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 "mc/mc.h"
  10 #include "src/mc/mc_replay.h"
  11 #include "src/surf/virtual_machine.hpp"
  12 #include "src/surf/HostImpl.hpp"
  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 = xbt_swag_new(xbt_swag_offset(proc, host_proc_hookup));
  30
  31   /* Update global variables */
  32   sg_host_simix_set(host, smx_host);
  33 }
  34
  35 /**
  36  * \brief Start the host if it is off
  37  *
  38  */
  39 void SIMIX_host_on(sg_host_t h)
  40 {
  41   smx_host_priv_t host = sg_host_simix(h);
  42
  43   xbt_assert((host != NULL), "Invalid parameters");
  44
  45   if (h->isOff()) {
  46     simgrid::surf::HostImpl* surf_host = h->extension<simgrid::surf::HostImpl>();
  47     surf_host->turnOn();
  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 /**
  85  * \brief Stop the host if it is on
  86  *
  87  */
  88 void SIMIX_host_off(sg_host_t h, smx_process_t issuer)
  89 {
  90   smx_host_priv_t host = sg_host_simix(h);
  91
  92   xbt_assert((host != NULL), "Invalid parameters");
  93
  94   if (h->isOn()) {
  95     simgrid::surf::HostImpl* surf_host = h->extension<simgrid::surf::HostImpl>();
  96     surf_host->turnOff();
  97
  98     /* Clean Simulator data */
  99     if (xbt_swag_size(host->process_list) != 0) {
 100       smx_process_t process = NULL;
 101       xbt_swag_foreach(process, host->process_list) {
 102         SIMIX_process_kill(process, issuer);
 103         XBT_DEBUG("Killing %s on %s by %s", process->name,  sg_host_get_name(process->host), issuer->name);
 104       }
 105     }
 106   } else {
 107     XBT_INFO("Host %s is already off",h->name().c_str());
 108   }
 109 }
 110
 111 /**
 112  * \brief Internal function to destroy a SIMIX host.
 113  *
 114  * \param h the host to destroy (a sg_host_t)
 115  */
 116 void SIMIX_host_destroy(void *h)
 117 {
 118   smx_host_priv_t host = (smx_host_priv_t) h;
 119
 120   xbt_assert((host != NULL), "Invalid parameters");
 121
 122   /* Clean Simulator data */
 123   if (xbt_swag_size(host->process_list) != 0) {
 124     char *msg = xbt_strdup("Shutting down host, but it's not empty:");
 125     char *tmp;
 126     smx_process_t process = NULL;
 127
 128     xbt_swag_foreach(process, host->process_list) {
 129       tmp = bprintf("%s\n\t%s", msg, process->name);
 130       free(msg);
 131       msg = tmp;
 132     }
 133     SIMIX_display_process_status();
 134     THROWF(arg_error, 0, "%s", msg);
 135   }
 136   xbt_dynar_free(&host->auto_restart_processes);
 137   xbt_dynar_free(&host->boot_processes);
 138   xbt_swag_free(host->process_list);
 139
 140   /* Clean host structure */
 141   free(host);
 142   return;
 143 }
 144
 145 sg_host_t SIMIX_host_self(void)
 146 {
 147   smx_process_t process = SIMIX_process_self();
 148   return (process == NULL) ? NULL : SIMIX_process_get_host(process);
 149 }
 150
 151 /* needs to be public and without simcall because it is called
 152    by exceptions and logging events */
 153 const char* SIMIX_host_self_get_name(void)
 154 {
 155   sg_host_t host = SIMIX_host_self();
 156   if (host == NULL || SIMIX_process_self() == simix_global->maestro_process)
 157     return "";
 158
 159   return sg_host_get_name(host);
 160 }
 161
 162 void _SIMIX_host_free_process_arg(void *data)
 163 {
 164   smx_process_arg_t arg = *(smx_process_arg_t*)data;
 165   int i;
 166   for (i = 0; i < arg->argc; i++)
 167     xbt_free(arg->argv[i]);
 168   xbt_free(arg->argv);
 169   xbt_free(arg->name);
 170   xbt_free(arg);
 171 }
 172 /**
 173  * \brief Add a process to the list of the processes that the host will restart when it comes back
 174  * This function add a process to the list of the processes that will be restarted when the host comes
 175  * back. It is expected that this function is called when the host is down.
 176  * The processes will only be restarted once, meaning that you will have to register the process
 177  * again to restart the process again.
