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/kernel/activity/SynchroExec.hpp"
  16 #include "src/kernel/activity/SynchroComm.hpp"
  17
  18 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_host, simix, "SIMIX hosts");
  19
  20 namespace simgrid {
  21   namespace simix {
  22     simgrid::xbt::Extension<simgrid::s4u::Host, Host> Host::EXTENSION_ID;
  23
  24     Host::Host()
  25     {
  26       if (!EXTENSION_ID.valid())
  27         EXTENSION_ID = simgrid::s4u::Host::extension_create<simgrid::simix::Host>();
  28
  29       simgrid::simix::ActorImpl act;
  30       process_list = xbt_swag_new(xbt_swag_offset(act, host_proc_hookup));
  31     }
  32
  33     Host::~Host()
  34     {
  35       /* Clean Simulator data */
  36       if (xbt_swag_size(process_list) != 0) {
  37         char *msg = xbt_strdup("Shutting down host, but it's not empty:");
  38         char *tmp;
  39         smx_actor_t process = nullptr;
  40
  41         xbt_swag_foreach(process, process_list) {
  42           tmp = bprintf("%s\n\t%s", msg, process->name.c_str());
  43           free(msg);
  44           msg = tmp;
  45         }
  46         SIMIX_display_process_status();
  47         THROWF(arg_error, 0, "%s", msg);
  48       }
  49       xbt_dynar_free(&auto_restart_processes);
  50       xbt_dynar_free(&boot_processes);
  51       xbt_swag_free(process_list);
  52     }
  53   }
  54 }
  55
  56 /** @brief Internal function to create a SIMIX host. */
  57 void SIMIX_host_create(sg_host_t host)
  58 {
  59   sg_host_simix_set(host, new simgrid::simix::Host());
  60 }
  61
  62 /** @brief Start the host if it is off */
  63 void SIMIX_host_on(sg_host_t h)
  64 {
  65   smx_host_priv_t host = sg_host_simix(h);
  66
  67   xbt_assert((host != nullptr), "Invalid parameters");
  68
  69   if (h->isOff()) {
  70     simgrid::surf::HostImpl* surf_host = h->extension<simgrid::surf::HostImpl>();
  71     surf_host->turnOn();
  72
  73     unsigned int cpt;
  74     smx_process_arg_t arg;
  75     xbt_dynar_foreach(host->boot_processes,cpt,arg) {
  76       XBT_DEBUG("Booting Process %s(%s) right now",
  77         arg->name.c_str(), arg->hostname);
  78       if (simix_global->create_process_function) {
  79         simix_global->create_process_function(arg->name.c_str(),
  80                                               arg->code,
  81                                               nullptr,
  82                                               arg->hostname,
  83                                               arg->kill_time,
  84                                               arg->properties,
  85                                               arg->auto_restart,
  86                                               nullptr);
  87       } else {
  88         simcall_process_create(arg->name.c_str(),
  89                                arg->code,
  90                                nullptr,
  91                                arg->hostname,
  92                                arg->kill_time,
  93                                arg->properties,
  94                                arg->auto_restart);
  95       }
  96     }
  97   }
  98 }
  99
 100 /** @brief Stop the host if it is on */
 101 void SIMIX_host_off(sg_host_t h, smx_actor_t issuer)
 102 {
 103   smx_host_priv_t host = sg_host_simix(h);
 104
 105   xbt_assert((host != nullptr), "Invalid parameters");
 106
 107   if (h->isOn()) {
 108     simgrid::surf::HostImpl* surf_host = h->extension<simgrid::surf::HostImpl>();
 109     surf_host->turnOff();
 110
 111     /* Clean Simulator data */
 112     if (xbt_swag_size(host->process_list) != 0) {
 113       smx_actor_t process = nullptr;
 114       xbt_swag_foreach(process, host->process_list) {
 115         SIMIX_process_kill(process, issuer);
 116         XBT_DEBUG("Killing %s on %s by %s",
 117           process->name.c_str(),  sg_host_get_name(process->host),
 118           issuer->name.c_str());
 119       }
 120     }
 121   } else {
 122     XBT_INFO("Host %s is already off", h->name().c_str());
 123   }
 124 }
 125
 126 sg_host_t SIMIX_host_self()
 127 {
 128   smx_actor_t process = SIMIX_process_self();
 129   return (process == nullptr) ? nullptr : process->host;
 130 }
 131
 132 /* needs to be public and without simcall for exceptions and logging events */
 133 const char* SIMIX_host_self_get_name()
 134 {
 135   sg_host_t host = SIMIX_host_self();
 136   if (host == nullptr || SIMIX_process_self() == simix_global->maestro_process)
 137     return "";
 138
 139   return sg_host_get_name(host);
 140 }
 141
 142 void _SIMIX_host_free_process_arg(void *data)
 143 {
 144   smx_process_arg_t arg = *(static_cast<smx_process_arg_t*>(data));
 145   delete arg;
 146 }
 147 /**
 148  * \brief Add a process to the list of the processes that the host will restart when it comes back
 149  * This function add a process to the list of the processes that will be restarted when the host comes
 150  * back. It is expected that this function is called when the host is down.
