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