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 /**
  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
  27   /* Host structure */
  28   simgrid::simix::Process proc;
  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   for (int i = 0; i < arg->argc; i++)
 166     xbt_free(arg->argv[i]);
 167   xbt_free(arg->argv);
 168   xbt_free(arg->name);
 169   delete arg;
 170 }
 171 /**
 172  * \brief Add a process to the list of the processes that the host will restart when it comes back
 173  * This function add a process to the list of the processes that will be restarted when the host comes
 174  * back. It is expected that this function is called when the host is down.
 175  * The processes will only be restarted once, meaning that you will have to register the process
 176  * again to restart the process again.
 177  */
 178 void SIMIX_host_add_auto_restart_process(sg_host_t host,
 179                                          const char *name,
 180                                          xbt_main_func_t code,
 181                                          void *data,
 182                                          const char *hostname,
 183                                          double kill_time,
 184                                          int argc, char **argv,
 185                                          xbt_dict_t properties,
 186                                          int auto_restart)
 187 {
 188   if (!sg_host_simix(host)->auto_restart_processes) {
 189     sg_host_simix(host)->auto_restart_processes = xbt_dynar_new(sizeof(smx_process_arg_t),_SIMIX_host_free_process_arg);
 190   }
 191   smx_process_arg_t arg = new s_smx_process_arg_t();
 192   arg->name = xbt_strdup(name);
 193   arg->code = code;
 194   arg->data = data;
 195   arg->hostname = hostname;
 196   arg->kill_time = kill_time;
 197   arg->argc = argc;
 198
 199   arg->argv = xbt_new(char*,argc + 1);
 200
 201   for (int i = 0; i < argc; i++)
 202     arg->argv[i] = xbt_strdup(argv[i]);
 203   arg->argv[argc] = NULL;
 204
 205   arg->properties = properties;
 206   arg->auto_restart = auto_restart;
 207
 208   if( host->isOff() && !xbt_dict_get_or_null(watched_hosts_lib,sg_host_get_name(host))){
 209     xbt_dict_set(watched_hosts_lib,sg_host_get_name(host),host,NULL);
 210     XBT_DEBUG("Push host %s to watched_hosts_lib because state == SURF_RESOURCE_OFF",sg_host_get_name(host));
 211   }
 212   xbt_dynar_push_as(sg_host_simix(host)->auto_restart_processes,smx_process_arg_t,arg);
 213 }
 214 /**
 215  * \brief Restart the list of processes that have been registered to the host
 216  */
 217 void SIMIX_host_autorestart(sg_host_t host)
 218 {
 219   unsigned int cpt;
 220   smx_process_arg_t arg;
 221   xbt_dynar_t process_list = sg_host_simix(host)->auto_restart_processes;
 222   if (!process_list)
 223     return;
 224
 225   xbt_dynar_foreach (process_list, cpt, arg) {
 226
 227     XBT_DEBUG("Restarting Process %s(%s) right now", arg->argv[0], arg->hostname);
 228     if (simix_global->create_process_function) {
 229       simix_global->create_process_function(arg->argv[0],
 230                                             arg->code,
 231                                             NULL,
 232                                             arg->hostname,
 233                                             arg->kill_time,
 234                                             arg->argc,
 235                                             arg->argv,
 236                                             arg->properties,
 237                                             arg->auto_restart,
 238                                             NULL);
 239     } else {
 240       simcall_process_create(arg->argv[0],
 241                              (xbt_main_func_t) 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