 151  * The processes will only be restarted once, meaning that you will have to register the process
 152  * again to restart the process again.
 153  */
 154 void SIMIX_host_add_auto_restart_process(
 155   sg_host_t host, const char *name, std::function<void()> code,
 156   void* data, const char *hostname, double kill_time,
 157   xbt_dict_t properties, int auto_restart)
 158 {
 159   if (!sg_host_simix(host)->auto_restart_processes) {
 160     sg_host_simix(host)->auto_restart_processes = xbt_dynar_new(sizeof(smx_process_arg_t),_SIMIX_host_free_process_arg);
 161   }
 162   smx_process_arg_t arg = new simgrid::simix::ProcessArg();
 163   arg->name = name;
 164   arg->code = std::move(code);
 165   arg->data = data;
 166   arg->hostname = hostname;
 167   arg->kill_time = kill_time;
 168   arg->properties = properties;
 169   arg->auto_restart = auto_restart;
 170
 171   if( host->isOff() && !xbt_dict_get_or_null(watched_hosts_lib,sg_host_get_name(host))){
 172     xbt_dict_set(watched_hosts_lib,sg_host_get_name(host),host,nullptr);
 173     XBT_DEBUG("Push host %s to watched_hosts_lib because state == SURF_RESOURCE_OFF",sg_host_get_name(host));
 174   }
 175   xbt_dynar_push_as(sg_host_simix(host)->auto_restart_processes,smx_process_arg_t,arg);
 176 }
 177 /** @brief Restart the list of processes that have been registered to the host */
 178 void SIMIX_host_autorestart(sg_host_t host)
 179 {
 180   unsigned int cpt;
 181   smx_process_arg_t arg;
 182   xbt_dynar_t process_list = sg_host_simix(host)->auto_restart_processes;
 183   if (!process_list)
 184     return;
 185
 186   xbt_dynar_foreach (process_list, cpt, arg) {
 187
 188     XBT_DEBUG("Restarting Process %s(%s) right now", arg->name.c_str(), arg->hostname);
 189     if (simix_global->create_process_function) {
 190       simix_global->create_process_function(arg->name.c_str(), arg->code, nullptr, arg->hostname, arg->kill_time,
 191                                             arg->properties, arg->auto_restart, nullptr);
 192     } else {
 193       simcall_process_create(arg->name.c_str(), arg->code, nullptr, arg->hostname, arg->kill_time, arg->properties,
 194                              arg->auto_restart);
 195     }
 196   }
 197   xbt_dynar_reset(process_list);
 198 }
 199
 200 smx_activity_t simcall_HANDLER_execution_start(smx_simcall_t simcall, const char* name, double flops_amount,
 201                                               double priority, double bound) {
 202   return SIMIX_execution_start(simcall->issuer, name,flops_amount,priority,bound);
 203 }
 204
 205 smx_activity_t SIMIX_execution_start(smx_actor_t issuer, const char *name, double flops_amount, double priority,
 206                                     double bound){
 207
 208   /* alloc structures and initialize */
 209   simgrid::kernel::activity::Exec *exec = new simgrid::kernel::activity::Exec(name, issuer->host);
 210
 211   /* set surf's action */
 212   if (!MC_is_active() && !MC_record_replay_is_active()) {
 213
 214     exec->surf_exec = issuer->host->pimpl_cpu->execution_start(flops_amount);
 215     exec->surf_exec->setData(exec);
 216     exec->surf_exec->setPriority(priority);
 217
 218     if (bound > 0)
 219       static_cast<simgrid::surf::CpuAction*>(exec->surf_exec)->setBound(bound);
 220   }
 221
 222   XBT_DEBUG("Create execute synchro %p: %s", exec, exec->name.c_str());
 223
 224   return exec;
 225 }
 226
 227 smx_activity_t SIMIX_execution_parallel_start(const char *name, int host_nb, sg_host_t *host_list, double *flops_amount,
 228                                              double *bytes_amount, double amount, double rate){
 229
 230   /* alloc structures and initialize */
 231   simgrid::kernel::activity::Exec *exec = new simgrid::kernel::activity::Exec(name, nullptr);
 232
 233   /* set surf's synchro */
 234   sg_host_t *host_list_cpy = xbt_new0(sg_host_t, host_nb);
 235   for (int i = 0; i < host_nb; i++)
 236     host_list_cpy[i] = host_list[i];
 237
 238   /* Check that we are not mixing VMs and PMs in the parallel task */
 239   simgrid::surf::HostImpl *host = host_list[0]->extension<simgrid::surf::HostImpl>();
 240   bool is_a_vm = (nullptr != dynamic_cast<simgrid::surf::VirtualMachine*>(host));
 241   for (int i = 1; i < host_nb; i++) {
