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       XBT_DEBUG("Booting Process %s(%s) right now",
  53         arg->name.c_str(), arg->hostname);
  54       if (simix_global->create_process_function) {
  55         simix_global->create_process_function(arg->name.c_str(),
  56                                               arg->code,
  57                                               NULL,
  58                                               arg->hostname,
  59                                               arg->kill_time,
  60                                               arg->args,
  61                                               arg->properties,
  62                                               arg->auto_restart,
  63                                               NULL);
  64       } else {
  65         simcall_process_create(arg->name.c_str(),
  66                                arg->code,
  67                                NULL,
  68                                arg->hostname,
  69                                arg->kill_time,
  70                                arg->args,
  71                                arg->properties,
  72                                arg->auto_restart);
  73       }
  74     }
  75   }
  76 }
  77
  78 /**
  79  * \brief Stop the host if it is on
  80  *
  81  */
  82 void SIMIX_host_off(sg_host_t h, smx_process_t issuer)
  83 {
  84   smx_host_priv_t host = sg_host_simix(h);
  85
  86   xbt_assert((host != NULL), "Invalid parameters");
  87
  88   if (h->isOn()) {
  89     simgrid::surf::HostImpl* surf_host = h->extension<simgrid::surf::HostImpl>();
  90     surf_host->turnOff();
  91
  92     /* Clean Simulator data */
  93     if (xbt_swag_size(host->process_list) != 0) {
  94       smx_process_t process = NULL;
  95       xbt_swag_foreach(process, host->process_list) {
  96         SIMIX_process_kill(process, issuer);
  97         XBT_DEBUG("Killing %s on %s by %s",
  98           process->name.c_str(),  sg_host_get_name(process->host),
  99           issuer->name.c_str());
 100       }
 101     }
 102   } else {
 103     XBT_INFO("Host %s is already off", h->name().c_str());
 104   }
 105 }
 106
 107 /**
 108  * \brief Internal function to destroy a SIMIX host.
 109  *
 110  * \param h the host to destroy (a sg_host_t)
 111  */
 112 void SIMIX_host_destroy(void *h)
 113 {
 114   smx_host_priv_t host = (smx_host_priv_t) h;
 115
 116   xbt_assert((host != NULL), "Invalid parameters");
 117
 118   /* Clean Simulator data */
 119   if (xbt_swag_size(host->process_list) != 0) {
 120     char *msg = xbt_strdup("Shutting down host, but it's not empty:");
 121     char *tmp;
 122     smx_process_t process = NULL;
 123
 124     xbt_swag_foreach(process, host->process_list) {
 125       tmp = bprintf("%s\n\t%s", msg, process->name.c_str());
 126       free(msg);
 127       msg = tmp;
 128     }
 129     SIMIX_display_process_status();
 130     THROWF(arg_error, 0, "%s", msg);
 131   }
 132   xbt_dynar_free(&host->auto_restart_processes);
 133   xbt_dynar_free(&host->boot_processes);
 134   xbt_swag_free(host->process_list);
 135
 136   /* Clean host structure */
 137   free(host);
 138   return;
 139 }
 140
 141 sg_host_t SIMIX_host_self(void)
 142 {
 143   smx_process_t process = SIMIX_process_self();
 144   return (process == NULL) ? NULL : SIMIX_process_get_host(process);
 145 }
 146
 147 /* needs to be public and without simcall because it is called
 148    by exceptions and logging events */
 149 const char* SIMIX_host_self_get_name(void)
 150 {
 151   sg_host_t host = SIMIX_host_self();
 152   if (host == NULL || SIMIX_process_self() == simix_global->maestro_process)
 153     return "";
 154
 155   return sg_host_get_name(host);
 156 }
 157
 158 void _SIMIX_host_free_process_arg(void *data)
 159 {
 160   smx_process_arg_t arg = *(smx_process_arg_t*)data;
 161   delete arg;
 162 }
 163 /**
 164  * \brief Add a process to the list of the processes that the host will restart when it comes back
 165  * This function add a process to the list of the processes that will be restarted when the host comes
 166  * back. It is expected that this function is called when the host is down.
