_exec->setData(synchro);
 385   }
 386   XBT_DEBUG("Create parallel execute synchro %p", synchro);
 387
 388   return synchro;
 389 }
 390
 391 void SIMIX_process_execution_destroy(smx_synchro_t synchro){
 392   XBT_DEBUG("Destroy synchro %p", synchro);
 393
 394   if (synchro->execution.surf_exec) {
 395     synchro->execution.surf_exec->unref();
 396     synchro->execution.surf_exec = NULL;
 397   }
 398   xbt_free(synchro->name);
 399   xbt_mallocator_release(simix_global->synchro_mallocator, synchro);
 400 }
 401
 402 void SIMIX_process_execution_cancel(smx_synchro_t synchro){
 403   XBT_DEBUG("Cancel synchro %p", synchro);
 404
 405   if (synchro->execution.surf_exec)
 406     synchro->execution.surf_exec->cancel();
 407 }
 408
 409 double SIMIX_process_execution_get_remains(smx_synchro_t synchro){
 410   double result = 0.0;
 411
 412   if (synchro->state == SIMIX_RUNNING)
 413     result = surf_action_get_remains(synchro->execution.surf_exec);
 414
 415   return result;
 416 }
 417
 418 e_smx_state_t SIMIX_process_execution_get_state(smx_synchro_t synchro){
 419   return synchro->state;
 420 }
 421
 422 void SIMIX_process_execution_set_priority(smx_synchro_t synchro, double priority){
 423
 424   if(synchro->execution.surf_exec)
 425         synchro->execution.surf_exec->setPriority(priority);
 426 }
 427
 428 void SIMIX_process_execution_set_bound(smx_synchro_t synchro, double bound){
 429
 430   if(synchro->execution.surf_exec)
 431         static_cast<simgrid::surf::CpuAction*>(synchro->execution.surf_exec)
 432     ->setBound(bound);
 433 }
 434
 435 void SIMIX_process_execution_set_affinity(smx_synchro_t synchro, sg_host_t host, unsigned long mask){
 436   xbt_assert(synchro->type == SIMIX_SYNC_EXECUTE);
 437
 438   if (synchro->execution.surf_exec) {
 439     /* just a double check to confirm that this host is the host where this task is running. */
 440     xbt_assert(synchro->execution.host == host);
 441     static_cast<simgrid::surf::CpuAction*>(synchro->execution.surf_exec)
 442       ->setAffinity(host->p_cpu, mask);
 443   }
 444 }
 445
 446 void simcall_HANDLER_process_execution_wait(smx_simcall_t simcall, smx_synchro_t synchro){
 447
 448   XBT_DEBUG("Wait for execution of synchro %p, state %d", synchro, (int)synchro->state);
 449
 450   /* Associate this simcall to the synchro */
 451   xbt_fifo_push(synchro->simcalls, simcall);
 452   simcall->issuer->waiting_synchro = synchro;
 453
 454   /* set surf's synchro */
 455   if (MC_is_active() || MC_record_replay_is_active()) {
 456     synchro->state = SIMIX_DONE;
 457     SIMIX_execution_finish(synchro);
 458     return;
 459   }
 460
 461   /* If the synchro is already finished then perform the error handling */
 462   if (synchro->state != SIMIX_RUNNING)
 463     SIMIX_execution_finish(synchro);
 464 }
 465
 466 void SIMIX_host_execution_suspend(smx_synchro_t synchro)
 467 {
 468   if(synchro->execution.surf_exec)
 469     synchro->execution.surf_exec->suspend();
 470 }
 471
 472 void SIMIX_host_execution_resume(smx_synchro_t synchro)
 473 {
 474   if(synchro->execution.surf_exec)
 475     synchro->execution.surf_exec->resume();
 476 }
 477
 478 void SIMIX_execution_finish(smx_synchro_t synchro)
 479 {
 480   xbt_fifo_item_t item;
 481   smx_simcall_t simcall;
 482
 483   xbt_fifo_foreach(synchro->simcalls, item, simcall, smx_simcall_t) {
 484
 485     switch (synchro->state) {
 486
 487       case SIMIX_DONE:
 488         /* do nothing, synchro done */
 489         XBT_DEBUG("SIMIX_execution_finished: execution successful");
 490         break;
 491
