1 /* Copyright (c) 2007-2015. The SimGrid Team.
2 * All rights reserved. */
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. */
7 #include "smx_private.h"
8 #include "xbt/sysdep.h"
10 #include "src/mc/mc_replay.h"
11 #include "src/surf/virtual_machine.hpp"
12 #include "src/surf/HostImpl.hpp"
14 #include "src/simix/SynchroExec.hpp"
15 #include "src/simix/SynchroComm.hpp"
17 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_host, simix, "SIMIX hosts");
19 static void SIMIX_execution_finish(smx_synchro_t synchro);
22 * \brief Internal function to create a SIMIX host.
23 * \param name name of the host to create
25 void SIMIX_host_create(sg_host_t host) // FIXME: braindead prototype. Take sg_host as parameter
27 smx_host_priv_t smx_host = xbt_new0(s_smx_host_priv_t, 1);
31 smx_host->process_list = xbt_swag_new(xbt_swag_offset(proc, host_proc_hookup));
33 /* Update global variables */
34 sg_host_simix_set(host, smx_host);
38 * \brief Start the host if it is off
41 void SIMIX_host_on(sg_host_t h)
43 smx_host_priv_t host = sg_host_simix(h);
45 xbt_assert((host != NULL), "Invalid parameters");
48 simgrid::surf::HostImpl* surf_host = h->extension<simgrid::surf::HostImpl>();
52 smx_process_arg_t arg;
53 xbt_dynar_foreach(host->boot_processes,cpt,arg) {
55 char** argv = xbt_new(char*, arg->argc);
56 for (int i=0; i<arg->argc; i++)
57 argv[i] = xbt_strdup(arg->argv[i]);
59 XBT_DEBUG("Booting Process %s(%s) right now", arg->argv[0], arg->hostname);
60 if (simix_global->create_process_function) {
61 simix_global->create_process_function(argv[0],
72 simcall_process_create(arg->argv[0],
87 * \brief Stop the host if it is on
90 void SIMIX_host_off(sg_host_t h, smx_process_t issuer)
92 smx_host_priv_t host = sg_host_simix(h);
94 xbt_assert((host != NULL), "Invalid parameters");
97 simgrid::surf::HostImpl* surf_host = h->extension<simgrid::surf::HostImpl>();
100 /* Clean Simulator data */
101 if (xbt_swag_size(host->process_list) != 0) {
102 smx_process_t process = NULL;
103 xbt_swag_foreach(process, host->process_list) {
104 SIMIX_process_kill(process, issuer);
105 XBT_DEBUG("Killing %s on %s by %s", process->name, sg_host_get_name(process->host), issuer->name);
109 XBT_INFO("Host %s is already off",h->name().c_str());
114 * \brief Internal function to destroy a SIMIX host.
116 * \param h the host to destroy (a sg_host_t)
118 void SIMIX_host_destroy(void *h)
120 smx_host_priv_t host = (smx_host_priv_t) h;
122 xbt_assert((host != NULL), "Invalid parameters");
124 /* Clean Simulator data */
125 if (xbt_swag_size(host->process_list) != 0) {
126 char *msg = xbt_strdup("Shutting down host, but it's not empty:");
128 smx_process_t process = NULL;
130 xbt_swag_foreach(process, host->process_list) {
131 tmp = bprintf("%s\n\t%s", msg, process->name);
135 SIMIX_display_process_status();
136 THROWF(arg_error, 0, "%s", msg);
138 xbt_dynar_free(&host->auto_restart_processes);
139 xbt_dynar_free(&host->boot_processes);
140 xbt_swag_free(host->process_list);
142 /* Clean host structure */
147 sg_host_t SIMIX_host_self(void)
149 smx_process_t process = SIMIX_process_self();
150 return (process == NULL) ? NULL : SIMIX_process_get_host(process);
153 /* needs to be public and without simcall because it is called
154 by exceptions and logging events */
155 const char* SIMIX_host_self_get_name(void)
157 sg_host_t host = SIMIX_host_self();
158 if (host == NULL || SIMIX_process_self() == simix_global->maestro_process)
161 return sg_host_get_name(host);
164 void _SIMIX_host_free_process_arg(void *data)
166 smx_process_arg_t arg = *(smx_process_arg_t*)data;
167 for (int i = 0; i < arg->argc; i++)
168 xbt_free(arg->argv[i]);
174 * \brief Add a process to the list of the processes that the host will restart when it comes back
175 * This function add a process to the list of the processes that will be restarted when the host comes
176 * back. It is expected that this function is called when the host is down.
