Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
91b7ef9f731a480c043180f4fa94ba2ac0cdc9b5
[simgrid.git] / 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/sysdep.h"
9 #include "xbt/log.h"
10 #include "xbt/dict.h"
11 #include "mc/mc.h"
12 #include "src/mc/mc_replay.h"
13 #include "src/surf/host_interface.hpp"
14
15 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_host, simix,
16                                 "SIMIX hosts");
17
18 static void SIMIX_execution_finish(smx_synchro_t synchro);
19
20 /**
21  * \brief Internal function to create a SIMIX host.
22  * \param name name of the host to create
23  */
24 void SIMIX_host_create(sg_host_t host) // FIXME: braindead prototype. Take sg_host as parameter
25 {
26   smx_host_priv_t smx_host = xbt_new0(s_smx_host_priv_t, 1);
27   s_smx_process_t proc;
28
29   /* Host structure */
30   smx_host->process_list =
31       xbt_swag_new(xbt_swag_offset(proc, host_proc_hookup));
32
33   /* Update global variables */
34   sg_host_simix_set(host, smx_host);
35 }
36
37 /**
38  * \brief Start the host if it is off
39  *
40  */
41 void SIMIX_host_on(sg_host_t h)
42 {
43   smx_host_priv_t host = sg_host_simix(h);
44
45   xbt_assert((host != NULL), "Invalid parameters");
46
47   if (h->isOff()) {
48     simgrid::surf::Host* surf_host = h->extension<simgrid::surf::Host>();
49     surf_host->turnOn();
50
51     unsigned int cpt;
52     smx_process_arg_t arg;
53     xbt_dynar_foreach(host->boot_processes,cpt,arg) {
54
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]);
58
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],
62                                               arg->code,
63                                               NULL,
64                                               arg->hostname,
65                                               arg->kill_time,
66                                               arg->argc,
67                                               argv,
68                                               arg->properties,
69                                               arg->auto_restart,
70                                               NULL);
71       } else {
72         simcall_process_create(arg->argv[0],
73                                arg->code,
74                                NULL,
75                                arg->hostname,
76                                arg->kill_time,
77                                arg->argc,
78                                argv,
79                                arg->properties,
80                                arg->auto_restart);
81       }
82     }
83   }
84 }
85
86 void simcall_HANDLER_host_off(smx_simcall_t simcall, sg_host_t h)
87 {
88  SIMIX_host_off(h, simcall->issuer);
89 }
90
91 /**
92  * \brief Stop the host if it is on
93  *
94  */
95 void SIMIX_host_off(sg_host_t h, smx_process_t issuer)
96 {
97   smx_host_priv_t host = sg_host_simix(h);
98
99   xbt_assert((host != NULL), "Invalid parameters");
100
101   if (h->isOn()) {
102     simgrid::surf::Host* surf_host = h->extension<simgrid::surf::Host>();
103     surf_host->turnOff();
104
105     /* Clean Simulator data */
106     if (xbt_swag_size(host->process_list) != 0) {
107       smx_process_t process = NULL;
108       xbt_swag_foreach(process, host->process_list) {
109         SIMIX_process_kill(process, issuer);
110         XBT_DEBUG("Killing %s on %s by %s", process->name,  sg_host_get_name(process->host), issuer->name);
111       }
112     }
113   } else {
114     XBT_INFO("Host %s is already off",h->getName().c_str());
115   }
116 }
117
118 /**
119  * \brief Internal function to destroy a SIMIX host.
