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