177 * The processes will only be restarted once, meaning that you will have to register the process
178 * again to restart the process again.
180 void SIMIX_host_add_auto_restart_process(sg_host_t host,
182 xbt_main_func_t code,
184 const char *hostname,
186 int argc, char **argv,
187 xbt_dict_t properties,
190 if (!sg_host_simix(host)->auto_restart_processes) {
191 sg_host_simix(host)->auto_restart_processes = xbt_dynar_new(sizeof(smx_process_arg_t),_SIMIX_host_free_process_arg);
193 smx_process_arg_t arg = xbt_new(s_smx_process_arg_t,1);
194 arg->name = xbt_strdup(name);
197 arg->hostname = hostname;
198 arg->kill_time = kill_time;
201 arg->argv = xbt_new(char*,argc + 1);
203 for (int i = 0; i < argc; i++)
204 arg->argv[i] = xbt_strdup(argv[i]);
205 arg->argv[argc] = NULL;
207 arg->properties = properties;
208 arg->auto_restart = auto_restart;
210 if( host->isOff() && !xbt_dict_get_or_null(watched_hosts_lib,sg_host_get_name(host))){
211 xbt_dict_set(watched_hosts_lib,sg_host_get_name(host),host,NULL);
212 XBT_DEBUG("Push host %s to watched_hosts_lib because state == SURF_RESOURCE_OFF",sg_host_get_name(host));
214 xbt_dynar_push_as(sg_host_simix(host)->auto_restart_processes,smx_process_arg_t,arg);
217 * \brief Restart the list of processes that have been registered to the host
219 void SIMIX_host_autorestart(sg_host_t host)
222 smx_process_arg_t arg;
223 xbt_dynar_t process_list = sg_host_simix(host)->auto_restart_processes;
227 xbt_dynar_foreach (process_list, cpt, arg) {
229 XBT_DEBUG("Restarting Process %s(%s) right now", arg->argv[0], arg->hostname);
230 if (simix_global->create_process_function) {
231 simix_global->create_process_function(arg->argv[0],
242 simcall_process_create(arg->argv[0],
243 (xbt_main_func_t) arg->code,
253 /* arg->argv is used by the process created above. Hide it to
254 * _SIMIX_host_free_process_arg() which is called by xbt_dynar_reset()
259 xbt_dynar_reset(process_list);
262 smx_synchro_t simcall_HANDLER_execution_start(smx_simcall_t simcall,
263 const char* name, double flops_amount, double priority, double bound, unsigned long affinity_mask) {
264 return SIMIX_execution_start(simcall->issuer, name,flops_amount,priority,bound,affinity_mask);
266 smx_synchro_t SIMIX_execution_start(smx_process_t issuer, const char *name,
267 double flops_amount, double priority, double bound, unsigned long affinity_mask){
269 /* alloc structures and initialize */
270 simgrid::simix::Exec *exec = new simgrid::simix::Exec();
271 exec->name = xbt_strdup(name);
272 exec->state = SIMIX_RUNNING;
273 exec->host = issuer->host;
275 /* set surf's action */
276 if (!MC_is_active() && !MC_record_replay_is_active()) {
278 exec->surf_exec = issuer->host->pimpl_cpu->execution_start(flops_amount);
279 exec->surf_exec->setData(exec);
280 exec->surf_exec->setPriority(priority);
283 static_cast<simgrid::surf::CpuAction*>(exec->surf_exec)->setBound(bound);
285 if (affinity_mask != 0) {
286 /* just a double check to confirm that this host is the host where this task is running. */
287 xbt_assert(exec->host == issuer->host);
288 static_cast<simgrid::surf::CpuAction*>(exec->surf_exec)
