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 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_host, simix,
17 static void SIMIX_execution_finish(smx_synchro_t synchro);
20 * \brief Internal function to create a SIMIX host.
21 * \param name name of the host to create
23 void SIMIX_host_create(sg_host_t host) // FIXME: braindead prototype. Take sg_host as parameter
25 smx_host_priv_t smx_host = xbt_new0(s_smx_host_priv_t, 1);
29 smx_host->process_list = xbt_swag_new(xbt_swag_offset(proc, host_proc_hookup));
31 /* Update global variables */
32 sg_host_simix_set(host, smx_host);
36 * \brief Start the host if it is off
39 void SIMIX_host_on(sg_host_t h)
41 smx_host_priv_t host = sg_host_simix(h);
43 xbt_assert((host != NULL), "Invalid parameters");
46 simgrid::surf::HostImpl* surf_host = h->extension<simgrid::surf::HostImpl>();
50 smx_process_arg_t arg;
51 xbt_dynar_foreach(host->boot_processes,cpt,arg) {
53 char** argv = xbt_new(char*, arg->argc);
54 for (int i=0; i<arg->argc; i++)
55 argv[i] = xbt_strdup(arg->argv[i]);
57 XBT_DEBUG("Booting Process %s(%s) right now", arg->argv[0], arg->hostname);
58 if (simix_global->create_process_function) {
59 simix_global->create_process_function(argv[0],
70 simcall_process_create(arg->argv[0],
85 * \brief Stop the host if it is on
88 void SIMIX_host_off(sg_host_t h, smx_process_t issuer)
90 smx_host_priv_t host = sg_host_simix(h);
92 xbt_assert((host != NULL), "Invalid parameters");
95 simgrid::surf::HostImpl* surf_host = h->extension<simgrid::surf::HostImpl>();
98 /* Clean Simulator data */
99 if (xbt_swag_size(host->process_list) != 0) {
100 smx_process_t process = NULL;
101 xbt_swag_foreach(process, host->process_list) {
102 SIMIX_process_kill(process, issuer);
103 XBT_DEBUG("Killing %s on %s by %s", process->name, sg_host_get_name(process->host), issuer->name);
107 XBT_INFO("Host %s is already off",h->name().c_str());
112 * \brief Internal function to destroy a SIMIX host.
114 * \param h the host to destroy (a sg_host_t)
116 void SIMIX_host_destroy(void *h)
118 smx_host_priv_t host = (smx_host_priv_t) h;
120 xbt_assert((host != NULL), "Invalid parameters");
122 /* Clean Simulator data */
123 if (xbt_swag_size(host->process_list) != 0) {
124 char *msg = xbt_strdup("Shutting down host, but it's not empty:");
126 smx_process_t process = NULL;
128 xbt_swag_foreach(process, host->process_list) {
129 tmp = bprintf("%s\n\t%s", msg, process->name);
133 SIMIX_display_process_status();
134 THROWF(arg_error, 0, "%s", msg);
136 xbt_dynar_free(&host->auto_restart_processes);
137 xbt_dynar_free(&host->boot_processes);
138 xbt_swag_free(host->process_list);
140 /* Clean host structure */
145 sg_host_t SIMIX_host_self(void)
147 smx_process_t process = SIMIX_process_self();
148 return (process == NULL) ? NULL : SIMIX_process_get_host(process);
151 /* needs to be public and without simcall because it is called
152 by exceptions and logging events */
153 const char* SIMIX_host_self_get_name(void)
155 sg_host_t host = SIMIX_host_self();
156 if (host == NULL || SIMIX_process_self() == simix_global->maestro_process)
159 return sg_host_get_name(host);
162 void _SIMIX_host_free_process_arg(void *data)
164 smx_process_arg_t arg = *(smx_process_arg_t*)data;
166 for (i = 0; i < arg->argc; i++)
167 xbt_free(arg->argv[i]);
173 * \brief Add a process to the list of the processes that the host will restart when it comes back
174 * This function add a process to the list of the processes that will be restarted when the host comes
175 * back. It is expected that this function is called when the host is down.
