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     /** Re-starts all the actors that are marked as restartable.
  55      *
  56      * Weird things will happen if you turn on an host that is already on. S4U is fool-proof, not this.
  57      */
  58     void Host::turnOn()
  59     {
  60       unsigned int cpt;
  61       smx_process_arg_t arg;
  62       xbt_dynar_foreach(boot_processes,cpt,arg) {
  63         XBT_DEBUG("Booting Process %s(%s) right now", arg->name.c_str(), arg->hostname);
  64         // FIXME: factorize this code by registering the simcall as default function
  65         if (simix_global->create_process_function) {
  66           simix_global->create_process_function(arg->name.c_str(),
  67                                                 arg->code,
  68                                                 nullptr,
  69                                                 arg->hostname,
  70                                                 arg->kill_time,
  71                                                 arg->properties,
  72                                                 arg->auto_restart,
  73                                                 nullptr);
  74         } else {
  75           simcall_process_create(arg->name.c_str(),
  76                                  arg->code,
  77                                  nullptr,
  78                                  arg->hostname,
  79                                  arg->kill_time,
  80                                  arg->properties,
  81                                  arg->auto_restart);
  82         }
  83       }
  84     }
  85
  86 }} // namespaces
  87
  88 /** @brief Stop the host if it is on */
  89 void SIMIX_host_off(sg_host_t h, smx_actor_t issuer)
  90 {
  91   smx_host_priv_t host = sg_host_simix(h);
  92
  93   xbt_assert((host != nullptr), "Invalid parameters");
  94
  95   if (h->isOn()) {
  96     simgrid::surf::HostImpl* surf_host = h->extension<simgrid::surf::HostImpl>();
  97     surf_host->turnOff();
  98
  99     /* Clean Simulator data */
 100     if (xbt_swag_size(host->process_list) != 0) {
 101       smx_actor_t process = nullptr;
 102       xbt_swag_foreach(process, host->process_list) {
 103         SIMIX_process_kill(process, issuer);
 104         XBT_DEBUG("Killing %s on %s by %s",
 105           process->name.c_str(),  sg_host_get_name(process->host),
 106           issuer->name.c_str());
 107       }
 108     }
 109   } else {
 110     XBT_INFO("Host %s is already off", h->name().c_str());
 111   }
 112 }
 113
 114 sg_host_t SIMIX_host_self()
 115 {
 116   smx_actor_t process = SIMIX_process_self();
 117   return (process == nullptr) ? nullptr : process->host;
 118 }
 119
 120 /* needs to be public and without simcall for exceptions and logging events */
 121 const char* SIMIX_host_self_get_name()
 122 {
 123   sg_host_t host = SIMIX_host_self();
 124   if (host == nullptr || SIMIX_process_self() == simix_global->maestro_process)
 125     return "";
 126
 127   return sg_host_get_name(host);
 128 }
 129
 130 void _SIMIX_host_free_process_arg(void *data)
 131 {
 132   smx_process_arg_t arg = *(static_cast<smx_process_arg_t*>(data));
 133   delete arg;
 134 }
 135 /**
 136  * \brief Add a process to the list of the processes that the host will restart when it comes back
 137  * This function add a process to the list of the processes that will be restarted when the host comes
 138  * back. It is expected that this function is called when the host is down.
 139  * The processes will only be restarted once, meaning that you will have to register the process
 140  * again to restart the process again.
