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"
12 #include "src/mc/mc_replay.h"
13 #include "src/surf/host_interface.hpp"
15 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_host, simix,
18 static void SIMIX_execution_finish(smx_synchro_t synchro);
21 * \brief Internal function to create a SIMIX host.
22 * \param name name of the host to create
24 void SIMIX_host_create(sg_host_t host) // FIXME: braindead prototype. Take sg_host as parameter
26 smx_host_priv_t smx_host = xbt_new0(s_smx_host_priv_t, 1);
30 smx_host->process_list =
31 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::Host* surf_host = h->extension<simgrid::surf::Host>();
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::Host* surf_host = h->extension<simgrid::surf::Host>();
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;
168 for (i = 0; i < arg->argc; i++)
169 xbt_free(arg->argv[i]);
175 * \brief Add a process to the list of the processes that the host will restart when it comes back
176 * This function add a process to the list of the processes that will be restarted when the host comes
177 * back. It is expected that this function is called when the host is down.
178 * The processes will only be restarted once, meaning that you will have to register the process
179 * again to restart the process again.
181 void SIMIX_host_add_auto_restart_process(sg_host_t host,
183 xbt_main_func_t code,
185 const char *hostname,
187 int argc, char **argv,
188 xbt_dict_t properties,
191 if (!sg_host_simix(host)->auto_restart_processes) {
192 sg_host_simix(host)->auto_restart_processes = xbt_dynar_new(sizeof(smx_process_arg_t),_SIMIX_host_free_process_arg);
194 smx_process_arg_t arg = xbt_new(s_smx_process_arg_t,1);
195 arg->name = xbt_strdup(name);
198 arg->hostname = hostname;
199 arg->kill_time = kill_time;
202 arg->argv = xbt_new(char*,argc + 1);
205 for (i = 0; i < argc; i++) {
206 arg->argv[i] = xbt_strdup(argv[i]);
208 arg->argv[argc] = NULL;
210 arg->properties = properties;
211 arg->auto_restart = auto_restart;
213 if( ! sg_host_is_on(host)
214 && !xbt_dict_get_or_null(watched_hosts_lib,sg_host_get_name(host))){
215 xbt_dict_set(watched_hosts_lib,sg_host_get_name(host),host,NULL);
216 XBT_DEBUG("Have pushed host %s to watched_hosts_lib because state == SURF_RESOURCE_OFF",sg_host_get_name(host));
218 xbt_dynar_push_as(sg_host_simix(host)->auto_restart_processes,smx_process_arg_t,arg);
221 * \brief Restart the list of processes that have been registered to the host
223 void SIMIX_host_autorestart(sg_host_t host)
226 smx_process_arg_t arg;
227 xbt_dynar_t process_list = sg_host_simix(host)->auto_restart_processes;
231 xbt_dynar_foreach (process_list, cpt, arg) {
233 XBT_DEBUG("Restarting Process %s(%s) right now", arg->argv[0], arg->hostname);
234 if (simix_global->create_process_function) {
235 simix_global->create_process_function(arg->argv[0],
246 simcall_process_create(arg->argv[0],
247 (xbt_main_func_t) arg->code,
257 /* arg->argv is used by the process created above. Hide it to
258 * _SIMIX_host_free_process_arg() which is called by xbt_dynar_reset()
263 xbt_dynar_reset(process_list);
266 smx_synchro_t simcall_HANDLER_execution_start(smx_simcall_t simcall,
267 const char* name, double flops_amount, double priority, double bound, unsigned long affinity_mask) {
268 return SIMIX_execution_start(simcall->issuer, name,flops_amount,priority,bound,affinity_mask);
270 smx_synchro_t SIMIX_execution_start(smx_process_t issuer, const char *name,
271 double flops_amount, double priority, double bound, unsigned long affinity_mask){
273 /* alloc structures and initialize */
274 smx_synchro_t synchro = (smx_synchro_t) xbt_mallocator_get(simix_global->synchro_mallocator);