176 * The processes will only be restarted once, meaning that you will have to register the process
177 * again to restart the process again.
179 void SIMIX_host_add_auto_restart_process(sg_host_t host,
181 xbt_main_func_t code,
183 const char *hostname,
185 int argc, char **argv,
186 xbt_dict_t properties,
189 if (!sg_host_simix(host)->auto_restart_processes) {
190 sg_host_simix(host)->auto_restart_processes = xbt_dynar_new(sizeof(smx_process_arg_t),_SIMIX_host_free_process_arg);
192 smx_process_arg_t arg = xbt_new(s_smx_process_arg_t,1);
193 arg->name = xbt_strdup(name);
196 arg->hostname = hostname;
197 arg->kill_time = kill_time;
200 arg->argv = xbt_new(char*,argc + 1);
202 for (int i = 0; i < argc; i++)
203 arg->argv[i] = xbt_strdup(argv[i]);
204 arg->argv[argc] = NULL;
206 arg->properties = properties;
207 arg->auto_restart = auto_restart;
209 if( host->isOff() && !xbt_dict_get_or_null(watched_hosts_lib,sg_host_get_name(host))){
210 xbt_dict_set(watched_hosts_lib,sg_host_get_name(host),host,NULL);
211 XBT_DEBUG("Push host %s to watched_hosts_lib because state == SURF_RESOURCE_OFF",sg_host_get_name(host));
213 xbt_dynar_push_as(sg_host_simix(host)->auto_restart_processes,smx_process_arg_t,arg);
216 * \brief Restart the list of processes that have been registered to the host
218 void SIMIX_host_autorestart(sg_host_t host)
221 smx_process_arg_t arg;
222 xbt_dynar_t process_list = sg_host_simix(host)->auto_restart_processes;
226 xbt_dynar_foreach (process_list, cpt, arg) {
228 XBT_DEBUG("Restarting Process %s(%s) right now", arg->argv[0], arg->hostname);
229 if (simix_global->create_process_function) {
230 simix_global->create_process_function(arg->argv[0],
241 simcall_process_create(arg->argv[0],
242 (xbt_main_func_t) arg->code,
252 /* arg->argv is used by the process created above. Hide it to
253 * _SIMIX_host_free_process_arg() which is called by xbt_dynar_reset()
258 xbt_dynar_reset(process_list);
261 smx_synchro_t simcall_HANDLER_execution_start(smx_simcall_t simcall,
262 const char* name, double flops_amount, double priority, double bound, unsigned long affinity_mask) {
263 return SIMIX_execution_start(simcall->issuer, name,flops_amount,priority,bound,affinity_mask);
265 smx_synchro_t SIMIX_execution_start(smx_process_t issuer, const char *name,
266 double flops_amount, double priority, double bound, unsigned long affinity_mask){
268 /* alloc structures and initialize */
269 smx_synchro_t synchro = (smx_synchro_t) xbt_mallocator_get(simix_global->synchro_mallocator);
270 synchro->type = SIMIX_SYNC_EXECUTE;
271 synchro->name = xbt_strdup(name);
272 synchro->state = SIMIX_RUNNING;
273 synchro->execution.host = issuer->host;
274 synchro->category = NULL;
276 /* set surf's action */
277 if (!MC_is_active() && !MC_record_replay_is_active()) {
279 synchro->execution.surf_exec = issuer->host->pimpl_cpu->execution_start(flops_amount);
280 synchro->execution.surf_exec->setData(synchro);
281 synchro->execution.surf_exec->setPriority(priority);
284 static_cast<simgrid::surf::CpuAction*>(synchro->execution.surf_exec)
287 if (affinity_mask != 0) {
288 /* just a double check to confirm that this host is the host where this task is running. */
289 xbt_assert(synchro->execution.host == issuer->host);
290 static_cast<simgrid::surf::CpuAction*>(synchro->execution.surf_exec)
