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/virtual_machine.hpp"
  14 #include "src/surf/host_interface.hpp"
  15
  16 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_host, simix,
  17                                 "SIMIX hosts");
  18
  19 static void SIMIX_execution_finish(smx_synchro_t synchro);
  20
  21 /**
  22  * \brief Internal function to create a SIMIX host.
  23  * \param name name of the host to create
  24  */
  25 void SIMIX_host_create(sg_host_t host) // FIXME: braindead prototype. Take sg_host as parameter
  26 {
  27   smx_host_priv_t smx_host = xbt_new0(s_smx_host_priv_t, 1);
  28   s_smx_process_t proc;
  29
  30   /* Host structure */
  31   smx_host->process_list =
  32       xbt_swag_new(xbt_swag_offset(proc, host_proc_hookup));
  33
  34   /* Update global variables */
  35   sg_host_simix_set(host, smx_host);
  36 }
  37
  38 /**
  39  * \brief Start the host if it is off
  40  *
  41  */
  42 void SIMIX_host_on(sg_host_t h)
  43 {
  44   smx_host_priv_t host = sg_host_simix(h);
  45
  46   xbt_assert((host != NULL), "Invalid parameters");
  47
  48   if (h->isOff()) {
  49     simgrid::surf::Host* surf_host = h->extension<simgrid::surf::Host>();
  50     surf_host->turnOn();
  51
  52     unsigned int cpt;
  53     smx_process_arg_t arg;
  54     xbt_dynar_foreach(host->boot_processes,cpt,arg) {
  55
  56       char** argv = xbt_new(char*, arg->argc);
  57       for (int i=0; i<arg->argc; i++)
  58         argv[i] = xbt_strdup(arg->argv[i]);
  59
  60       XBT_DEBUG("Booting Process %s(%s) right now", arg->argv[0], arg->hostname);
  61       if (simix_global->create_process_function) {
  62         simix_global->create_process_function(argv[0],
  63                                               arg->code,
  64                                               NULL,
  65                                               arg->hostname,
  66                                               arg->kill_time,
  67                                               arg->argc,
  68                                               argv,
  69                                               arg->properties,
  70                                               arg->auto_restart,
  71                                               NULL);
  72       } else {
  73         simcall_process_create(arg->argv[0],
  74                                arg->code,
  75                                NULL,
  76                                arg->hostname,
  77                                arg->kill_time,
  78                                arg->argc,
  79                                argv,
  80                                arg->properties,
  81                                arg->auto_restart);
  82       }
  83     }
  84   }
  85 }
  86
  87 /**
  88  * \brief Stop the host if it is on
  89  *
  90  */
  91 void SIMIX_host_off(sg_host_t h, smx_process_t issuer)
  92 {
  93   smx_host_priv_t host = sg_host_simix(h);
  94
  95   xbt_assert((host != NULL), "Invalid parameters");
  96
  97   if (h->isOn()) {
  98     simgrid::surf::Host* surf_host = h->extension<simgrid::surf::Host>();
  99     surf_host->turnOff();
 100
 101     /* Clean Simulator data */
 102     if (xbt_swag_size(host->process_list) != 0) {
 103       smx_process_t process = NULL;
 104       xbt_swag_foreach(process, host->process_list) {
 105         SIMIX_process_kill(process, issuer);
 106         XBT_DEBUG("Killing %s on %s by %s", process->name,  sg_host_get_name(process->host), issuer->name);
 107       }
 108     }
 109   } else {
 110     XBT_INFO("Host %s is already off",h->name().c_str());
 111   }
 112 }
 113
 114 /**
 115  * \brief Internal function to destroy a SIMIX host.