275 synchro->type = SIMIX_SYNC_EXECUTE;
276 synchro->name = xbt_strdup(name);
277 synchro->state = SIMIX_RUNNING;
278 synchro->execution.host = issuer->host;
279 synchro->category = NULL;
281 /* set surf's action */
282 if (!MC_is_active() && !MC_record_replay_is_active()) {
284 synchro->execution.surf_exec = issuer->host->pimpl_cpu->execution_start(flops_amount);
285 synchro->execution.surf_exec->setData(synchro);
286 synchro->execution.surf_exec->setPriority(priority);
289 static_cast<simgrid::surf::CpuAction*>(synchro->execution.surf_exec)
292 if (affinity_mask != 0) {
293 /* just a double check to confirm that this host is the host where this task is running. */
294 xbt_assert(synchro->execution.host == issuer->host);
295 static_cast<simgrid::surf::CpuAction*>(synchro->execution.surf_exec)
296 ->setAffinity(issuer->host->pimpl_cpu, affinity_mask);
300 XBT_DEBUG("Create execute synchro %p: %s", synchro, synchro->name);
305 smx_synchro_t SIMIX_execution_parallel_start(const char *name,
306 int host_nb, sg_host_t *host_list,
307 double *flops_amount, double *bytes_amount,
308 double amount, double rate){
310 sg_host_t*host_list_cpy = NULL;
313 /* alloc structures and initialize */
314 smx_synchro_t synchro = (smx_synchro_t) xbt_mallocator_get(simix_global->synchro_mallocator);
315 synchro->type = SIMIX_SYNC_PARALLEL_EXECUTE;
316 synchro->name = xbt_strdup(name);
317 synchro->state = SIMIX_RUNNING;
318 synchro->execution.host = NULL; /* FIXME: do we need the list of hosts? */
319 synchro->category = NULL;
321 /* set surf's synchro */
322 host_list_cpy = xbt_new0(sg_host_t, host_nb);
323 for (i = 0; i < host_nb; i++)
324 host_list_cpy[i] = host_list[i];
327 /* FIXME: what happens if host_list contains VMs and PMs. If
328 * execute_parallel_task() does not change the state of the model, we can mix
330 surf_host_model_t ws_model =
331 host_list[0]->extension<simgrid::surf::Host>()->getModel();
332 for (i = 1; i < host_nb; i++) {
333 surf_host_model_t ws_model_tmp =
334 host_list[i]->extension<simgrid::surf::Host>()->getModel();
335 if (ws_model_tmp != ws_model) {
336 XBT_CRITICAL("mixing VMs and PMs is not supported");
341 /* set surf's synchro */
342 if (!MC_is_active() && !MC_record_replay_is_active()) {
343 synchro->execution.surf_exec =
344 surf_host_model->executeParallelTask(
345 host_nb, host_list_cpy, flops_amount, bytes_amount, rate);
347 synchro->execution.surf_exec->setData(synchro);
349 XBT_DEBUG("Create parallel execute synchro %p", synchro);
354 void SIMIX_execution_destroy(smx_synchro_t synchro)
356 XBT_DEBUG("Destroy synchro %p", synchro);
358 if (synchro->execution.surf_exec) {
359 synchro->execution.surf_exec->unref();
360 synchro->execution.surf_exec = NULL;
362 xbt_free(synchro->name);
363 xbt_mallocator_release(simix_global->synchro_mallocator, synchro);
366 void SIMIX_execution_cancel(smx_synchro_t synchro)
368 XBT_DEBUG("Cancel synchro %p", synchro);
370 if (synchro->execution.surf_exec)
371 synchro->execution.surf_exec->cancel();
374 double SIMIX_execution_get_remains(smx_synchro_t synchro)
378 if (synchro->state == SIMIX_RUNNING)
379 result = surf_action_get_remains(synchro->execution.surf_exec);
384 e_smx_state_t SIMIX_execution_get_state(smx_synchro_t synchro)
386 return synchro->state;
389 void SIMIX_execution_set_priority(smx_synchro_t synchro, double priority)
391 if(synchro->execution.surf_exec)
392 synchro->execution.surf_exec->setPriority(priority);
395 void SIMIX_execution_set_bound(smx_synchro_t synchro, double bound)
397 if(synchro->execution.surf_exec)
398 static_cast<simgrid::surf::CpuAction*>(synchro->execution.surf_exec)->setBound(bound);