291 ->setAffinity(issuer->host->pimpl_cpu, affinity_mask);
295 XBT_DEBUG("Create execute synchro %p: %s", synchro, synchro->name);
300 smx_synchro_t SIMIX_execution_parallel_start(const char *name,
301 int host_nb, sg_host_t *host_list,
302 double *flops_amount, double *bytes_amount,
303 double amount, double rate){
305 sg_host_t*host_list_cpy = NULL;
308 /* alloc structures and initialize */
309 smx_synchro_t synchro = (smx_synchro_t) xbt_mallocator_get(simix_global->synchro_mallocator);
310 synchro->type = SIMIX_SYNC_PARALLEL_EXECUTE;
311 synchro->name = xbt_strdup(name);
312 synchro->state = SIMIX_RUNNING;
313 synchro->execution.host = NULL; /* FIXME: do we need the list of hosts? */
314 synchro->category = NULL;
316 /* set surf's synchro */
317 host_list_cpy = xbt_new0(sg_host_t, host_nb);
318 for (i = 0; i < host_nb; i++)
319 host_list_cpy[i] = host_list[i];
321 /* Check that we are not mixing VMs and PMs in the parallel task */
322 simgrid::surf::HostImpl *host = host_list[0]->extension<simgrid::surf::HostImpl>();
323 bool is_a_vm = (nullptr != dynamic_cast<simgrid::surf::VirtualMachine*>(host));
324 for (i = 1; i < host_nb; i++) {
325 bool tmp_is_a_vm = (nullptr != dynamic_cast<simgrid::surf::VirtualMachine*>(host_list[i]->extension<simgrid::surf::HostImpl>()));
326 xbt_assert(is_a_vm == tmp_is_a_vm, "parallel_execute: mixing VMs and PMs is not supported (yet).");
329 /* set surf's synchro */
330 if (!MC_is_active() && !MC_record_replay_is_active()) {
331 synchro->execution.surf_exec =
332 surf_host_model->executeParallelTask(
333 host_nb, host_list_cpy, flops_amount, bytes_amount, rate);
335 synchro->execution.surf_exec->setData(synchro);
337 XBT_DEBUG("Create parallel execute synchro %p", synchro);
342 void SIMIX_execution_destroy(smx_synchro_t synchro)
344 XBT_DEBUG("Destroy synchro %p", synchro);
346 if (synchro->execution.surf_exec) {
347 synchro->execution.surf_exec->unref();
348 synchro->execution.surf_exec = NULL;
350 xbt_free(synchro->name);
351 xbt_mallocator_release(simix_global->synchro_mallocator, synchro);
354 void SIMIX_execution_cancel(smx_synchro_t synchro)
356 XBT_DEBUG("Cancel synchro %p", synchro);
358 if (synchro->execution.surf_exec)
359 synchro->execution.surf_exec->cancel();
362 double SIMIX_execution_get_remains(smx_synchro_t synchro)
366 if (synchro->state == SIMIX_RUNNING)
367 result = synchro->execution.surf_exec->getRemains();
372 e_smx_state_t SIMIX_execution_get_state(smx_synchro_t synchro)
374 return synchro->state;
377 void SIMIX_execution_set_priority(smx_synchro_t synchro, double priority)
379 if(synchro->execution.surf_exec)
380 synchro->execution.surf_exec->setPriority(priority);
383 void SIMIX_execution_set_bound(smx_synchro_t synchro, double bound)
385 if(synchro->execution.surf_exec)
386 static_cast<simgrid::surf::CpuAction*>(synchro->execution.surf_exec)->setBound(bound);
389 void SIMIX_execution_set_affinity(smx_synchro_t synchro, sg_host_t host, unsigned long mask)
391 xbt_assert(synchro->type == SIMIX_SYNC_EXECUTE);
393 if (synchro->execution.surf_exec) {
394 /* just a double check to confirm that this host is the host where this task is running. */
395 xbt_assert(synchro->execution.host == host);
396 static_cast<simgrid::surf::CpuAction*>(synchro->execution.surf_exec)