 178  */
 179 void SIMIX_host_add_auto_restart_process(sg_host_t host,
 180                                          const char *name,
 181                                          xbt_main_func_t code,
 182                                          void *data,
 183                                          const char *hostname,
 184                                          double kill_time,
 185                                          int argc, char **argv,
 186                                          xbt_dict_t properties,
 187                                          int auto_restart)
 188 {
 189   if (!sg_host_simix(host)->auto_restart_processes) {
 190     sg_host_simix(host)->auto_restart_processes = xbt_dynar_new(sizeof(smx_process_arg_t),_SIMIX_host_free_process_arg);
 191   }
 192   smx_process_arg_t arg = xbt_new(s_smx_process_arg_t,1);
 193   arg->name = xbt_strdup(name);
 194   arg->code = code;
 195   arg->data = data;
 196   arg->hostname = hostname;
 197   arg->kill_time = kill_time;
 198   arg->argc = argc;
 199
 200   arg->argv = xbt_new(char*,argc + 1);
 201
 202   for (int i = 0; i < argc; i++)
 203     arg->argv[i] = xbt_strdup(argv[i]);
 204   arg->argv[argc] = NULL;
 205
 206   arg->properties = properties;
 207   arg->auto_restart = auto_restart;
 208
 209   if( host->isOff() && !xbt_dict_get_or_null(watched_hosts_lib,sg_host_get_name(host))){
 210     xbt_dict_set(watched_hosts_lib,sg_host_get_name(host),host,NULL);
 211     XBT_DEBUG("Push host %s to watched_hosts_lib because state == SURF_RESOURCE_OFF",sg_host_get_name(host));
 212   }
 213   xbt_dynar_push_as(sg_host_simix(host)->auto_restart_processes,smx_process_arg_t,arg);
 214 }
 215 /**
 216  * \brief Restart the list of processes that have been registered to the host
 217  */
 218 void SIMIX_host_autorestart(sg_host_t host)
 219 {
 220   unsigned int cpt;
 221   smx_process_arg_t arg;
 222   xbt_dynar_t process_list = sg_host_simix(host)->auto_restart_processes;
 223   if (!process_list)
 224     return;
 225
 226   xbt_dynar_foreach (process_list, cpt, arg) {
 227
 228     XBT_DEBUG("Restarting Process %s(%s) right now", arg->argv[0], arg->hostname);
 229     if (simix_global->create_process_function) {
 230       simix_global->create_process_function(arg->argv[0],
 231                                             arg->code,
 232                                             NULL,
 233                                             arg->hostname,
 234                                             arg->kill_time,
 235                                             arg->argc,
 236                                             arg->argv,
 237                                             arg->properties,
 238                                             arg->auto_restart,
 239                                             NULL);
 240     } else {
 241       simcall_process_create(arg->argv[0],
 242                              (xbt_main_func_t) arg->code,
 243                              NULL,
 244                              arg->hostname,
 245                              arg->kill_time,
 246                              arg->argc,
 247                              arg->argv,
 248                              arg->properties,
 249                              arg->auto_restart);
 250
 251     }
 252     /* arg->argv is used by the process created above.  Hide it to
 253      * _SIMIX_host_free_process_arg() which is called by xbt_dynar_reset()
 254      * below. */
 255     arg->argc = 0;
 256     arg->argv = NULL;
 257   }
 258   xbt_dynar_reset(process_list);
 259 }
 260
 261 smx_synchro_t simcall_HANDLER_execution_start(smx_simcall_t simcall,
 262     const char* name, double flops_amount, double priority, double bound, unsigned long affinity_mask) {
 263   return SIMIX_execution_start(simcall->issuer, name,flops_amount,priority,bound,affinity_mask);
 264 }
 265 smx_synchro_t SIMIX_execution_start(smx_process_t issuer, const char *name,
 266      double flops_amount, double priority, double bound, unsigned long affinity_mask){
 267
 268   /* alloc structures and initialize */
 269   smx_synchro_t synchro = (smx_synchro_t) xbt_mallocator_get(simix_global->synchro_mallocator);
 270   synchro->type = SIMIX_SYNC_EXECUTE;
 271   synchro->name = xbt_strdup(name);
 272   synchro->state = SIMIX_RUNNING;
 273   synchro->execution.host = issuer->host;