 492       case SIMIX_FAILED:
 493         XBT_DEBUG("SIMIX_execution_finished: host '%s' failed", sg_host_get_name(simcall->issuer->host));
 494         simcall->issuer->context->iwannadie = 1;
 495         SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
 496         break;
 497
 498       case SIMIX_CANCELED:
 499         XBT_DEBUG("SIMIX_execution_finished: execution canceled");
 500         SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Canceled");
 501         break;
 502
 503       default:
 504         xbt_die("Internal error in SIMIX_execution_finish: unexpected synchro state %d",
 505             (int)synchro->state);
 506     }
 507     /* check if the host is down */
 508     if (simcall->issuer->host->isOff()) {
 509       simcall->issuer->context->iwannadie = 1;
 510     }
 511
 512     simcall->issuer->waiting_synchro =    NULL;
 513     simcall_process_execution_wait__set__result(simcall, synchro->state);
 514     SIMIX_simcall_answer(simcall);
 515   }
 516
 517   /* We no longer need it */
 518   SIMIX_process_execution_destroy(synchro);
 519 }
 520
 521
 522 void SIMIX_post_host_execute(smx_synchro_t synchro)
 523 {
 524   if (synchro->type == SIMIX_SYNC_EXECUTE && /* FIMXE: handle resource failure
 525                                                * for parallel tasks too */
 526       synchro->execution.host->isOff()) {
 527     /* If the host running the synchro failed, notice it so that the asking
 528      * process can be killed if it runs on that host itself */
 529     synchro->state = SIMIX_FAILED;
 530   } else if (synchro->execution.surf_exec->getState() == SURF_ACTION_FAILED) {
 531     /* If the host running the synchro didn't fail, then the synchro was
 532      * canceled */
 533     synchro->state = SIMIX_CANCELED;
 534   } else {
 535     synchro->state = SIMIX_DONE;
 536   }
 537
 538   if (synchro->execution.surf_exec) {
 539     synchro->execution.surf_exec->unref();
 540     synchro->execution.surf_exec = NULL;
 541   }
 542
 543   /* If there are simcalls associated with the synchro, then answer them */
 544   if (xbt_fifo_size(synchro->simcalls)) {
 545     SIMIX_execution_finish(synchro);
 546   }
 547 }
 548
 549
 550 void SIMIX_set_category(smx_synchro_t synchro, const char *category)
 551 {
 552   if (synchro->state != SIMIX_RUNNING) return;
 553   if (synchro->type == SIMIX_SYNC_EXECUTE){
 554     synchro->execution.surf_exec->setCategory(category);
 555   }else if (synchro->type == SIMIX_SYNC_COMMUNICATE){
 556     synchro->comm.surf_comm->setCategory(category);
 557   }
 558 }
 559
 560 /**
 561  * \brief Function to get the parameters of the given the SIMIX host.
 562  *
 563  * \param host the host to get_phys_host (a sg_host_t)
 564  * \param param the parameter object space to be overwritten (a ws_params_t)
 565  */
 566 void SIMIX_host_get_params(sg_host_t ind_vm, vm_params_t params)
 567 {
 568   ind_vm->extension<simgrid::surf::Host>()->getParams(params);
 569 }
 570
 571 void SIMIX_host_set_params(sg_host_t ind_vm, vm_params_t params)
 572 {
 573   ind_vm->extension<simgrid::surf::Host>()->setParams(params);
 574 }
 575
 576 xbt_dict_t SIMIX_host_get_mounted_storage_list(sg_host_t host)
 577 {
 578   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 579   return host->extension<simgrid::surf::Host>()->getMountedStorageList();
 580 }
 581
 582 xbt_dynar_t SIMIX_host_get_attached_storage_list(sg_host_t host)
 583 {
 584   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
 585   return host->extension<simgrid::surf::Host>()->getAttachedStorageList();
 586 }