 250     }
 251     /* arg->argv is used by the process created above.  Hide it to
 252      * _SIMIX_host_free_process_arg() which is called by xbt_dynar_reset()
 253      * below. */
 254     arg->argc = 0;
 255     arg->argv = NULL;
 256   }
 257   xbt_dynar_reset(process_list);
 258 }
 259
 260 smx_synchro_t simcall_HANDLER_execution_start(smx_simcall_t simcall,
 261     const char* name, double flops_amount, double priority, double bound, unsigned long affinity_mask) {
 262   return SIMIX_execution_start(simcall->issuer, name,flops_amount,priority,bound,affinity_mask);
 263 }
 264 smx_synchro_t SIMIX_execution_start(smx_process_t issuer, const char *name,
 265      double flops_amount, double priority, double bound, unsigned long affinity_mask){
 266
 267   /* alloc structures and initialize */
 268   simgrid::simix::Exec *exec = new simgrid::simix::Exec();
 269   exec->name = xbt_strdup(name);
 270   exec->state = SIMIX_RUNNING;
 271   exec->host = issuer->host;
 272
 273   /* set surf's action */
 274   if (!MC_is_active() && !MC_record_replay_is_active()) {
 275
 276     exec->surf_exec = issuer->host->pimpl_cpu->execution_start(flops_amount);
 277     exec->surf_exec->setData(exec);
 278     exec->surf_exec->setPriority(priority);
 279
 280     if (bound != 0)
 281       static_cast<simgrid::surf::CpuAction*>(exec->surf_exec)->setBound(bound);
 282
 283     if (affinity_mask != 0) {
 284       /* just a double check to confirm that this host is the host where this task is running. */
 285       xbt_assert(exec->host == issuer->host);
 286       static_cast<simgrid::surf::CpuAction*>(exec->surf_exec)
 287         ->setAffinity(issuer->host->pimpl_cpu, affinity_mask);
 288     }
 289   }
 290
 291   XBT_DEBUG("Create execute synchro %p: %s", exec, exec->name);
 292
 293   return exec;
 294 }
 295
 296 smx_synchro_t SIMIX_execution_parallel_start(const char *name,
 297     int host_nb, sg_host_t *host_list,
 298     double *flops_amount, double *bytes_amount,
 299     double amount, double rate){
 300
 301   sg_host_t *host_list_cpy = NULL;
 302   int i;
 303
 304   /* alloc structures and initialize */
 305   simgrid::simix::Exec *exec = new simgrid::simix::Exec();
 306   exec->name = xbt_strdup(name);
 307   exec->state = SIMIX_RUNNING;
 308   exec->host = nullptr; /* FIXME: do we need the list of hosts? */
 309
 310   /* set surf's synchro */
 311   host_list_cpy = xbt_new0(sg_host_t, host_nb);
 312   for (i = 0; i < host_nb; i++)
 313     host_list_cpy[i] = host_list[i];
 314
 315   /* Check that we are not mixing VMs and PMs in the parallel task */
 316   simgrid::surf::HostImpl *host = host_list[0]->extension<simgrid::surf::HostImpl>();
 317   bool is_a_vm = (nullptr != dynamic_cast<simgrid::surf::VirtualMachine*>(host));
 318   for (i = 1; i < host_nb; i++) {
 319     bool tmp_is_a_vm = (nullptr != dynamic_cast<simgrid::surf::VirtualMachine*>(host_list[i]->extension<simgrid::surf::HostImpl>()));
 320     xbt_assert(is_a_vm == tmp_is_a_vm, "parallel_execute: mixing VMs and PMs is not supported (yet).");
 321   }
 322
 323   /* set surf's synchro */
 324   if (!MC_is_active() && !MC_record_replay_is_active()) {
 325     exec->surf_exec = surf_host_model->executeParallelTask(host_nb, host_list_cpy, flops_amount, bytes_amount, rate);
 326     exec->surf_exec->setData(exec);
 327   }
 328   XBT_DEBUG("Create parallel execute synchro %p", exec);
 329
 330   return exec;
 331 }
 332
 333 void SIMIX_execution_cancel(smx_synchro_t synchro)
 334 {
 335   XBT_DEBUG("Cancel synchro %p", synchro);
 336   simgrid::simix::Exec *exec = static_cast<simgrid::simix::Exec *>(synchro);
 337
 338   if (exec->surf_exec)
 339     exec->surf_exec->cancel();
 340 }
 341