401 void SIMIX_execution_set_affinity(smx_synchro_t synchro, sg_host_t host, unsigned long mask)
403 xbt_assert(synchro->type == SIMIX_SYNC_EXECUTE);
405 if (synchro->execution.surf_exec) {
406 /* just a double check to confirm that this host is the host where this task is running. */
407 xbt_assert(synchro->execution.host == host);
408 static_cast<simgrid::surf::CpuAction*>(synchro->execution.surf_exec)
409 ->setAffinity(host->pimpl_cpu, mask);
413 void simcall_HANDLER_execution_wait(smx_simcall_t simcall, smx_synchro_t synchro)
416 XBT_DEBUG("Wait for execution of synchro %p, state %d", synchro, (int)synchro->state);
418 /* Associate this simcall to the synchro */
419 xbt_fifo_push(synchro->simcalls, simcall);
420 simcall->issuer->waiting_synchro = synchro;
422 /* set surf's synchro */
423 if (MC_is_active() || MC_record_replay_is_active()) {
424 synchro->state = SIMIX_DONE;
425 SIMIX_execution_finish(synchro);
429 /* If the synchro is already finished then perform the error handling */
430 if (synchro->state != SIMIX_RUNNING)
431 SIMIX_execution_finish(synchro);
434 void SIMIX_execution_suspend(smx_synchro_t synchro)
436 if(synchro->execution.surf_exec)
437 synchro->execution.surf_exec->suspend();
440 void SIMIX_execution_resume(smx_synchro_t synchro)
442 if(synchro->execution.surf_exec)
443 synchro->execution.surf_exec->resume();
446 void SIMIX_execution_finish(smx_synchro_t synchro)
448 xbt_fifo_item_t item;
449 smx_simcall_t simcall;
451 xbt_fifo_foreach(synchro->simcalls, item, simcall, smx_simcall_t) {
453 switch (synchro->state) {
456 /* do nothing, synchro done */
457 XBT_DEBUG("SIMIX_execution_finished: execution successful");
461 XBT_DEBUG("SIMIX_execution_finished: host '%s' failed", sg_host_get_name(simcall->issuer->host));
462 simcall->issuer->context->iwannadie = 1;
463 SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
467 XBT_DEBUG("SIMIX_execution_finished: execution canceled");
468 SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Canceled");
472 xbt_die("Internal error in SIMIX_execution_finish: unexpected synchro state %d",
473 (int)synchro->state);
475 /* check if the host is down */
476 if (simcall->issuer->host->is_off()) {
477 simcall->issuer->context->iwannadie = 1;
480 simcall->issuer->waiting_synchro = NULL;
481 simcall_execution_wait__set__result(simcall, synchro->state);
482 SIMIX_simcall_answer(simcall);
485 /* We no longer need it */
486 SIMIX_execution_destroy(synchro);
490 void SIMIX_post_host_execute(smx_synchro_t synchro)
492 if (synchro->type == SIMIX_SYNC_EXECUTE && /* FIMXE: handle resource failure
493 * for parallel tasks too */
494 synchro->execution.host->is_off()) {
495 /* If the host running the synchro failed, notice it so that the asking
496 * process can be killed if it runs on that host itself */
497 synchro->state = SIMIX_FAILED;
498 } else if (synchro->execution.surf_exec->getState() == SURF_ACTION_FAILED) {
499 /* If the host running the synchro didn't fail, then the synchro was
501 synchro->state = SIMIX_CANCELED;
503 synchro->state = SIMIX_DONE;
506 if (synchro->execution.surf_exec) {
507 synchro->execution.surf_exec->unref();
508 synchro->execution.surf_exec = NULL;
511 /* If there are simcalls associated with the synchro, then answer them */
512 if (xbt_fifo_size(synchro->simcalls)) {
513 SIMIX_execution_finish(synchro);
518 void SIMIX_set_category(smx_synchro_t synchro, const char *category)
520 if (synchro->state != SIMIX_RUNNING) return;
521 if (synchro->type == SIMIX_SYNC_EXECUTE){
522 synchro->execution.surf_exec->setCategory(category);
523 }else if (synchro->type == SIMIX_SYNC_COMMUNICATE){
524 synchro->comm.surf_comm->setCategory(category);