120  *
121  * \param h the host to destroy (a sg_host_t)
122  */
123 void SIMIX_host_destroy(void *h)
124 {
125   smx_host_priv_t host = (smx_host_priv_t) h;
126
127   xbt_assert((host != NULL), "Invalid parameters");
128
129   /* Clean Simulator data */
130   if (xbt_swag_size(host->process_list) != 0) {
131     char *msg = xbt_strdup("Shutting down host, but it's not empty:");
132     char *tmp;
133     smx_process_t process = NULL;
134
135     xbt_swag_foreach(process, host->process_list) {
136       tmp = bprintf("%s\n\t%s", msg, process->name);
137       free(msg);
138       msg = tmp;
139     }
140     SIMIX_display_process_status();
141     THROWF(arg_error, 0, "%s", msg);
142   }
143   xbt_dynar_free(&host->auto_restart_processes);
144   xbt_dynar_free(&host->boot_processes);
145   xbt_swag_free(host->process_list);
146
147   /* Clean host structure */
148   free(host);
149   return;
150 }
151
152 sg_host_t SIMIX_host_self(void)
153 {
154   smx_process_t process = SIMIX_process_self();
155   return (process == NULL) ? NULL : SIMIX_process_get_host(process);
156 }
157
158 /* needs to be public and without simcall because it is called
159    by exceptions and logging events */
160 const char* SIMIX_host_self_get_name(void)
161 {
162   sg_host_t host = SIMIX_host_self();
163   if (host == NULL || SIMIX_process_self() == simix_global->maestro_process)
164     return "";
165
166   return SIMIX_host_get_name(host);
167 }
168
169 xbt_dict_t SIMIX_host_get_properties(sg_host_t host){
170   return sg_host_get_properties(host);
171 }
172
173
174 xbt_swag_t SIMIX_host_get_process_list(sg_host_t host){
175   smx_host_priv_t host_priv = sg_host_simix(host);
176
177   return host_priv->process_list;
178 }
179
180
181 double SIMIX_host_get_current_power_peak(sg_host_t host) {
182           return surf_host_get_current_power_peak(host);
183 }
184
185 double SIMIX_host_get_power_peak_at(sg_host_t host, int pstate_index) {
186           return surf_host_get_power_peak_at(host, pstate_index);
187 }
188
189 void _SIMIX_host_free_process_arg(void *data)
190 {
191   smx_process_arg_t arg = *(smx_process_arg_t*)data;
192   int i;
193   for (i = 0; i < arg->argc; i++)
194     xbt_free(arg->argv[i]);
195   xbt_free(arg->argv);
196   xbt_free(arg->name);
197   xbt_free(arg);
198 }
199 /**
200  * \brief Add a process to the list of the processes that the host will restart when it comes back
201  * This function add a process to the list of the processes that will be restarted when the host comes
202  * back. It is expected that this function is called when the host is down.
203  * The processes will only be restarted once, meaning that you will have to register the process
204  * again to restart the process again.
205  */
206 void SIMIX_host_add_auto_restart_process(sg_host_t host,
207                                          const char *name,
208                                          xbt_main_func_t code,
209                                          void *data,
210                                          const char *hostname,
211                                          double kill_time,
212                                          int argc, char **argv,
213                                          xbt_dict_t properties,
214                                          int auto_restart)
215 {
216   if (!sg_host_simix(host)->auto_restart_processes) {
217     sg_host_simix(host)->auto_restart_processes = xbt_dynar_new(sizeof(smx_process_arg_t),_SIMIX_host_free_process_arg);
218   }
219   smx_process_arg_t arg = xbt_new(s_smx_process_arg_t,1);
220   arg->name = xbt_strdup(name);
221   arg->code = code;
222   arg->data = data;
223   arg->hostname = hostname;
224   arg->kill_time = kill_time;
225   arg->argc = argc;
226
227   arg->argv = xbt_new(char*,argc + 1);
228
229   int i;
230   for (i = 0; i < argc; i++) {
231     arg->argv[i] = xbt_strdup(argv[i]);
232   }
233   arg->argv[argc] = NULL;
234
235   arg->properties = properties;
236   arg->auto_restart = auto_restart;
237
238   if( ! sg_host_is_on(host)
239       && !xbt_dict_get_or_null(watched_hosts_lib,sg_host_get_name(host))){
240     xbt_dict_set(watched_hosts_lib,sg_host_get_name(host),host,NULL);
