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