 141  */
 142 void SIMIX_host_add_auto_restart_process(
 143   sg_host_t host, const char *name, std::function<void()> code,
 144   void* data, const char *hostname, double kill_time,
 145   xbt_dict_t properties, int auto_restart)
 146 {
 147   if (!sg_host_simix(host)->auto_restart_processes) {
 148     sg_host_simix(host)->auto_restart_processes = xbt_dynar_new(sizeof(smx_process_arg_t),_SIMIX_host_free_process_arg);
 149   }
 150   smx_process_arg_t arg = new simgrid::simix::ProcessArg();
 151   arg->name = name;
 152   arg->code = std::move(code);
 153   arg->data = data;
 154   arg->hostname = hostname;
 155   arg->kill_time = kill_time;
 156   arg->properties = properties;
 157   arg->auto_restart = auto_restart;
 158
 159   if( host->isOff() && !xbt_dict_get_or_null(watched_hosts_lib,sg_host_get_name(host))){
 160     xbt_dict_set(watched_hosts_lib,sg_host_get_name(host),host,nullptr);
 161     XBT_DEBUG("Push host %s to watched_hosts_lib because state == SURF_RESOURCE_OFF",sg_host_get_name(host));
 162   }
 163   xbt_dynar_push_as(sg_host_simix(host)->auto_restart_processes,smx_process_arg_t,arg);
 164 }
 165 /** @brief Restart the list of processes that have been registered to the host */
 166 void SIMIX_host_autorestart(sg_host_t host)
 167 {
 168   unsigned int cpt;
 169   smx_process_arg_t arg;
 170   xbt_dynar_t process_list = sg_host_simix(host)->auto_restart_processes;
 171   if (!process_list)
 172     return;
 173
 174   xbt_dynar_foreach (process_list, cpt, arg) {
 175
 176     XBT_DEBUG("Restarting Process %s(%s) right now", arg->name.c_str(), arg->hostname);
 177     if (simix_global->create_process_function) {
 178       simix_global->create_process_function(arg->name.c_str(), arg->code, nullptr, arg->hostname, arg->kill_time,
 179                                             arg->properties, arg->auto_restart, nullptr);
 180     } else {
 181       simcall_process_create(arg->name.c_str(), arg->code, nullptr, arg->hostname, arg->kill_time, arg->properties,
 182                              arg->auto_restart);
 183     }
 184   }
 185   xbt_dynar_reset(process_list);
 186 }
 187
 188 smx_activity_t simcall_HANDLER_execution_start(smx_simcall_t simcall, const char* name, double flops_amount,
 189                                               double priority, double bound) {
 190   return SIMIX_execution_start(simcall->issuer, name,flops_amount,priority,bound);
 191 }
 192
 193 smx_activity_t SIMIX_execution_start(smx_actor_t issuer, const char *name, double flops_amount, double priority,
 194                                     double bound){
 195
 196   /* alloc structures and initialize */
 197   simgrid::kernel::activity::Exec *exec = new simgrid::kernel::activity::Exec(name, issuer->host);
 198
 199   /* set surf's action */
 200   if (!MC_is_active() && !MC_record_replay_is_active()) {
 201
 202     exec->surf_exec = issuer->host->pimpl_cpu->execution_start(flops_amount);
 203     exec->surf_exec->setData(exec);
 204     exec->surf_exec->setPriority(priority);
 205
 206     if (bound > 0)
 207       static_cast<simgrid::surf::CpuAction*>(exec->surf_exec)->setBound(bound);
 208   }
 209
 210   XBT_DEBUG("Create execute synchro %p: %s", exec, exec->name.c_str());
 211
 212   return exec;
 213 }
 214
 215 smx_activity_t SIMIX_execution_parallel_start(const char *name, int host_nb, sg_host_t *host_list, double *flops_amount,
 216                                              double *bytes_amount, double amount, double rate){
 217
 218   /* alloc structures and initialize */
 219   simgrid::kernel::activity::Exec *exec = new simgrid::kernel::activity::Exec(name, nullptr);
 220
 221   /* set surf's synchro */
 222   sg_host_t *host_list_cpy = xbt_new0(sg_host_t, host_nb);
 223   for (int i = 0; i < host_nb; i++)
 224     host_list_cpy[i] = host_list[i];
 225
 226   /* Check that we are not mixing VMs and PMs in the parallel task */
 227   simgrid::surf::HostImpl *host = host_list[0]->extension<simgrid::surf::HostImpl>();
 228   bool is_a_vm = (nullptr != dynamic_cast<simgrid::surf::VirtualMachine*>(host));
 229   for (int i = 1; i < host_nb; i++) {
 230     bool tmp_is_a_vm = (nullptr != dynamic_cast<simgrid::surf::VirtualMachine*>(host_list[i]->extension<simgrid::surf::HostImpl>()));