 167  * The processes will only be restarted once, meaning that you will have to register the process
 168  * again to restart the process again.
 169  */
 170 void SIMIX_host_add_auto_restart_process(sg_host_t host,
 171                                          const char *name,
 172                                          xbt_main_func_t code,
 173                                          void *data,
 174                                          const char *hostname,
 175                                          double kill_time,
 176                                          int argc, char **argv,
 177                                          xbt_dict_t properties,
 178                                          int auto_restart)
 179 {
 180   if (!sg_host_simix(host)->auto_restart_processes) {
 181     sg_host_simix(host)->auto_restart_processes = xbt_dynar_new(sizeof(smx_process_arg_t),_SIMIX_host_free_process_arg);
 182   }
 183   smx_process_arg_t arg = new simgrid::simix::ProcessArg();
 184   arg->name = name;
 185   arg->code = code;
 186   arg->data = data;
 187   arg->hostname = hostname;
 188   arg->kill_time = kill_time;
 189   arg->args.assign(argc, argv);
 190   arg->properties = properties;
 191   arg->auto_restart = auto_restart;
 192
 193   if( host->isOff() && !xbt_dict_get_or_null(watched_hosts_lib,sg_host_get_name(host))){
 194     xbt_dict_set(watched_hosts_lib,sg_host_get_name(host),host,NULL);
 195     XBT_DEBUG("Push host %s to watched_hosts_lib because state == SURF_RESOURCE_OFF",sg_host_get_name(host));
 196   }
 197   xbt_dynar_push_as(sg_host_simix(host)->auto_restart_processes,smx_process_arg_t,arg);
 198 }
 199 /**
 200  * \brief Restart the list of processes that have been registered to the host
 201  */
 202 void SIMIX_host_autorestart(sg_host_t host)
 203 {
 204   unsigned int cpt;
 205   smx_process_arg_t arg;
 206   xbt_dynar_t process_list = sg_host_simix(host)->auto_restart_processes;
 207   if (!process_list)
 208     return;
 209
 210   xbt_dynar_foreach (process_list, cpt, arg) {
 211
 212     XBT_DEBUG("Restarting Process %s(%s) right now", arg->name.c_str(), arg->hostname);
 213     if (simix_global->create_process_function) {
 214       simix_global->create_process_function(arg->name.c_str(),
 215                                             arg->code,
 216                                             NULL,
 217                                             arg->hostname,
 218                                             arg->kill_time,
 219                                             arg->args,
 220                                             arg->properties,
 221                                             arg->auto_restart,
 222                                             NULL);
 223     } else {
 224       simcall_process_create(arg->name.c_str(),
 225                              (xbt_main_func_t) arg->code,
 226                              NULL,
 227                              arg->hostname,
 228                              arg->kill_time,
 229                              arg->args,
 230                              arg->properties,
 231                              arg->auto_restart);
 232     }
 233   }
 234   xbt_dynar_reset(process_list);
 235 }
 236
 237 smx_synchro_t simcall_HANDLER_execution_start(smx_simcall_t simcall,
 238     const char* name, double flops_amount, double priority, double bound, unsigned long affinity_mask) {
 239   return SIMIX_execution_start(simcall->issuer, name,flops_amount,priority,bound,affinity_mask);