241     XBT_DEBUG("Have pushed host %s to watched_hosts_lib because state == SURF_RESOURCE_OFF",sg_host_get_name(host));
242   }
243   xbt_dynar_push_as(sg_host_simix(host)->auto_restart_processes,smx_process_arg_t,arg);
244 }
245 /**
246  * \brief Restart the list of processes that have been registered to the host
247  */
248 void SIMIX_host_restart_processes(sg_host_t host)
249 {
250   unsigned int cpt;
251   smx_process_arg_t arg;
252   xbt_dynar_t process_list = sg_host_simix(host)->auto_restart_processes;
253   if (!process_list)
254     return;
255
256   xbt_dynar_foreach (process_list, cpt, arg) {
257
258     XBT_DEBUG("Restarting Process %s(%s) right now", arg->argv[0], arg->hostname);
259     if (simix_global->create_process_function) {
260       simix_global->create_process_function(arg->argv[0],
261                                             arg->code,
262                                             NULL,
263                                             arg->hostname,
264                                             arg->kill_time,
265                                             arg->argc,
266                                             arg->argv,
267                                             arg->properties,
268                                             arg->auto_restart,
269                                             NULL);
270     } else {
271       simcall_process_create(arg->argv[0],
272                              (xbt_main_func_t) arg->code,
273                              NULL,
274                              arg->hostname,
275                              arg->kill_time,
276                              arg->argc,
277                              arg->argv,
278                              arg->properties,
279                              arg->auto_restart);
280
281     }
282     /* arg->argv is used by the process created above.  Hide it to
283      * _SIMIX_host_free_process_arg() which is called by xbt_dynar_reset()
284      * below. */
285     arg->argc = 0;
286     arg->argv = NULL;
287   }
288   xbt_dynar_reset(process_list);
289 }
290
291 void SIMIX_host_autorestart(sg_host_t host)
292 {
293   if(simix_global->autorestart)
294     simix_global->autorestart(host);
295   else
296     xbt_die("No function for simix_global->autorestart");
297 }
298 smx_synchro_t simcall_HANDLER_process_execute(smx_simcall_t simcall,
299                 const char* name, double flops_amount, double priority, double bound, unsigned long affinity_mask) {
300         return SIMIX_process_execute(simcall->issuer, name,flops_amount,priority,bound,affinity_mask);
301 }
302 smx_synchro_t SIMIX_process_execute(smx_process_t issuer, const char *name,
303      double flops_amount, double priority, double bound, unsigned long affinity_mask){
304
305   /* alloc structures and initialize */
306   smx_synchro_t synchro = (smx_synchro_t) xbt_mallocator_get(simix_global->synchro_mallocator);
307   synchro->type = SIMIX_SYNC_EXECUTE;
308   synchro->name = xbt_strdup(name);
309   synchro->state = SIMIX_RUNNING;
310   synchro->execution.host = issuer->host;
311   synchro->category = NULL;
312
313   /* set surf's action */
314   if (!MC_is_active() && !MC_record_replay_is_active()) {
315
316     synchro->execution.surf_exec = issuer->host->p_cpu->execute(flops_amount);
317     synchro->execution.surf_exec->setData(synchro);
318     synchro->execution.surf_exec->setPriority(priority);
319
320     if (bound != 0)
321       static_cast<simgrid::surf::CpuAction*>(synchro->execution.surf_exec)
322         ->setBound(bound);
323
324     if (affinity_mask != 0) {
325       /* just a double check to confirm that this host is the host where this task is running. */
326       xbt_assert(synchro->execution.host == issuer->host);
327       static_cast<simgrid::surf::CpuAction*>(synchro->execution.surf_exec)
328         ->setAffinity(issuer->host->p_cpu, affinity_mask);
329     }
330   }
331
332   XBT_DEBUG("Create execute synchro %p: %s", synchro, synchro->name);
333
334   return synchro;
335 }
336
337 smx_synchro_t SIMIX_process_parallel_execute(const char *name,
338     int host_nb, sg_host_t *host_list,
339     double *flops_amount, double *bytes_amount,