397 ->setAffinity(host->pimpl_cpu, mask);
401 void simcall_HANDLER_execution_wait(smx_simcall_t simcall, smx_synchro_t synchro)
404 XBT_DEBUG("Wait for execution of synchro %p, state %d", synchro, (int)synchro->state);
406 /* Associate this simcall to the synchro */
407 xbt_fifo_push(synchro->simcalls, simcall);
408 simcall->issuer->waiting_synchro = synchro;
410 /* set surf's synchro */
411 if (MC_is_active() || MC_record_replay_is_active()) {
412 synchro->state = SIMIX_DONE;
413 SIMIX_execution_finish(synchro);
417 /* If the synchro is already finished then perform the error handling */
418 if (synchro->state != SIMIX_RUNNING)
419 SIMIX_execution_finish(synchro);
422 void SIMIX_execution_suspend(smx_synchro_t synchro)
424 if(synchro->execution.surf_exec)
425 synchro->execution.surf_exec->suspend();
428 void SIMIX_execution_resume(smx_synchro_t synchro)
430 if(synchro->execution.surf_exec)
431 synchro->execution.surf_exec->resume();
434 void SIMIX_execution_finish(smx_synchro_t synchro)
436 xbt_fifo_item_t item;
437 smx_simcall_t simcall;
439 xbt_fifo_foreach(synchro->simcalls, item, simcall, smx_simcall_t) {
441 switch (synchro->state) {
444 /* do nothing, synchro done */
445 XBT_DEBUG("SIMIX_execution_finished: execution successful");
449 XBT_DEBUG("SIMIX_execution_finished: host '%s' failed", sg_host_get_name(simcall->issuer->host));
450 simcall->issuer->context->iwannadie = 1;
451 SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
455 XBT_DEBUG("SIMIX_execution_finished: execution canceled");
456 SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Canceled");
460 xbt_die("Internal error in SIMIX_execution_finish: unexpected synchro state %d",
461 (int)synchro->state);
463 /* check if the host is down */
464 if (simcall->issuer->host->isOff()) {
465 simcall->issuer->context->iwannadie = 1;
468 simcall->issuer->waiting_synchro = NULL;
469 simcall_execution_wait__set__result(simcall, synchro->state);
470 SIMIX_simcall_answer(simcall);
473 /* We no longer need it */
474 SIMIX_execution_destroy(synchro);
478 void SIMIX_post_host_execute(smx_synchro_t synchro)
480 if (synchro->type == SIMIX_SYNC_EXECUTE && /* FIMXE: handle resource failure
481 * for parallel tasks too */
482 synchro->execution.host->isOff()) {
483 /* If the host running the synchro failed, notice it so that the asking
484 * process can be killed if it runs on that host itself */
485 synchro->state = SIMIX_FAILED;
486 } else if (synchro->execution.surf_exec->getState() == simgrid::surf::Action::State::failed) {
487 /* If the host running the synchro didn't fail, then the synchro was
489 synchro->state = SIMIX_CANCELED;
491 synchro->state = SIMIX_DONE;
494 if (synchro->execution.surf_exec) {
495 synchro->execution.surf_exec->unref();
496 synchro->execution.surf_exec = NULL;
499 /* If there are simcalls associated with the synchro, then answer them */
500 if (xbt_fifo_size(synchro->simcalls)) {
501 SIMIX_execution_finish(synchro);
506 void SIMIX_set_category(smx_synchro_t synchro, const char *category)
508 if (synchro->state != SIMIX_RUNNING) return;
509 if (synchro->type == SIMIX_SYNC_EXECUTE){
510 synchro->execution.surf_exec->setCategory(category);
511 }else if (synchro->type == SIMIX_SYNC_COMMUNICATE){
512 synchro->comm.surf_comm->setCategory(category);