 116  *
 117  * \param h the host to destroy (a sg_host_t)
 118  */
 119 void SIMIX_host_destroy(void *h)
 120 {
 121   smx_host_priv_t host = (smx_host_priv_t) h;
 122
 123   xbt_assert((host != NULL), "Invalid parameters");
 124
 125   /* Clean Simulator data */
 126   if (xbt_swag_size(host->process_list) != 0) {
 127     char *msg = xbt_strdup("Shutting down host, but it's not empty:");
 128     char *tmp;
 129     smx_process_t process = NULL;
 130
 131     xbt_swag_foreach(process, host->process_list) {
 132       tmp = bprintf("%s\n\t%s", msg, process->name);
 133       free(msg);
 134       msg = tmp;
 135     }
 136     SIMIX_display_process_status();
 137     THROWF(arg_error, 0, "%s", msg);
 138   }
 139   xbt_dynar_free(&host->auto_restart_processes);
 140   xbt_dynar_free(&host->boot_processes);
 141   xbt_swag_free(host->process_list);
 142
 143   /* Clean host structure */
 144   free(host);
 145   return;
 146 }
 147
 148 sg_host_t SIMIX_host_self(void)
 149 {
 150   smx_process_t process = SIMIX_process_self();
 151   return (process == NULL) ? NULL : SIMIX_process_get_host(process);
 152 }
 153
 154 /* needs to be public and without simcall because it is called
 155    by exceptions and logging events */
 156 const char* SIMIX_host_self_get_name(void)
 157 {
 158   sg_host_t host = SIMIX_host_self();
 159   if (host == NULL || SIMIX_process_self() == simix_global->maestro_process)
 160     return "";
 161
 162   return sg_host_get_name(host);
 163 }
 164
 165 void _SIMIX_host_free_process_arg(void *data)
 166 {
 167   smx_process_arg_t arg = *(smx_process_arg_t*)data;
 168   int i;
 169   for (i = 0; i < arg->argc; i++)
 170     xbt_free(arg->argv[i]);
 171   xbt_free(arg->argv);
 172   xbt_free(arg->name);
 173   xbt_free(arg);
 174 }
 175 /**
 176  * \brief Add a process to the list of the processes that the host will restart when it comes back
 177  * This function add a process to the list of the processes that will be restarted when the host comes
 178  * back. It is expected that this function is called when the host is down.
 179  * The processes will only be restarted once, meaning that you will have to register the process
 180  * again to restart the process again.
 181  */
 182 void SIMIX_host_add_auto_restart_process(sg_host_t host,
 183                                          const char *name,
 184                                          xbt_main_func_t code,
 185                                          void *data,
 186                                          const char *hostname,
 187                                          double kill_time,
 188                                          int argc, char **argv,
 189                                          xbt_dict_t properties,
 190                                          int auto_restart)
 191 {
 192   if (!sg_host_simix(host)->auto_restart_processes) {
 193     sg_host_simix(host)->auto_restart_processes = xbt_dynar_new(sizeof(smx_process_arg_t),_SIMIX_host_free_process_arg);
 194   }
 195   smx_process_arg_t arg = xbt_new(s_smx_process_arg_t,1);
 196   arg->name = xbt_strdup(name);
 197   arg->code = code;
 198   arg->data = data;
 199   arg->hostname = hostname;
 200   arg->kill_time = kill_time;
 201   arg->argc = argc;
 202
 203   arg->argv = xbt_new(char*,argc + 1);
 204
 205   int i;
 206   for (i = 0; i < argc; i++) {
 207     arg->argv[i] = xbt_strdup(argv[i]);
 208   }
 209   arg->argv[argc] = NULL;
 210
 211   arg->properties = properties;
 212   arg->auto_restart = auto_restart;
 213
 214   if( ! sg_host_is_on(host)
 215       && !xbt_dict_get_or_null(watched_hosts_lib,sg_host_get_name(host))){
 216     xbt_dict_set(watched_hosts_lib,sg_host_get_name(host),host,NULL);
 217     XBT_DEBUG("Have pushed host %s to watched_hosts_lib because state == SURF_RESOURCE_OFF",sg_host_get_name(host));