 231     xbt_assert(is_a_vm == tmp_is_a_vm, "parallel_execute: mixing VMs and PMs is not supported (yet).");
 232   }
 233
 234   /* set surf's synchro */
 235   if (!MC_is_active() && !MC_record_replay_is_active()) {
 236     exec->surf_exec = surf_host_model->executeParallelTask(host_nb, host_list_cpy, flops_amount, bytes_amount, rate);
 237     exec->surf_exec->setData(exec);
 238   }
 239   XBT_DEBUG("Create parallel execute synchro %p", exec);
 240
 241   return exec;
 242 }
 243
 244 void SIMIX_execution_cancel(smx_activity_t synchro)
 245 {
 246   XBT_DEBUG("Cancel synchro %p", synchro);
 247   simgrid::kernel::activity::Exec *exec = static_cast<simgrid::kernel::activity::Exec *>(synchro);
 248
 249   if (exec->surf_exec)
 250     exec->surf_exec->cancel();
 251 }
 252
 253 void SIMIX_execution_set_priority(smx_activity_t synchro, double priority)
 254 {
 255   simgrid::kernel::activity::Exec *exec = static_cast<simgrid::kernel::activity::Exec *>(synchro);
 256   if(exec->surf_exec)
 257     exec->surf_exec->setPriority(priority);
 258 }
 259
 260 void SIMIX_execution_set_bound(smx_activity_t synchro, double bound)
 261 {
 262   simgrid::kernel::activity::Exec *exec = static_cast<simgrid::kernel::activity::Exec *>(synchro);
 263   if(exec->surf_exec)
 264     static_cast<simgrid::surf::CpuAction*>(exec->surf_exec)->setBound(bound);
 265 }
 266
 267 void simcall_HANDLER_execution_wait(smx_simcall_t simcall, smx_activity_t synchro)
 268 {
 269   simgrid::kernel::activity::Exec *exec = static_cast<simgrid::kernel::activity::Exec *>(synchro);
 270   XBT_DEBUG("Wait for execution of synchro %p, state %d", synchro, (int)synchro->state);
 271
 272   /* Associate this simcall to the synchro */
 273   synchro->simcalls.push_back(simcall);
 274   simcall->issuer->waiting_synchro = synchro;
 275
 276   /* set surf's synchro */
 277   if (MC_is_active() || MC_record_replay_is_active()) {
 278     synchro->state = SIMIX_DONE;
 279     SIMIX_execution_finish(exec);
 280     return;
 281   }
 282
 283   /* If the synchro is already finished then perform the error handling */
 284   if (synchro->state != SIMIX_RUNNING)
 285     SIMIX_execution_finish(exec);
 286 }
 287
 288 void SIMIX_execution_finish(simgrid::kernel::activity::Exec *exec)
 289 {
 290   for (smx_simcall_t simcall : exec->simcalls) {
 291     switch (exec->state) {
 292
 293       case SIMIX_DONE:
 294         /* do nothing, synchro done */
 295         XBT_DEBUG("SIMIX_execution_finished: execution successful");
 296         break;
 297
 298       case SIMIX_FAILED:
 299         XBT_DEBUG("SIMIX_execution_finished: host '%s' failed", sg_host_get_name(simcall->issuer->host));
 300         simcall->issuer->context->iwannadie = 1;
 301         SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
 302         break;
 303
 304       case SIMIX_CANCELED:
 305         XBT_DEBUG("SIMIX_execution_finished: execution canceled");
 306         SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Canceled");
 307         break;
 308
 309       default:
 310         xbt_die("Internal error in SIMIX_execution_finish: unexpected synchro state %d",
 311             (int)exec->state);
 312     }
 313     /* Fail the process if the host is down */
 314     if (simcall->issuer->host->isOff())
 315       simcall->issuer->context->iwannadie = 1;
 316
 317     simcall->issuer->waiting_synchro = nullptr;
 318     simcall_execution_wait__set__result(simcall, exec->state);
 319     SIMIX_simcall_answer(simcall);
 320   }
 321
 322   /* We no longer need it */
 323   exec->unref();
 324 }
 325
 326 void SIMIX_set_category(smx_activity_t synchro, const char *category)
 327 {
 328   if (synchro->state != SIMIX_RUNNING)
 329     return;
 330
 331   simgrid::kernel::activity::Exec *exec = dynamic_cast<simgrid::kernel::activity::Exec *>(synchro);
 332   if (exec != nullptr) {
 333     exec->surf_exec->setCategory(category);
 334     return;
 335   }
 336
 337   simgrid::kernel::activity::Comm *comm = dynamic_cast<simgrid::kernel::activity::Comm *>(synchro);
 338   if (comm != nullptr) {
 339     comm->surf_comm->setCategory(category);
 340     return;
 341   }
 342 }