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/ex.hpp>
   9 #include "xbt/sysdep.h"
  10 #include "mc/mc.h"
  11 #include "src/mc/mc_replay.h"
  12 #include "src/surf/virtual_machine.hpp"
  13 #include "src/surf/HostImpl.hpp"
  14
  15 #include "src/simix/SynchroExec.hpp"
  16 #include "src/simix/SynchroComm.hpp"
  17
  18 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_host, simix, "SIMIX hosts");
  19
  20 /** @brief Internal function to create a SIMIX host. */
  21 void SIMIX_host_create(sg_host_t host) // FIXME: braindead prototype. Take sg_host as parameter
  22 {
  23   smx_host_priv_t smx_host = xbt_new0(s_smx_host_priv_t, 1);
  24
  25   /* Host structure */
  26   simgrid::simix::Process proc;
  27   smx_host->process_list = xbt_swag_new(xbt_swag_offset(proc, host_proc_hookup));
  28
  29   /* Update global variables */
  30   sg_host_simix_set(host, smx_host);
  31 }
  32
  33 /** @brief Start the host if it is off */
  34 void SIMIX_host_on(sg_host_t h)
  35 {
  36   smx_host_priv_t host = sg_host_simix(h);
  37
  38   xbt_assert((host != nullptr), "Invalid parameters");
  39
  40   if (h->isOff()) {
  41     simgrid::surf::HostImpl* surf_host = h->extension<simgrid::surf::HostImpl>();
  42     surf_host->turnOn();
  43
  44     unsigned int cpt;
  45     smx_process_arg_t arg;
  46     xbt_dynar_foreach(host->boot_processes,cpt,arg) {
  47       XBT_DEBUG("Booting Process %s(%s) right now",
  48         arg->name.c_str(), arg->hostname);
  49       if (simix_global->create_process_function) {
  50         simix_global->create_process_function(arg->name.c_str(),
  51                                               arg->code,
  52                                               nullptr,
  53                                               arg->hostname,
  54                                               arg->kill_time,
  55                                               arg->properties,
  56                                               arg->auto_restart,
  57                                               nullptr);
  58       } else {
  59         simcall_process_create(arg->name.c_str(),
  60                                arg->code,
  61                                nullptr,
  62                                arg->hostname,
  63                                arg->kill_time,
  64                                arg->properties,
  65                                arg->auto_restart);
  66       }
  67     }
  68   }
  69 }
  70
  71 /** @brief Stop the host if it is on */
  72 void SIMIX_host_off(sg_host_t h, smx_process_t issuer)
  73 {
  74   smx_host_priv_t host = sg_host_simix(h);
  75
  76   xbt_assert((host != nullptr), "Invalid parameters");
  77
  78   if (h->isOn()) {
  79     simgrid::surf::HostImpl* surf_host = h->extension<simgrid::surf::HostImpl>();
  80     surf_host->turnOff();
  81
  82     /* Clean Simulator data */
  83     if (xbt_swag_size(host->process_list) != 0) {
  84       smx_process_t process = nullptr;
  85       xbt_swag_foreach(process, host->process_list) {
  86         SIMIX_process_kill(process, issuer);
  87         XBT_DEBUG("Killing %s on %s by %s",
  88           process->name.c_str(),  sg_host_get_name(process->host),
  89           issuer->name.c_str());
  90       }
  91     }
  92   } else {
  93     XBT_INFO("Host %s is already off", h->name().c_str());
  94   }
  95 }
  96
  97 /**
  98  * \brief Internal function to destroy a SIMIX host.