 218   }
 219   xbt_dynar_push_as(sg_host_simix(host)->auto_restart_processes,smx_process_arg_t,arg);
 220 }
 221 /**
 222  * \brief Restart the list of processes that have been registered to the host
 223  */
 224 void SIMIX_host_autorestart(sg_host_t host)
 225 {
 226   unsigned int cpt;
 227   smx_process_arg_t arg;
 228   xbt_dynar_t process_list = sg_host_simix(host)->auto_restart_processes;
 229   if (!process_list)
 230     return;
 231
 232   xbt_dynar_foreach (process_list, cpt, arg) {
 233
 234     XBT_DEBUG("Restarting Process %s(%s) right now", arg->argv[0], arg->hostname);
 235     if (simix_global->create_process_function) {
 236       simix_global->create_process_function(arg->argv[0],
 237                                             arg->code,
 238                                             NULL,
 239                                             arg->hostname,
 240                                             arg->kill_time,
 241                                             arg->argc,
 242                                             arg->argv,
 243                                             arg->properties,
 244                                             arg->auto_restart,
 245                                             NULL);
 246     } else {
 247       simcall_process_create(arg->argv[0],
 248                              (xbt_main_func_t) arg->code,
 249                              NULL,
 250                              arg->hostname,
 251                              arg->kill_time,
 252                              arg->argc,
 253                              arg->argv,
 254                              arg->properties,
 255                              arg->auto_restart);
 256
 257     }
 258     /* arg->argv is used by the process created above.  Hide it to
 259      * _SIMIX_host_free_process_arg() which is called by xbt_dynar_reset()
 260      * below. */
 261     arg->argc = 0;
 262     arg->argv = NULL;
 263   }
 264   xbt_dynar_reset(process_list);
 265 }
 266
 267 smx_synchro_t simcall_HANDLER_execution_start(smx_simcall_t simcall,
 268     const char* name, double flops_amount, double priority, double bound, unsigned long affinity_mask) {
 269   return SIMIX_execution_start(simcall->issuer, name,flops_amount,priority,bound,affinity_mask);
 270 }
 271 smx_synchro_t SIMIX_execution_start(smx_process_t issuer, const char *name,
 272      double flops_amount, double priority, double bound, unsigned long affinity_mask){
 273
 274   /* alloc structures and initialize */
 275   smx_synchro_t synchro = (smx_synchro_t) xbt_mallocator_get(simix_global->synchro_mallocator);
 276   synchro->type = SIMIX_SYNC_EXECUTE;
 277   synchro->name = xbt_strdup(name);
 278   synchro->state = SIMIX_RUNNING;
 279   synchro->execution.host = issuer->host;
 280   synchro->category = NULL;
 281
 282   /* set surf's action */
 283   if (!MC_is_active() && !MC_record_replay_is_active()) {
 284
 285     synchro->execution.surf_exec = issuer->host->pimpl_cpu->execution_start(flops_amount);
 286     synchro->execution.surf_exec->setData(synchro);
 287     synchro->execution.surf_exec->setPriority(priority);
 288
 289     if (bound != 0)
 290       static_cast<simgrid::surf::CpuAction*>(synchro->execution.surf_exec)
 291         ->setBound(bound);
 292
 293     if (affinity_mask != 0) {
 294       /* just a double check to confirm that this host is the host where this task is running. */
 295       xbt_assert(synchro->execution.host == issuer->host);
 296       static_cast<simgrid::surf::CpuAction*>(synchro->execution.surf_exec)
 297         ->setAffinity(issuer->host->pimpl_cpu, affinity_mask);
 298     }
 299   }
 300
 301   XBT_DEBUG("Create execute synchro %p: %s", synchro, synchro->name);
 302
 303   return synchro;
 304 }
 305
 306 smx_synchro_t SIMIX_execution_parallel_start(const char *name,
