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