289 ->setAffinity(issuer->host->pimpl_cpu, affinity_mask);
293 XBT_DEBUG("Create execute synchro %p: %s", exec, exec->name);
298 smx_synchro_t SIMIX_execution_parallel_start(const char *name,
299 int host_nb, sg_host_t *host_list,
300 double *flops_amount, double *bytes_amount,
301 double amount, double rate){
303 sg_host_t *host_list_cpy = NULL;
306 /* alloc structures and initialize */
307 simgrid::simix::Exec *exec = new simgrid::simix::Exec();
308 exec->name = xbt_strdup(name);
309 exec->state = SIMIX_RUNNING;
310 exec->host = nullptr; /* FIXME: do we need the list of hosts? */
312 /* set surf's synchro */
313 host_list_cpy = xbt_new0(sg_host_t, host_nb);
314 for (i = 0; i < host_nb; i++)
315 host_list_cpy[i] = host_list[i];
317 /* Check that we are not mixing VMs and PMs in the parallel task */
318 simgrid::surf::HostImpl *host = host_list[0]->extension<simgrid::surf::HostImpl>();
319 bool is_a_vm = (nullptr != dynamic_cast<simgrid::surf::VirtualMachine*>(host));
320 for (i = 1; i < host_nb; i++) {
321 bool tmp_is_a_vm = (nullptr != dynamic_cast<simgrid::surf::VirtualMachine*>(host_list[i]->extension<simgrid::surf::HostImpl>()));
322 xbt_assert(is_a_vm == tmp_is_a_vm, "parallel_execute: mixing VMs and PMs is not supported (yet).");
325 /* set surf's synchro */
326 if (!MC_is_active() && !MC_record_replay_is_active()) {
327 exec->surf_exec = surf_host_model->executeParallelTask(host_nb, host_list_cpy, flops_amount, bytes_amount, rate);
328 exec->surf_exec->setData(exec);
330 XBT_DEBUG("Create parallel execute synchro %p", exec);
335 void SIMIX_execution_destroy(smx_synchro_t synchro)
337 XBT_DEBUG("Destroy synchro %p", synchro);
338 simgrid::simix::Exec *exec = static_cast<simgrid::simix::Exec *>(synchro);
340 if (exec->surf_exec) {
341 exec->surf_exec->unref();
342 exec->surf_exec = NULL;
347 void SIMIX_execution_cancel(smx_synchro_t synchro)
349 XBT_DEBUG("Cancel synchro %p", synchro);
350 simgrid::simix::Exec *exec = static_cast<simgrid::simix::Exec *>(synchro);
353 exec->surf_exec->cancel();
356 void SIMIX_execution_set_priority(smx_synchro_t synchro, double priority)
358 simgrid::simix::Exec *exec = static_cast<simgrid::simix::Exec *>(synchro);
360 exec->surf_exec->setPriority(priority);
363 void SIMIX_execution_set_bound(smx_synchro_t synchro, double bound)
365 simgrid::simix::Exec *exec = static_cast<simgrid::simix::Exec *>(synchro);
367 static_cast<simgrid::surf::CpuAction*>(exec->surf_exec)->setBound(bound);
370 void SIMIX_execution_set_affinity(smx_synchro_t synchro, sg_host_t host, unsigned long mask)
372 simgrid::simix::Exec *exec = static_cast<simgrid::simix::Exec *>(synchro);
373 if(exec->surf_exec) {
374 /* just a double check to confirm that this host is the host where this task is running. */
375 xbt_assert(exec->host == host);
376 static_cast<simgrid::surf::CpuAction*>(exec->surf_exec)->setAffinity(host->pimpl_cpu, mask);
380 void simcall_HANDLER_execution_wait(smx_simcall_t simcall, smx_synchro_t synchro)
383 XBT_DEBUG("Wait for execution of synchro %p, state %d", synchro, (int)synchro->state);