  99  *
 100  * \param h the host to destroy (a sg_host_t)
 101  */
 102 void SIMIX_host_destroy(void *h)
 103 {
 104   smx_host_priv_t host = (smx_host_priv_t) h;
 105
 106   xbt_assert((host != nullptr), "Invalid parameters");
 107
 108   /* Clean Simulator data */
 109   if (xbt_swag_size(host->process_list) != 0) {
 110     char *msg = xbt_strdup("Shutting down host, but it's not empty:");
 111     char *tmp;
 112     smx_process_t process = nullptr;
 113
 114     xbt_swag_foreach(process, host->process_list) {
 115       tmp = bprintf("%s\n\t%s", msg, process->name.c_str());
 116       free(msg);
 117       msg = tmp;
 118     }
 119     SIMIX_display_process_status();
 120     THROWF(arg_error, 0, "%s", msg);
 121   }
 122   xbt_dynar_free(&host->auto_restart_processes);
 123   xbt_dynar_free(&host->boot_processes);
 124   xbt_swag_free(host->process_list);
 125
 126   /* Clean host structure */
 127   free(host);
 128   return;
 129 }
 130
 131 sg_host_t SIMIX_host_self(void)
 132 {
 133   smx_process_t process = SIMIX_process_self();
 134   return (process == nullptr) ? nullptr : SIMIX_process_get_host(process);
 135 }
 136
 137 /* needs to be public and without simcall for exceptions and logging events */
 138 const char* SIMIX_host_self_get_name(void)
 139 {
 140   sg_host_t host = SIMIX_host_self();
 141   if (host == nullptr || SIMIX_process_self() == simix_global->maestro_process)
 142     return "";
 143
 144   return sg_host_get_name(host);
 145 }
 146
 147 void _SIMIX_host_free_process_arg(void *data)
 148 {
 149   smx_process_arg_t arg = *(smx_process_arg_t*)data;
 150   delete arg;
 151 }
 152 /**
 153  * \brief Add a process to the list of the processes that the host will restart when it comes back
 154  * This function add a process to the list of the processes that will be restarted when the host comes
 155  * back. It is expected that this function is called when the host is down.
 156  * The processes will only be restarted once, meaning that you will have to register the process
 157  * again to restart the process again.
 158  */
 159 void SIMIX_host_add_auto_restart_process(
 160   sg_host_t host, const char *name, std::function<void()> code,
 161   void* data, const char *hostname, double kill_time,
 162   xbt_dict_t properties, int auto_restart)
 163 {
 164   if (!sg_host_simix(host)->auto_restart_processes) {
 165     sg_host_simix(host)->auto_restart_processes = xbt_dynar_new(sizeof(smx_process_arg_t),_SIMIX_host_free_process_arg);
 166   }
 167   smx_process_arg_t arg = new simgrid::simix::ProcessArg();
 168   arg->name = name;
 169   arg->code = std::move(code);
 170   arg->data = data;
 171   arg->hostname = hostname;
 172   arg->kill_time = kill_time;
 173   arg->properties = properties;
 174   arg->auto_restart = auto_restart;
 175
 176   if( host->isOff() && !xbt_dict_get_or_null(watched_hosts_lib,sg_host_get_name(host))){
 177     xbt_dict_set(watched_hosts_lib,sg_host_get_name(host),host,nullptr);
 178     XBT_DEBUG("Push host %s to watched_hosts_lib because state == SURF_RESOURCE_OFF",sg_host_get_name(host));
 179   }
 180   xbt_dynar_push_as(sg_host_simix(host)->auto_restart_processes,smx_process_arg_t,arg);
 181 }
 182 /** @brief Restart the list of processes that have been registered to the host */
 183 void SIMIX_host_autorestart(sg_host_t host)
 184 {
 185   unsigned int cpt;
 186   smx_process_arg_t arg;
 187   xbt_dynar_t process_list = sg_host_simix(host)->auto_restart_processes;
 188   if (!process_list)
 189     return;
 190
 191   xbt_dynar_foreach (process_list, cpt, arg) {
 192
 193     XBT_DEBUG("Restarting Process %s(%s) right now", arg->name.c_str(), arg->hostname);
 194     if (simix_global->create_process_function) {
 195       simix_global->create_process_function(arg->name.c_str(),
 196                                             arg->code,
