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