385 /* Associate this simcall to the synchro */
386 xbt_fifo_push(synchro->simcalls, simcall);
387 simcall->issuer->waiting_synchro = synchro;
389 /* set surf's synchro */
390 if (MC_is_active() || MC_record_replay_is_active()) {
391 synchro->state = SIMIX_DONE;
392 SIMIX_execution_finish(synchro);
396 /* If the synchro is already finished then perform the error handling */
397 if (synchro->state != SIMIX_RUNNING)
398 SIMIX_execution_finish(synchro);
401 void SIMIX_execution_suspend(smx_synchro_t synchro)
403 synchro->suspend(); // FIXME: USELESS
406 void SIMIX_execution_resume(smx_synchro_t synchro)
408 synchro->resume(); // FIXME: USELESS
411 void SIMIX_execution_finish(smx_synchro_t synchro)
413 xbt_fifo_item_t item;
414 smx_simcall_t simcall;
416 xbt_fifo_foreach(synchro->simcalls, item, simcall, smx_simcall_t) {
418 switch (synchro->state) {
421 /* do nothing, synchro done */
422 XBT_DEBUG("SIMIX_execution_finished: execution successful");
426 XBT_DEBUG("SIMIX_execution_finished: host '%s' failed", sg_host_get_name(simcall->issuer->host));
427 simcall->issuer->context->iwannadie = 1;
428 SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
432 XBT_DEBUG("SIMIX_execution_finished: execution canceled");
433 SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Canceled");
437 xbt_die("Internal error in SIMIX_execution_finish: unexpected synchro state %d",
438 (int)synchro->state);
440 /* check if the host is down */
441 if (simcall->issuer->host->isOff()) {
442 simcall->issuer->context->iwannadie = 1;
445 simcall->issuer->waiting_synchro = NULL;
446 simcall_execution_wait__set__result(simcall, synchro->state);
447 SIMIX_simcall_answer(simcall);
450 /* We no longer need it */
451 SIMIX_execution_destroy(synchro);
455 void SIMIX_post_host_execute(smx_synchro_t synchro)
457 simgrid::simix::Exec *exec = dynamic_cast<simgrid::simix::Exec *>(synchro);
459 if (exec != nullptr && exec->host && /* FIMXE: handle resource failure for parallel tasks too */
460 exec->host->isOff()) {
461 /* If the host running the synchro failed, notice it. This way, the asking
462 * process can be killed if it runs on that host itself */
463 synchro->state = SIMIX_FAILED;
464 } else if (exec->surf_exec->getState() == simgrid::surf::Action::State::failed) {
465 /* If the host running the synchro didn't fail, then the synchro was canceled */
466 synchro->state = SIMIX_CANCELED;
468 synchro->state = SIMIX_DONE;
471 if (exec != nullptr && exec->surf_exec) {
472 exec->surf_exec->unref();
473 exec->surf_exec = NULL;
476 /* If there are simcalls associated with the synchro, then answer them */
477 if (xbt_fifo_size(synchro->simcalls)) {
478 SIMIX_execution_finish(synchro);
483 void SIMIX_set_category(smx_synchro_t synchro, const char *category)
485 if (synchro->state != SIMIX_RUNNING) return;
487 simgrid::simix::Exec *exec = dynamic_cast<simgrid::simix::Exec *>(synchro);
488 if (exec != nullptr) {
489 exec->surf_exec->setCategory(category);
493 simgrid::simix::Comm *comm = dynamic_cast<simgrid::simix::Comm *>(synchro);
494 if (comm != nullptr) {
495 comm->surf_comm->setCategory(category);