 342 void SIMIX_execution_set_priority(smx_synchro_t synchro, double priority)
 343 {
 344   simgrid::simix::Exec *exec = static_cast<simgrid::simix::Exec *>(synchro);
 345   if(exec->surf_exec)
 346     exec->surf_exec->setPriority(priority);
 347 }
 348
 349 void SIMIX_execution_set_bound(smx_synchro_t synchro, double bound)
 350 {
 351   simgrid::simix::Exec *exec = static_cast<simgrid::simix::Exec *>(synchro);
 352   if(exec->surf_exec)
 353     static_cast<simgrid::surf::CpuAction*>(exec->surf_exec)->setBound(bound);
 354 }
 355
 356 void SIMIX_execution_set_affinity(smx_synchro_t synchro, sg_host_t host, unsigned long mask)
 357 {
 358   simgrid::simix::Exec *exec = static_cast<simgrid::simix::Exec *>(synchro);
 359   if(exec->surf_exec) {
 360     /* just a double check to confirm that this host is the host where this task is running. */
 361     xbt_assert(exec->host == host);
 362     static_cast<simgrid::surf::CpuAction*>(exec->surf_exec)->setAffinity(host->pimpl_cpu, mask);
 363   }
 364 }
 365
 366 void simcall_HANDLER_execution_wait(smx_simcall_t simcall, smx_synchro_t synchro)
 367 {
 368   simgrid::simix::Exec *exec = static_cast<simgrid::simix::Exec *>(synchro);
 369   XBT_DEBUG("Wait for execution of synchro %p, state %d", synchro, (int)synchro->state);
 370
 371   /* Associate this simcall to the synchro */
 372   xbt_fifo_push(synchro->simcalls, simcall);
 373   simcall->issuer->waiting_synchro = synchro;
 374
 375   /* set surf's synchro */
 376   if (MC_is_active() || MC_record_replay_is_active()) {
 377     synchro->state = SIMIX_DONE;
 378     SIMIX_execution_finish(exec);
 379     return;
 380   }
 381
 382   /* If the synchro is already finished then perform the error handling */
 383   if (synchro->state != SIMIX_RUNNING)
 384     SIMIX_execution_finish(exec);
 385 }
 386
 387 void SIMIX_execution_finish(simgrid::simix::Exec *exec)
 388 {
 389   xbt_fifo_item_t item;
 390   smx_simcall_t simcall;
 391
 392   xbt_fifo_foreach(exec->simcalls, item, simcall, smx_simcall_t) {
 393
 394     switch (exec->state) {
 395
 396       case SIMIX_DONE:
 397         /* do nothing, synchro done */
 398         XBT_DEBUG("SIMIX_execution_finished: execution successful");
 399         break;
 400
 401       case SIMIX_FAILED:
 402         XBT_DEBUG("SIMIX_execution_finished: host '%s' failed", sg_host_get_name(simcall->issuer->host));
 403         simcall->issuer->context->iwannadie = 1;
 404         SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
 405         break;
 406
 407       case SIMIX_CANCELED:
 408         XBT_DEBUG("SIMIX_execution_finished: execution canceled");
 409         SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Canceled");
 410         break;
 411
 412       default:
 413         xbt_die("Internal error in SIMIX_execution_finish: unexpected synchro state %d",
 414             (int)exec->state);
 415     }
 416     /* Fail the process if the host is down */
 417     if (simcall->issuer->host->isOff())
 418       simcall->issuer->context->iwannadie = 1;
 419
 420     simcall->issuer->waiting_synchro = NULL;
 421     simcall_execution_wait__set__result(simcall, exec->state);
 422     SIMIX_simcall_answer(simcall);
 423   }
 424
 425   /* We no longer need it */
 426   exec->unref();
 427 }
 428
 429 void SIMIX_set_category(smx_synchro_t synchro, const char *category)
 430 {
 431   if (synchro->state != SIMIX_RUNNING) return;
 432
 433   simgrid::simix::Exec *exec = dynamic_cast<simgrid::simix::Exec *>(synchro);
 434   if (exec != nullptr) {
 435     exec->surf_exec->setCategory(category);
 436     return;
 437   }
 438
 439   simgrid::simix::Comm *comm = dynamic_cast<simgrid::simix::Comm *>(synchro);
 440   if (comm != nullptr) {
 441     comm->surf_comm->setCategory(category);
 442     return;
 443   }
 444 }