 242     bool tmp_is_a_vm = (nullptr != dynamic_cast<simgrid::surf::VirtualMachine*>(host_list[i]->extension<simgrid::surf::HostImpl>()));
 243     xbt_assert(is_a_vm == tmp_is_a_vm, "parallel_execute: mixing VMs and PMs is not supported (yet).");
 244   }
 245
 246   /* set surf's synchro */
 247   if (!MC_is_active() && !MC_record_replay_is_active()) {
 248     exec->surf_exec = surf_host_model->executeParallelTask(host_nb, host_list_cpy, flops_amount, bytes_amount, rate);
 249     exec->surf_exec->setData(exec);
 250   }
 251   XBT_DEBUG("Create parallel execute synchro %p", exec);
 252
 253   return exec;
 254 }
 255
 256 void SIMIX_execution_cancel(smx_activity_t synchro)
 257 {
 258   XBT_DEBUG("Cancel synchro %p", synchro);
 259   simgrid::kernel::activity::Exec *exec = static_cast<simgrid::kernel::activity::Exec *>(synchro);
 260
 261   if (exec->surf_exec)
 262     exec->surf_exec->cancel();
 263 }
 264
 265 void SIMIX_execution_set_priority(smx_activity_t synchro, double priority)
 266 {
 267   simgrid::kernel::activity::Exec *exec = static_cast<simgrid::kernel::activity::Exec *>(synchro);
 268   if(exec->surf_exec)
 269     exec->surf_exec->setPriority(priority);
 270 }
 271
 272 void SIMIX_execution_set_bound(smx_activity_t synchro, double bound)
 273 {
 274   simgrid::kernel::activity::Exec *exec = static_cast<simgrid::kernel::activity::Exec *>(synchro);
 275   if(exec->surf_exec)
 276     static_cast<simgrid::surf::CpuAction*>(exec->surf_exec)->setBound(bound);
 277 }
 278
 279 void simcall_HANDLER_execution_wait(smx_simcall_t simcall, smx_activity_t synchro)
 280 {
 281   simgrid::kernel::activity::Exec *exec = static_cast<simgrid::kernel::activity::Exec *>(synchro);
 282   XBT_DEBUG("Wait for execution of synchro %p, state %d", synchro, (int)synchro->state);
 283
 284   /* Associate this simcall to the synchro */
 285   synchro->simcalls.push_back(simcall);
 286   simcall->issuer->waiting_synchro = synchro;
 287
 288   /* set surf's synchro */
 289   if (MC_is_active() || MC_record_replay_is_active()) {
 290     synchro->state = SIMIX_DONE;
 291     SIMIX_execution_finish(exec);
 292     return;
 293   }
 294
 295   /* If the synchro is already finished then perform the error handling */
 296   if (synchro->state != SIMIX_RUNNING)
 297     SIMIX_execution_finish(exec);
 298 }
 299
 300 void SIMIX_execution_finish(simgrid::kernel::activity::Exec *exec)
 301 {
 302   for (smx_simcall_t simcall : exec->simcalls) {
 303     switch (exec->state) {
 304
 305       case SIMIX_DONE:
 306         /* do nothing, synchro done */
 307         XBT_DEBUG("SIMIX_execution_finished: execution successful");
 308         break;
 309
 310       case SIMIX_FAILED:
 311         XBT_DEBUG("SIMIX_execution_finished: host '%s' failed", sg_host_get_name(simcall->issuer->host));
 312         simcall->issuer->context->iwannadie = 1;
 313         SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
 314         break;
 315
 316       case SIMIX_CANCELED:
 317         XBT_DEBUG("SIMIX_execution_finished: execution canceled");
 318         SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Canceled");
 319         break;
 320
 321       default:
 322         xbt_die("Internal error in SIMIX_execution_finish: unexpected synchro state %d",
 323             (int)exec->state);
 324     }
 325     /* Fail the process if the host is down */
 326     if (simcall->issuer->host->isOff())
 327       simcall->issuer->context->iwannadie = 1;
 328
 329     simcall->issuer->waiting_synchro = nullptr;
 330     simcall_execution_wait__set__result(simcall, exec->state);
 331     SIMIX_simcall_answer(simcall);
 332   }
 333
 334   /* We no longer need it */
 335   exec->unref();
 336 }
 337
 338 void SIMIX_set_category(smx_activity_t synchro, const char *category)
 339 {
 340   if (synchro->state != SIMIX_RUNNING)
 341     return;
 342
 343   simgrid::kernel::activity::Exec *exec = dynamic_cast<simgrid::kernel::activity::Exec *>(synchro);
 344   if (exec != nullptr) {
 345     exec->surf_exec->setCategory(category);
 346     return;
 347   }
 348
 349   simgrid::kernel::activity::Comm *comm = dynamic_cast<simgrid::kernel::activity::Comm *>(synchro);
 350   if (comm != nullptr) {
 351     comm->surf_comm->setCategory(category);
 352     return;
 353   }
 354 }