 307     int host_nb, sg_host_t *host_list,
 308     double *flops_amount, double *bytes_amount,
 309     double amount, double rate){
 310
 311   sg_host_t*host_list_cpy = NULL;
 312   int i;
 313
 314   /* alloc structures and initialize */
 315   smx_synchro_t synchro = (smx_synchro_t) xbt_mallocator_get(simix_global->synchro_mallocator);
 316   synchro->type = SIMIX_SYNC_PARALLEL_EXECUTE;
 317   synchro->name = xbt_strdup(name);
 318   synchro->state = SIMIX_RUNNING;
 319   synchro->execution.host = NULL; /* FIXME: do we need the list of hosts? */
 320   synchro->category = NULL;
 321
 322   /* set surf's synchro */
 323   host_list_cpy = xbt_new0(sg_host_t, host_nb);
 324   for (i = 0; i < host_nb; i++)
 325     host_list_cpy[i] = host_list[i];
 326
 327   /* Check that we are not mixing VMs and PMs in the parallel task */
 328   simgrid::surf::Host *host = host_list[0]->extension<simgrid::surf::Host>();
 329   bool is_a_vm = (nullptr != dynamic_cast<simgrid::surf::VirtualMachine*>(host));
 330   for (i = 1; i < host_nb; i++) {
 331     bool tmp_is_a_vm = (nullptr != dynamic_cast<simgrid::surf::VirtualMachine*>(host_list[i]->extension<simgrid::surf::Host>()));
 332     xbt_assert(is_a_vm == tmp_is_a_vm, "parallel_execute: mixing VMs and PMs is not supported (yet).");
 333   }
 334
 335   /* set surf's synchro */
 336   if (!MC_is_active() && !MC_record_replay_is_active()) {
 337     synchro->execution.surf_exec =
 338       surf_host_model->executeParallelTask(
 339           host_nb, host_list_cpy, flops_amount, bytes_amount, rate);
 340
 341     synchro->execution.surf_exec->setData(synchro);
 342   }
 343   XBT_DEBUG("Create parallel execute synchro %p", synchro);
 344
 345   return synchro;
 346 }
 347
 348 void SIMIX_execution_destroy(smx_synchro_t synchro)
 349 {
 350   XBT_DEBUG("Destroy synchro %p", synchro);
 351
 352   if (synchro->execution.surf_exec) {
 353     synchro->execution.surf_exec->unref();
 354     synchro->execution.surf_exec = NULL;
 355   }
 356   xbt_free(synchro->name);
 357   xbt_mallocator_release(simix_global->synchro_mallocator, synchro);
 358 }
 359
 360 void SIMIX_execution_cancel(smx_synchro_t synchro)
 361 {
 362   XBT_DEBUG("Cancel synchro %p", synchro);
 363
 364   if (synchro->execution.surf_exec)
 365     synchro->execution.surf_exec->cancel();
 366 }
 367
 368 double SIMIX_execution_get_remains(smx_synchro_t synchro)
 369 {
 370   double result = 0.0;
 371
 372   if (synchro->state == SIMIX_RUNNING)
 373     result = synchro->execution.surf_exec->getRemains();
 374
 375   return result;
 376 }
 377
 378 e_smx_state_t SIMIX_execution_get_state(smx_synchro_t synchro)
 379 {
 380   return synchro->state;
 381 }
 382
 383 void SIMIX_execution_set_priority(smx_synchro_t synchro, double priority)
 384 {
 385   if(synchro->execution.surf_exec)
 386     synchro->execution.surf_exec->setPriority(priority);
 387 }
 388
 389 void SIMIX_execution_set_bound(smx_synchro_t synchro, double bound)
 390 {
 391   if(synchro->execution.surf_exec)
 392     static_cast<simgrid::surf::CpuAction*>(synchro->execution.surf_exec)->setBound(bound);
 393 }
 394
 395 void SIMIX_execution_set_affinity(smx_synchro_t synchro, sg_host_t host, unsigned long mask)
 396 {
 397   xbt_assert(synchro->type == SIMIX_SYNC_EXECUTE);
 398
 399   if (synchro->execution.surf_exec) {
 400     /* just a double check to confirm that this host is the host where this task is running. */
 401     xbt_assert(synchro->execution.host == host);
 402     static_cast<simgrid::surf::CpuAction*>(synchro->execution.surf_exec)
 403       ->setAffinity(host->pimpl_cpu, mask);
 404   }
 405 }
 406
 407 void simcall_HANDLER_execution_wait(smx_simcall_t simcall, smx_synchro_t synchro)
 408 {
 409
 410   XBT_DEBUG("Wait for execution of synchro %p, state %d", synchro, (int)synchro->state);
 411