340     double amount, double rate){
341
342   sg_host_t*host_list_cpy = NULL;
343   int i;
344
345   /* alloc structures and initialize */
346   smx_synchro_t synchro = (smx_synchro_t) xbt_mallocator_get(simix_global->synchro_mallocator);
347   synchro->type = SIMIX_SYNC_PARALLEL_EXECUTE;
348   synchro->name = xbt_strdup(name);
349   synchro->state = SIMIX_RUNNING;
350   synchro->execution.host = NULL; /* FIXME: do we need the list of hosts? */
351   synchro->category = NULL;
352
353   /* set surf's synchro */
354   host_list_cpy = xbt_new0(sg_host_t, host_nb);
355   for (i = 0; i < host_nb; i++)
356     host_list_cpy[i] = host_list[i];
357
358
359   /* FIXME: what happens if host_list contains VMs and PMs. If
360    * execute_parallel_task() does not change the state of the model, we can mix
361    * them. */
362   surf_host_model_t ws_model = surf_host_get_model(host_list[0]);
363   for (i = 1; i < host_nb; i++) {
364     surf_host_model_t ws_model_tmp = surf_host_get_model(host_list[0]);
365     if (ws_model_tmp != ws_model) {
366       XBT_CRITICAL("mixing VMs and PMs is not supported");
367       DIE_IMPOSSIBLE;
368     }
369   }
370
371   /* set surf's synchro */
372   if (!MC_is_active() && !MC_record_replay_is_active()) {
373     synchro->execution.surf_exec =
374       surf_host_model_execute_parallel_task(surf_host_model,
375                   host_nb, host_list_cpy, flops_amount, bytes_amount, rate);
376
377     synchro->execution.surf_exec->setData(synchro);
378   }
379   XBT_DEBUG("Create parallel execute synchro %p", synchro);
380
381   return synchro;
382 }
383
384 void SIMIX_process_execution_destroy(smx_synchro_t synchro){
385   XBT_DEBUG("Destroy synchro %p", synchro);
386
387   if (synchro->execution.surf_exec) {
388     synchro->execution.surf_exec->unref();
389     synchro->execution.surf_exec = NULL;
390   }
391   xbt_free(synchro->name);
392   xbt_mallocator_release(simix_global->synchro_mallocator, synchro);
393 }
394
395 void SIMIX_process_execution_cancel(smx_synchro_t synchro){
396   XBT_DEBUG("Cancel synchro %p", synchro);
397
398   if (synchro->execution.surf_exec)
399     synchro->execution.surf_exec->cancel();
400 }
401
402 double SIMIX_process_execution_get_remains(smx_synchro_t synchro){
403   double result = 0.0;
404
405   if (synchro->state == SIMIX_RUNNING)
406     result = surf_action_get_remains(synchro->execution.surf_exec);
407
408   return result;
409 }
410
411 e_smx_state_t SIMIX_process_execution_get_state(smx_synchro_t synchro){
412   return synchro->state;
413 }
414
415 void SIMIX_process_execution_set_priority(smx_synchro_t synchro, double priority){
416
417   if(synchro->execution.surf_exec)
418         synchro->execution.surf_exec->setPriority(priority);
419 }
420
421 void SIMIX_process_execution_set_bound(smx_synchro_t synchro, double bound){
422
423   if(synchro->execution.surf_exec)
424         static_cast<simgrid::surf::CpuAction*>(synchro->execution.surf_exec)
425     ->setBound(bound);
426 }
427
428 void SIMIX_process_execution_set_affinity(smx_synchro_t synchro, sg_host_t host, unsigned long mask){
429   xbt_assert(synchro->type == SIMIX_SYNC_EXECUTE);
430
431   if (synchro->execution.surf_exec) {
432     /* just a double check to confirm that this host is the host where this task is running. */
433     xbt_assert(synchro->execution.host == host);
434     static_cast<simgrid::surf::CpuAction*>(synchro->execution.surf_exec)
435       ->setAffinity(host->p_cpu, mask);
436   }
437 }
438
439 void simcall_HANDLER_process_execution_wait(smx_simcall_t simcall, smx_synchro_t synchro){
440
441   XBT_DEBUG("Wait for execution of synchro %p, state %d", synchro, (int)synchro->state);
442
443   /* Associate this simcall to the synchro */
444   xbt_fifo_push(synchro->simcalls, simcall);
445   simcall->issuer->waiting_synchro = synchro;
446
447   /* set surf's synchro */
448   if (MC_is_active() || MC_record_replay_is_active()) {
449     synchro->state = SIMIX_DONE;
450     SIMIX_execution_finish(synchro);
451     return;
452   }
453
454   /* If the synchro is already finished then perform the error handling */
455   if (synchro->state != SIMIX_RUNNING)
456     SIMIX_execution_finish(synchro);
457 }
458