 240 }
 241 smx_synchro_t SIMIX_execution_start(smx_process_t issuer, const char *name,
 242      double flops_amount, double priority, double bound, unsigned long affinity_mask){
 243
 244   /* alloc structures and initialize */
 245   simgrid::simix::Exec *exec = new simgrid::simix::Exec();
 246   exec->name = xbt_strdup(name);
 247   exec->state = SIMIX_RUNNING;
 248   exec->host = issuer->host;
 249
 250   /* set surf's action */
 251   if (!MC_is_active() && !MC_record_replay_is_active()) {
 252
 253     exec->surf_exec = issuer->host->pimpl_cpu->execution_start(flops_amount);
 254     exec->surf_exec->setData(exec);
 255     exec->surf_exec->setPriority(priority);
 256
 257     if (bound != 0)
 258       static_cast<simgrid::surf::CpuAction*>(exec->surf_exec)->setBound(bound);
 259
 260     if (affinity_mask != 0) {
 261       /* just a double check to confirm that this host is the host where this task is running. */
 262       xbt_assert(exec->host == issuer->host);
 263       static_cast<simgrid::surf::CpuAction*>(exec->surf_exec)
 264         ->setAffinity(issuer->host->pimpl_cpu, affinity_mask);
 265     }
 266   }
 267
 268   XBT_DEBUG("Create execute synchro %p: %s", exec, exec->name);
 269
 270   return exec;
 271 }
 272
 273 smx_synchro_t SIMIX_execution_parallel_start(const char *name,
 274     int host_nb, sg_host_t *host_list,
 275     double *flops_amount, double *bytes_amount,
 276     double amount, double rate){
 277
 278   sg_host_t *host_list_cpy = NULL;
 279   int i;
 280
 281   /* alloc structures and initialize */
 282   simgrid::simix::Exec *exec = new simgrid::simix::Exec();
 283   exec->name = xbt_strdup(name);
 284   exec->state = SIMIX_RUNNING;
 285   exec->host = nullptr; /* FIXME: do we need the list of hosts? */
 286
 287   /* set surf's synchro */
 288   host_list_cpy = xbt_new0(sg_host_t, host_nb);
 289   for (i = 0; i < host_nb; i++)
 290     host_list_cpy[i] = host_list[i];
 291
 292   /* Check that we are not mixing VMs and PMs in the parallel task */
 293   simgrid::surf::HostImpl *host = host_list[0]->extension<simgrid::surf::HostImpl>();
 294   bool is_a_vm = (nullptr != dynamic_cast<simgrid::surf::VirtualMachine*>(host));
 295   for (i = 1; i < host_nb; i++) {
 296     bool tmp_is_a_vm = (nullptr != dynamic_cast<simgrid::surf::VirtualMachine*>(host_list[i]->extension<simgrid::surf::HostImpl>()));
 297     xbt_assert(is_a_vm == tmp_is_a_vm, "parallel_execute: mixing VMs and PMs is not supported (yet).");
 298   }
 299
 300   /* set surf's synchro */
 301   if (!MC_is_active() && !MC_record_replay_is_active()) {
 302     exec->surf_exec = surf_host_model->executeParallelTask(host_nb, host_list_cpy, flops_amount, bytes_amount, rate);
 303     exec->surf_exec->setData(exec);
 304   }
 305   XBT_DEBUG("Create parallel execute synchro %p", exec);
 306
 307   return exec;
 308 }
 309
 310 void SIMIX_execution_cancel(smx_synchro_t synchro)
 311 {
 312   XBT_DEBUG("Cancel synchro %p", synchro);
 313   simgrid::simix::Exec *exec = static_cast<simgrid::simix::Exec *>(synchro);
 314
 315   if (exec->surf_exec)
 316     exec->surf_exec->cancel();
 317 }
 318
 319 void SIMIX_execution_set_priority(smx_synchro_t synchro, double priority)
 320 {
 321   simgrid::simix::Exec *exec = static_cast<simgrid::simix::Exec *>(synchro);
 322   if(exec->surf_exec)
 323     exec->surf_exec->setPriority(priority);
 324 }
 325
 326 void SIMIX_execution_set_bound(smx_synchro_t synchro, double bound)