 197                                             nullptr,
 198                                             arg->hostname,
 199                                             arg->kill_time,
 200                                             arg->properties,
 201                                             arg->auto_restart,
 202                                             nullptr);
 203     } else {
 204       simcall_process_create(arg->name.c_str(),
 205                              arg->code,
 206                              nullptr,
 207                              arg->hostname,
 208                              arg->kill_time,
 209                              arg->properties,
 210                              arg->auto_restart);
 211     }
 212   }
 213   xbt_dynar_reset(process_list);
 214 }
 215
 216 smx_synchro_t simcall_HANDLER_execution_start(smx_simcall_t simcall,
 217     const char* name, double flops_amount, double priority, double bound, unsigned long affinity_mask) {
 218   return SIMIX_execution_start(simcall->issuer, name,flops_amount,priority,bound,affinity_mask);
 219 }
 220 smx_synchro_t SIMIX_execution_start(smx_process_t issuer, const char *name,
 221      double flops_amount, double priority, double bound, unsigned long affinity_mask){
 222
 223   /* alloc structures and initialize */
 224   simgrid::simix::Exec *exec = new simgrid::simix::Exec(name, issuer->host);
 225
 226   /* set surf's action */
 227   if (!MC_is_active() && !MC_record_replay_is_active()) {
 228
 229     exec->surf_exec = issuer->host->pimpl_cpu->execution_start(flops_amount);
 230     exec->surf_exec->setData(exec);
 231     exec->surf_exec->setPriority(priority);
 232
 233     if (bound != 0)
 234       static_cast<simgrid::surf::CpuAction*>(exec->surf_exec)->setBound(bound);
 235
 236     if (affinity_mask != 0) {
 237       /* just a double check to confirm that this host is the host where this task is running. */
 238       xbt_assert(exec->host == issuer->host);
 239       static_cast<simgrid::surf::CpuAction*>(exec->surf_exec)
 240         ->setAffinity(issuer->host->pimpl_cpu, affinity_mask);
 241     }
 242   }
 243
 244   XBT_DEBUG("Create execute synchro %p: %s", exec, exec->name.c_str());
 245
 246   return exec;
 247 }
 248
 249 smx_synchro_t SIMIX_execution_parallel_start(const char *name,
 250     int host_nb, sg_host_t *host_list,
 251     double *flops_amount, double *bytes_amount,
 252     double amount, double rate){
 253
 254   sg_host_t *host_list_cpy = nullptr;
 255   int i;
 256
 257   /* alloc structures and initialize */
 258   simgrid::simix::Exec *exec = new simgrid::simix::Exec(name, nullptr);
 259
 260   /* set surf's synchro */
 261   host_list_cpy = xbt_new0(sg_host_t, host_nb);
 262   for (i = 0; i < host_nb; i++)
 263     host_list_cpy[i] = host_list[i];
 264
 265   /* Check that we are not mixing VMs and PMs in the parallel task */
 266   simgrid::surf::HostImpl *host = host_list[0]->extension<simgrid::surf::HostImpl>();
 267   bool is_a_vm = (nullptr != dynamic_cast<simgrid::surf::VirtualMachine*>(host));
 268   for (i = 1; i < host_nb; i++) {
 269     bool tmp_is_a_vm = (nullptr != dynamic_cast<simgrid::surf::VirtualMachine*>(host_list[i]->extension<simgrid::surf::HostImpl>()));
 270     xbt_assert(is_a_vm == tmp_is_a_vm, "parallel_execute: mixing VMs and PMs is not supported (yet).");
 271   }
 272
 273   /* set surf's synchro */
 274   if (!MC_is_active() && !MC_record_replay_is_active()) {
 275     exec->surf_exec = surf_host_model->executeParallelTask(host_nb, host_list_cpy, flops_amount, bytes_amount, rate);
 276     exec->surf_exec->setData(exec);
 277   }
 278   XBT_DEBUG("Create parallel execute synchro %p", exec);
 279
 280   return exec;
 281 }
 282
 283 void SIMIX_execution_cancel(smx_synchro_t synchro)
 284 {
 285   XBT_DEBUG("Cancel synchro %p", synchro);