 274   synchro->category = NULL;
 275
 276   /* set surf's action */
 277   if (!MC_is_active() && !MC_record_replay_is_active()) {
 278
 279     synchro->execution.surf_exec = issuer->host->pimpl_cpu->execution_start(flops_amount);
 280     synchro->execution.surf_exec->setData(synchro);
 281     synchro->execution.surf_exec->setPriority(priority);
 282
 283     if (bound != 0)
 284       static_cast<simgrid::surf::CpuAction*>(synchro->execution.surf_exec)
 285         ->setBound(bound);
 286
 287     if (affinity_mask != 0) {
 288       /* just a double check to confirm that this host is the host where this task is running. */
 289       xbt_assert(synchro->execution.host == issuer->host);
 290       static_cast<simgrid::surf::CpuAction*>(synchro->execution.surf_exec)
 291         ->setAffinity(issuer->host->pimpl_cpu, affinity_mask);
 292     }
 293   }
 294
 295   XBT_DEBUG("Create execute synchro %p: %s", synchro, synchro->name);
 296
 297   return synchro;
 298 }
 299
 300 smx_synchro_t SIMIX_execution_parallel_start(const char *name,
 301     int host_nb, sg_host_t *host_list,
 302     double *flops_amount, double *bytes_amount,
 303     double amount, double rate){
 304
 305   sg_host_t*host_list_cpy = NULL;
 306   int i;
 307
 308   /* alloc structures and initialize */
 309   smx_synchro_t synchro = (smx_synchro_t) xbt_mallocator_get(simix_global->synchro_mallocator);
 310   synchro->type = SIMIX_SYNC_PARALLEL_EXECUTE;
 311   synchro->name = xbt_strdup(name);
 312   synchro->state = SIMIX_RUNNING;
 313   synchro->execution.host = NULL; /* FIXME: do we need the list of hosts? */
 314   synchro->category = NULL;
 315
 316   /* set surf's synchro */
 317   host_list_cpy = xbt_new0(sg_host_t, host_nb);
 318   for (i = 0; i < host_nb; i++)
 319     host_list_cpy[i] = host_list[i];
 320
 321   /* Check that we are not mixing VMs and PMs in the parallel task */
 322   simgrid::surf::HostImpl *host = host_list[0]->extension<simgrid::surf::HostImpl>();
 323   bool is_a_vm = (nullptr != dynamic_cast<simgrid::surf::VirtualMachine*>(host));
 324   for (i = 1; i < host_nb; i++) {
 325     bool tmp_is_a_vm = (nullptr != dynamic_cast<simgrid::surf::VirtualMachine*>(host_list[i]->extension<simgrid::surf::HostImpl>()));
 326     xbt_assert(is_a_vm == tmp_is_a_vm, "parallel_execute: mixing VMs and PMs is not supported (yet).");
 327   }
 328
 329   /* set surf's synchro */
 330   if (!MC_is_active() && !MC_record_replay_is_active()) {
 331     synchro->execution.surf_exec =
 332       surf_host_model->executeParallelTask(
 333           host_nb, host_list_cpy, flops_amount, bytes_amount, rate);
 334
 335     synchro->execution.surf_exec->setData(synchro);
 336   }
 337   XBT_DEBUG("Create parallel execute synchro %p", synchro);
 338
 339   return synchro;
 340 }
 341
 342 void SIMIX_execution_destroy(smx_synchro_t synchro)
 343 {
 344   XBT_DEBUG("Destroy synchro %p", synchro);
 345
 346   if (synchro->execution.surf_exec) {
 347     synchro->execution.surf_exec->unref();
 348     synchro->execution.surf_exec = NULL;
 349   }
 350   xbt_free(synchro->name);
 351   xbt_mallocator_release(simix_global->synchro_mallocator, synchro);
 352 }
 353
 354 void SIMIX_execution_cancel(smx_synchro_t synchro)
 355 {
 356   XBT_DEBUG("Cancel synchro %p", synchro);
 357
 358   if (synchro->execution.surf_exec)
 359     synchro->execution.surf_exec->cancel();
 360 }
 361
 362 double SIMIX_execution_get_remains(smx_synchro_t synchro)
 363 {
 364   double result = 0.0;
 365
 366   if (synchro->state == SIMIX_RUNNING)
 367     result = synchro->execution.surf_exec->getRemains();
 368
 369   return result;
 370 }
 371
 372 e_smx_state_t SIMIX_execution_get_state(smx_synchro_t synchro)
 373 {
 374   return synchro->state;
 375 }
 376
 377 void SIMIX_execution_set_priority(smx_synchro_t synchro, double priority)
 378 {
 379   if(synchro->execution.surf_exec)
 380     synchro->execution.surf_exec->setPriority(priority);
 381 }
 382
 383 void SIMIX_execution_set_bound(smx_synchro_t synchro, double bound)
 384 {
 385   if(synchro->execution.surf_exec)
 386     static_cast<simgrid::surf::CpuAction*>(synchro->execution.surf_exec)->setBound(bound);
 387 }
 388
 389 void SIMIX_execution_set_affinity(smx_synchro_t synchro, sg_host_t host, unsigned long mask)
 390 {
 391   xbt_assert(synchro->type == SIMIX_SYNC_EXECUTE);
 392