 327 {
 328   simgrid::simix::Exec *exec = static_cast<simgrid::simix::Exec *>(synchro);
 329   if(exec->surf_exec)
 330     static_cast<simgrid::surf::CpuAction*>(exec->surf_exec)->setBound(bound);
 331 }
 332
 333 void SIMIX_execution_set_affinity(smx_synchro_t synchro, sg_host_t host, unsigned long mask)
 334 {
 335   simgrid::simix::Exec *exec = static_cast<simgrid::simix::Exec *>(synchro);
 336   if(exec->surf_exec) {
 337     /* just a double check to confirm that this host is the host where this task is running. */
 338     xbt_assert(exec->host == host);
 339     static_cast<simgrid::surf::CpuAction*>(exec->surf_exec)->setAffinity(host->pimpl_cpu, mask);
 340   }
 341 }
 342
 343 void simcall_HANDLER_execution_wait(smx_simcall_t simcall, smx_synchro_t synchro)
 344 {
 345   simgrid::simix::Exec *exec = static_cast<simgrid::simix::Exec *>(synchro);
 346   XBT_DEBUG("Wait for execution of synchro %p, state %d", synchro, (int)synchro->state);
 347
 348   /* Associate this simcall to the synchro */
 349   xbt_fifo_push(synchro->simcalls, simcall);
 350   simcall->issuer->waiting_synchro = synchro;
 351
 352   /* set surf's synchro */
 353   if (MC_is_active() || MC_record_replay_is_active()) {
 354     synchro->state = SIMIX_DONE;
 355     SIMIX_execution_finish(exec);
 356     return;
 357   }
 358
 359   /* If the synchro is already finished then perform the error handling */
 360   if (synchro->state != SIMIX_RUNNING)
 361     SIMIX_execution_finish(exec);
 362 }
 363
 364 void SIMIX_execution_finish(simgrid::simix::Exec *exec)
 365 {
 366   xbt_fifo_item_t item;
 367   smx_simcall_t simcall;
 368
 369   xbt_fifo_foreach(exec->simcalls, item, simcall, smx_simcall_t) {
 370
 371     switch (exec->state) {
 372
 373       case SIMIX_DONE:
 374         /* do nothing, synchro done */
 375         XBT_DEBUG("SIMIX_execution_finished: execution successful");
 376         break;
 377
 378       case SIMIX_FAILED:
 379         XBT_DEBUG("SIMIX_execution_finished: host '%s' failed", sg_host_get_name(simcall->issuer->host));
 380         simcall->issuer->context->iwannadie = 1;
 381         SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
 382         break;
 383
 384       case SIMIX_CANCELED:
 385         XBT_DEBUG("SIMIX_execution_finished: execution canceled");
 386         SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Canceled");
 387         break;
 388
 389       default:
 390         xbt_die("Internal error in SIMIX_execution_finish: unexpected synchro state %d",
 391             (int)exec->state);
 392     }
 393     /* Fail the process if the host is down */
 394     if (simcall->issuer->host->isOff())
 395       simcall->issuer->context->iwannadie = 1;
 396
 397     simcall->issuer->waiting_synchro = NULL;
 398     simcall_execution_wait__set__result(simcall, exec->state);
 399     SIMIX_simcall_answer(simcall);
 400   }
 401
 402   /* We no longer need it */
 403   exec->unref();
 404 }
 405
 406 void SIMIX_set_category(smx_synchro_t synchro, const char *category)
 407 {
 408   if (synchro->state != SIMIX_RUNNING) return;
 409
 410   simgrid::simix::Exec *exec = dynamic_cast<simgrid::simix::Exec *>(synchro);
 411   if (exec != nullptr) {
 412     exec->surf_exec->setCategory(category);
 413     return;
 414   }
 415
 416   simgrid::simix::Comm *comm = dynamic_cast<simgrid::simix::Comm *>(synchro);
 417   if (comm != nullptr) {
 418     comm->surf_comm->setCategory(category);
 419     return;
 420   }
 421 }