 286   simgrid::simix::Exec *exec = static_cast<simgrid::simix::Exec *>(synchro);
 287
 288   if (exec->surf_exec)
 289     exec->surf_exec->cancel();
 290 }
 291
 292 void SIMIX_execution_set_priority(smx_synchro_t synchro, double priority)
 293 {
 294   simgrid::simix::Exec *exec = static_cast<simgrid::simix::Exec *>(synchro);
 295   if(exec->surf_exec)
 296     exec->surf_exec->setPriority(priority);
 297 }
 298
 299 void SIMIX_execution_set_bound(smx_synchro_t synchro, double bound)
 300 {
 301   simgrid::simix::Exec *exec = static_cast<simgrid::simix::Exec *>(synchro);
 302   if(exec->surf_exec)
 303     static_cast<simgrid::surf::CpuAction*>(exec->surf_exec)->setBound(bound);
 304 }
 305
 306 void SIMIX_execution_set_affinity(smx_synchro_t synchro, sg_host_t host, unsigned long mask)
 307 {
 308   simgrid::simix::Exec *exec = static_cast<simgrid::simix::Exec *>(synchro);
 309   if(exec->surf_exec) {
 310     /* just a double check to confirm that this host is the host where this task is running. */
 311     xbt_assert(exec->host == host);
 312     static_cast<simgrid::surf::CpuAction*>(exec->surf_exec)->setAffinity(host->pimpl_cpu, mask);
 313   }
 314 }
 315
 316 void simcall_HANDLER_execution_wait(smx_simcall_t simcall, smx_synchro_t synchro)
 317 {
 318   simgrid::simix::Exec *exec = static_cast<simgrid::simix::Exec *>(synchro);
 319   XBT_DEBUG("Wait for execution of synchro %p, state %d", synchro, (int)synchro->state);
 320
 321   /* Associate this simcall to the synchro */
 322   synchro->simcalls.push_back(simcall);
 323   simcall->issuer->waiting_synchro = synchro;
 324
 325   /* set surf's synchro */
 326   if (MC_is_active() || MC_record_replay_is_active()) {
 327     synchro->state = SIMIX_DONE;
 328     SIMIX_execution_finish(exec);
 329     return;
 330   }
 331
 332   /* If the synchro is already finished then perform the error handling */
 333   if (synchro->state != SIMIX_RUNNING)
 334     SIMIX_execution_finish(exec);
 335 }
 336
 337 void SIMIX_execution_finish(simgrid::simix::Exec *exec)
 338 {
 339   for (smx_simcall_t simcall : exec->simcalls) {
 340     switch (exec->state) {
 341
 342       case SIMIX_DONE:
 343         /* do nothing, synchro done */
 344         XBT_DEBUG("SIMIX_execution_finished: execution successful");
 345         break;
 346
 347       case SIMIX_FAILED:
 348         XBT_DEBUG("SIMIX_execution_finished: host '%s' failed", sg_host_get_name(simcall->issuer->host));
 349         simcall->issuer->context->iwannadie = 1;
 350         SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
 351         break;
 352
 353       case SIMIX_CANCELED:
 354         XBT_DEBUG("SIMIX_execution_finished: execution canceled");
 355         SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Canceled");
 356         break;
 357
 358       default:
 359         xbt_die("Internal error in SIMIX_execution_finish: unexpected synchro state %d",
 360             (int)exec->state);
 361     }
 362     /* Fail the process if the host is down */
 363     if (simcall->issuer->host->isOff())
 364       simcall->issuer->context->iwannadie = 1;
 365
 366     simcall->issuer->waiting_synchro = nullptr;
 367     simcall_execution_wait__set__result(simcall, exec->state);
 368     SIMIX_simcall_answer(simcall);
 369   }
 370
 371   /* We no longer need it */
 372   exec->unref();
 373 }
 374
 375 void SIMIX_set_category(smx_synchro_t synchro, const char *category)
 376 {
 377   if (synchro->state != SIMIX_RUNNING) return;
 378
 379   simgrid::simix::Exec *exec = dynamic_cast<simgrid::simix::Exec *>(synchro);
 380   if (exec != nullptr) {
 381     exec->surf_exec->setCategory(category);
 382     return;
 383   }
 384
 385   simgrid::simix::Comm *comm = dynamic_cast<simgrid::simix::Comm *>(synchro);
 386   if (comm != nullptr) {
 387     comm->surf_comm->setCategory(category);
 388     return;
 389   }
 390 }