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