 412   /* Associate this simcall to the synchro */
 413   xbt_fifo_push(synchro->simcalls, simcall);
 414   simcall->issuer->waiting_synchro = synchro;
 415
 416   /* set surf's synchro */
 417   if (MC_is_active() || MC_record_replay_is_active()) {
 418     synchro->state = SIMIX_DONE;
 419     SIMIX_execution_finish(synchro);
 420     return;
 421   }
 422
 423   /* If the synchro is already finished then perform the error handling */
 424   if (synchro->state != SIMIX_RUNNING)
 425     SIMIX_execution_finish(synchro);
 426 }
 427
 428 void SIMIX_execution_suspend(smx_synchro_t synchro)
 429 {
 430   if(synchro->execution.surf_exec)
 431     synchro->execution.surf_exec->suspend();
 432 }
 433
 434 void SIMIX_execution_resume(smx_synchro_t synchro)
 435 {
 436   if(synchro->execution.surf_exec)
 437     synchro->execution.surf_exec->resume();
 438 }
 439
 440 void SIMIX_execution_finish(smx_synchro_t synchro)
 441 {
 442   xbt_fifo_item_t item;
 443   smx_simcall_t simcall;
 444
 445   xbt_fifo_foreach(synchro->simcalls, item, simcall, smx_simcall_t) {
 446
 447     switch (synchro->state) {
 448
 449       case SIMIX_DONE:
 450         /* do nothing, synchro done */
 451         XBT_DEBUG("SIMIX_execution_finished: execution successful");
 452         break;
 453
 454       case SIMIX_FAILED:
 455         XBT_DEBUG("SIMIX_execution_finished: host '%s' failed", sg_host_get_name(simcall->issuer->host));
 456         simcall->issuer->context->iwannadie = 1;
 457         SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
 458         break;
 459
 460       case SIMIX_CANCELED:
 461         XBT_DEBUG("SIMIX_execution_finished: execution canceled");
 462         SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Canceled");
 463         break;
 464
 465       default:
 466         xbt_die("Internal error in SIMIX_execution_finish: unexpected synchro state %d",
 467             (int)synchro->state);
 468     }
 469     /* check if the host is down */
 470     if (simcall->issuer->host->isOff()) {
 471       simcall->issuer->context->iwannadie = 1;
 472     }
 473
 474     simcall->issuer->waiting_synchro =    NULL;
 475     simcall_execution_wait__set__result(simcall, synchro->state);
 476     SIMIX_simcall_answer(simcall);
 477   }
 478
 479   /* We no longer need it */
 480   SIMIX_execution_destroy(synchro);
 481 }
 482
 483
 484 void SIMIX_post_host_execute(smx_synchro_t synchro)
 485 {
 486   if (synchro->type == SIMIX_SYNC_EXECUTE && /* FIMXE: handle resource failure
 487                                                * for parallel tasks too */
 488       synchro->execution.host->isOff()) {
 489     /* If the host running the synchro failed, notice it so that the asking
 490      * process can be killed if it runs on that host itself */
 491     synchro->state = SIMIX_FAILED;
 492   } else if (synchro->execution.surf_exec->getState() == SURF_ACTION_FAILED) {
 493     /* If the host running the synchro didn't fail, then the synchro was
 494      * canceled */
 495     synchro->state = SIMIX_CANCELED;
 496   } else {
 497     synchro->state = SIMIX_DONE;
 498   }
 499
 500   if (synchro->execution.surf_exec) {
 501     synchro->execution.surf_exec->unref();
 502     synchro->execution.surf_exec = NULL;
 503   }
 504
 505   /* If there are simcalls associated with the synchro, then answer them */
 506   if (xbt_fifo_size(synchro->simcalls)) {
 507     SIMIX_execution_finish(synchro);
 508   }
 509 }
 510
 511
 512 void SIMIX_set_category(smx_synchro_t synchro, const char *category)
 513 {
 514   if (synchro->state != SIMIX_RUNNING) return;
 515   if (synchro->type == SIMIX_SYNC_EXECUTE){
 516     synchro->execution.surf_exec->setCategory(category);
 517   }else if (synchro->type == SIMIX_SYNC_COMMUNICATE){
 518     synchro->comm.surf_comm->setCategory(category);
 519   }
 520 }