 393   if (synchro->execution.surf_exec) {
 394     /* just a double check to confirm that this host is the host where this task is running. */
 395     xbt_assert(synchro->execution.host == host);
 396     static_cast<simgrid::surf::CpuAction*>(synchro->execution.surf_exec)->setAffinity(host->pimpl_cpu, mask);
 397   }
 398 }
 399
 400 void simcall_HANDLER_execution_wait(smx_simcall_t simcall, smx_synchro_t synchro)
 401 {
 402
 403   XBT_DEBUG("Wait for execution of synchro %p, state %d", synchro, (int)synchro->state);
 404
 405   /* Associate this simcall to the synchro */
 406   xbt_fifo_push(synchro->simcalls, simcall);
 407   simcall->issuer->waiting_synchro = synchro;
 408
 409   /* set surf's synchro */
 410   if (MC_is_active() || MC_record_replay_is_active()) {
 411     synchro->state = SIMIX_DONE;
 412     SIMIX_execution_finish(synchro);
 413     return;
 414   }
 415
 416   /* If the synchro is already finished then perform the error handling */
 417   if (synchro->state != SIMIX_RUNNING)
 418     SIMIX_execution_finish(synchro);
 419 }
 420
 421 void SIMIX_execution_suspend(smx_synchro_t synchro)
 422 {
 423   if(synchro->execution.surf_exec)
 424     synchro->execution.surf_exec->suspend();
 425 }
 426
 427 void SIMIX_execution_resume(smx_synchro_t synchro)
 428 {
 429   if(synchro->execution.surf_exec)
 430     synchro->execution.surf_exec->resume();
 431 }
 432
 433 void SIMIX_execution_finish(smx_synchro_t synchro)
 434 {
 435   xbt_fifo_item_t item;
 436   smx_simcall_t simcall;
 437
 438   xbt_fifo_foreach(synchro->simcalls, item, simcall, smx_simcall_t) {
 439
 440     switch (synchro->state) {
 441
 442       case SIMIX_DONE:
 443         /* do nothing, synchro done */
 444         XBT_DEBUG("SIMIX_execution_finished: execution successful");
 445         break;
 446
 447       case SIMIX_FAILED:
 448         XBT_DEBUG("SIMIX_execution_finished: host '%s' failed", sg_host_get_name(simcall->issuer->host));
 449         simcall->issuer->context->iwannadie = 1;
 450         SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
 451         break;
 452
 453       case SIMIX_CANCELED:
 454         XBT_DEBUG("SIMIX_execution_finished: execution canceled");
 455         SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Canceled");
 456         break;
 457
 458       default:
 459         xbt_die("Internal error in SIMIX_execution_finish: unexpected synchro state %d",
 460             (int)synchro->state);
 461     }
 462     /* check if the host is down */
 463     if (simcall->issuer->host->isOff()) {
 464       simcall->issuer->context->iwannadie = 1;
 465     }
 466
 467     simcall->issuer->waiting_synchro =    NULL;
 468     simcall_execution_wait__set__result(simcall, synchro->state);
 469     SIMIX_simcall_answer(simcall);
 470   }
 471
 472   /* We no longer need it */
 473   SIMIX_execution_destroy(synchro);
 474 }
 475
 476
 477 void SIMIX_post_host_execute(smx_synchro_t synchro)
 478 {
 479   if (synchro->type == SIMIX_SYNC_EXECUTE && /* FIMXE: handle resource failure
 480                                                * for parallel tasks too */
 481       synchro->execution.host->isOff()) {
 482     /* If the host running the synchro failed, notice it so that the asking
 483      * process can be killed if it runs on that host itself */
 484     synchro->state = SIMIX_FAILED;
 485   } else if (synchro->execution.surf_exec->getState() == simgrid::surf::Action::State::failed) {
 486     /* If the host running the synchro didn't fail, then the synchro was
 487      * canceled */
 488     synchro->state = SIMIX_CANCELED;
 489   } else {
 490     synchro->state = SIMIX_DONE;
 491   }
 492
 493   if (synchro->execution.surf_exec) {
 494     synchro->execution.surf_exec->unref();
 495     synchro->execution.surf_exec = NULL;
 496   }
 497
 498   /* If there are simcalls associated with the synchro, then answer them */
 499   if (xbt_fifo_size(synchro->simcalls)) {
 500     SIMIX_execution_finish(synchro);
 501   }
 502 }
 503
 504
 505 void SIMIX_set_category(smx_synchro_t synchro, const char *category)
 506 {
 507   if (synchro->state != SIMIX_RUNNING) return;
 508   if (synchro->type == SIMIX_SYNC_EXECUTE){
 509     synchro->execution.surf_exec->setCategory(category);
 510   }else if (synchro->type == SIMIX_SYNC_COMMUNICATE){
 511     synchro->comm.surf_comm->setCategory(category);
 512   }
 513 }