459 void SIMIX_host_execution_suspend(smx_synchro_t synchro)
460 {
461   if(synchro->execution.surf_exec)
462     synchro->execution.surf_exec->suspend();
463 }
464
465 void SIMIX_host_execution_resume(smx_synchro_t synchro)
466 {
467   if(synchro->execution.surf_exec)
468     synchro->execution.surf_exec->resume();
469 }
470
471 void SIMIX_execution_finish(smx_synchro_t synchro)
472 {
473   xbt_fifo_item_t item;
474   smx_simcall_t simcall;
475
476   xbt_fifo_foreach(synchro->simcalls, item, simcall, smx_simcall_t) {
477
478     switch (synchro->state) {
479
480       case SIMIX_DONE:
481         /* do nothing, synchro done */
482         XBT_DEBUG("SIMIX_execution_finished: execution successful");
483         break;
484
485       case SIMIX_FAILED:
486         XBT_DEBUG("SIMIX_execution_finished: host '%s' failed", sg_host_get_name(simcall->issuer->host));
487         simcall->issuer->context->iwannadie = 1;
488         SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
489         break;
490
491       case SIMIX_CANCELED:
492         XBT_DEBUG("SIMIX_execution_finished: execution canceled");
493         SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Canceled");
494         break;
495
496       default:
497         xbt_die("Internal error in SIMIX_execution_finish: unexpected synchro state %d",
498             (int)synchro->state);
499     }
500     /* check if the host is down */
501     if (simcall->issuer->host->isOff()) {
502       simcall->issuer->context->iwannadie = 1;
503     }
504
505     simcall->issuer->waiting_synchro =    NULL;
506     simcall_process_execution_wait__set__result(simcall, synchro->state);
507     SIMIX_simcall_answer(simcall);
508   }
509
510   /* We no longer need it */
511   SIMIX_process_execution_destroy(synchro);
512 }
513
514
515 void SIMIX_post_host_execute(smx_synchro_t synchro)
516 {
517   if (synchro->type == SIMIX_SYNC_EXECUTE && /* FIMXE: handle resource failure
518                                                * for parallel tasks too */
519       synchro->execution.host->isOff()) {
520     /* If the host running the synchro failed, notice it so that the asking
521      * process can be killed if it runs on that host itself */
522     synchro->state = SIMIX_FAILED;
523   } else if (synchro->execution.surf_exec->getState() == SURF_ACTION_FAILED) {
524     /* If the host running the synchro didn't fail, then the synchro was
525      * canceled */
526     synchro->state = SIMIX_CANCELED;
527   } else {
528     synchro->state = SIMIX_DONE;
529   }
530
531   if (synchro->execution.surf_exec) {
532     synchro->execution.surf_exec->unref();
533     synchro->execution.surf_exec = NULL;
534   }
535
536   /* If there are simcalls associated with the synchro, then answer them */
537   if (xbt_fifo_size(synchro->simcalls)) {
538     SIMIX_execution_finish(synchro);
539   }
540 }
541
542
543 void SIMIX_set_category(smx_synchro_t synchro, const char *category)
544 {
545   if (synchro->state != SIMIX_RUNNING) return;
546   if (synchro->type == SIMIX_SYNC_EXECUTE){
547     synchro->execution.surf_exec->setCategory(category);
548   }else if (synchro->type == SIMIX_SYNC_COMMUNICATE){
549     synchro->comm.surf_comm->setCategory(category);
550   }
551 }
552
553 /**
554  * \brief Function to get the parameters of the given the SIMIX host.
555  *
556  * \param host the host to get_phys_host (a sg_host_t)
557  * \param param the parameter object space to be overwritten (a ws_params_t)
558  */
559 void SIMIX_host_get_params(sg_host_t ind_vm, vm_params_t params)
560 {
561   ind_vm->extension<simgrid::surf::Host>()->getParams(params);
562 }
563
564 void SIMIX_host_set_params(sg_host_t ind_vm, vm_params_t params)
565 {
566   ind_vm->extension<simgrid::surf::Host>()->setParams(params);
567 }
568
569 xbt_dict_t SIMIX_host_get_mounted_storage_list(sg_host_t host)
570 {
571   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
572   return host->extension<simgrid::surf::Host>()->getMountedStorageList();
573 }
574
575 xbt_dynar_t SIMIX_host_get_attached_storage_list(sg_host_t host)
576 {
577   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
578   return host->extension<simgrid::surf::Host>()->getAttachedStorageList();
579 }