Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
further energy interface cleanups (see changelog)
[simgrid.git] / src / simix / smx_host.c
1 /* Copyright (c) 2007-2014. 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 "mc/mc_replay.h"
13
14 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_host, simix,
15                                 "SIMIX hosts");
16
17 static void SIMIX_execution_finish(smx_synchro_t synchro);
18
19 /**
20  * \brief Internal function to create a SIMIX host.
21  * \param name name of the host to create
22  * \param workstation the SURF workstation to encapsulate
23  * \param data some user data (may be NULL)
24  */
25 smx_host_t SIMIX_host_create(const char *name,
26                                void *workstation, void *data)
27 {
28   smx_host_priv_t smx_host = xbt_new0(s_smx_host_priv_t, 1);
29   s_smx_process_t proc;
30
31   /* Host structure */
32   smx_host->process_list =
33       xbt_swag_new(xbt_swag_offset(proc, host_proc_hookup));
34
35   /* Update global variables */
36   xbt_lib_set(host_lib,name,SIMIX_HOST_LEVEL,smx_host);
37
38   return xbt_lib_get_elm_or_null(host_lib, name);
39 }
40
41 /**
42  * \brief Start the host if it is off
43  *
44  */
45 void SIMIX_host_on(smx_host_t h)
46 {
47   smx_host_priv_t host = SIMIX_host_priv(h);
48
49   xbt_assert((host != NULL), "Invalid parameters");
50
51   if (surf_resource_get_state(surf_workstation_resource_priv(h))==SURF_RESOURCE_OFF) {
52     surf_resource_set_state(surf_workstation_resource_priv(h), SURF_RESOURCE_ON);
53
54     unsigned int cpt;
55     smx_process_arg_t arg;
56     xbt_dynar_foreach(host->boot_processes,cpt,arg) {
57       smx_process_t process;
58
59       char** argv = xbt_new(char*, arg->argc);
60       for (int i=0; i<arg->argc; i++)
61         argv[i] = xbt_strdup(arg->argv[i]);
62
63       XBT_DEBUG("Booting Process %s(%s) right now", arg->argv[0], arg->hostname);
64       if (simix_global->create_process_function) {
65         simix_global->create_process_function(&process,
66                                               argv[0],
67                                               arg->code,
68                                               NULL,
69                                               arg->hostname,
70                                               arg->kill_time,
71                                               arg->argc,
72                                               argv,
73                                               arg->properties,
74                                               arg->auto_restart,
75                                               NULL);
76       } else {
77         simcall_process_create(&process,
78                                arg->argv[0],
79                                arg->code,
80                                NULL,
81                                arg->hostname,
82                                arg->kill_time,
83                                arg->argc,
84                                argv,
85                                arg->properties,
86                                arg->auto_restart);
87       }
88     }
89   }
90 }
91
92 void simcall_HANDLER_host_off(smx_simcall_t simcall, smx_host_t h)
93 {
94  SIMIX_host_off(h, simcall->issuer);
95 }
96
97 /**
98  * \brief Stop the host if it is on
99  *
100  */
101 void SIMIX_host_off(smx_host_t h, smx_process_t issuer)
102 {
103   smx_host_priv_t host = SIMIX_host_priv(h);
104
105   xbt_assert((host != NULL), "Invalid parameters");
106
107   if (surf_resource_get_state(surf_workstation_resource_priv(h))==SURF_RESOURCE_ON) {
108         surf_resource_set_state(surf_workstation_resource_priv(h), SURF_RESOURCE_OFF);
109
110     /* Clean Simulator data */
111     if (xbt_swag_size(host->process_list) != 0) {
112       smx_process_t process = NULL;
113       xbt_swag_foreach(process, host->process_list) {
114         SIMIX_process_kill(process, issuer);
115         XBT_DEBUG("Killing %s on %s by %s", process->name,  sg_host_name(process->smx_host), issuer->name);
116       }
117     }
118   }
119 }
120
121 /**
122  * \brief Internal function to destroy a SIMIX host.
123  *
124  * \param h the host to destroy (a smx_host_t)
125  */
126 void SIMIX_host_destroy(void *h)
127 {
128   smx_host_priv_t host = (smx_host_priv_t) h;
129
130   xbt_assert((host != NULL), "Invalid parameters");
131
132   /* Clean Simulator data */
133   if (xbt_swag_size(host->process_list) != 0) {
134     char *msg = xbt_strdup("Shutting down host, but it's not empty:");
135     char *tmp;
136     smx_process_t process = NULL;
137
138     xbt_swag_foreach(process, host->process_list) {
139       tmp = bprintf("%s\n\t%s", msg, process->name);
140       free(msg);
141       msg = tmp;
142     }
143     SIMIX_display_process_status();
144     THROWF(arg_error, 0, "%s", msg);
145   }
146   xbt_dynar_free(&host->auto_restart_processes);
147   xbt_dynar_free(&host->boot_processes);
148   xbt_swag_free(host->process_list);
149
150   /* Clean host structure */
151   free(host);
152   return;
153 }
154
155 smx_host_t SIMIX_host_get_by_name(const char *name){
156   xbt_assert(((simix_global != NULL)
157                && (host_lib != NULL)),
158               "Environment not set yet");
159
160   return xbt_lib_get_elm_or_null(host_lib, name);
161 }
162
163 smx_host_t SIMIX_host_self(void)
164 {
165   smx_process_t process = SIMIX_process_self();
166   return (process == NULL) ? NULL : SIMIX_process_get_host(process);
167 }
168
169 /* needs to be public and without simcall because it is called
170    by exceptions and logging events */
171 const char* SIMIX_host_self_get_name(void)
172 {
173   smx_host_t host = SIMIX_host_self();
174   if (host == NULL || SIMIX_process_self() == simix_global->maestro_process)
175     return "";
176
177   return SIMIX_host_get_name(host);
178 }
179
180 const char* SIMIX_host_get_name(smx_host_t host){
181   xbt_assert((host != NULL), "Invalid parameters");
182
183   return sg_host_name(host);
184 }
185
186 xbt_dict_t SIMIX_host_get_properties(smx_host_t host){
187   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
188
189   return surf_resource_get_properties(surf_workstation_resource_priv(host));
190 }
191
192 double SIMIX_host_get_speed(smx_host_t host){
193   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
194   return surf_workstation_get_speed(host, 1.0);
195 }
196
197 int SIMIX_host_get_core(smx_host_t host){
198   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
199
200   return surf_workstation_get_core(host);
201 }
202
203 xbt_swag_t SIMIX_host_get_process_list(smx_host_t host){
204   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
205   smx_host_priv_t host_priv = SIMIX_host_priv(host);
206
207   return host_priv->process_list;
208 }
209
210
211 double SIMIX_host_get_available_speed(smx_host_t host){
212   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
213
214   return surf_workstation_get_available_speed(host);
215 }
216
217 double SIMIX_host_get_current_power_peak(smx_host_t host) {
218           xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
219           return surf_workstation_get_current_power_peak(host);
220 }
221
222 double SIMIX_host_get_power_peak_at(smx_host_t host, int pstate_index) {
223           xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
224
225           return surf_workstation_get_power_peak_at(host, pstate_index);
226 }
227
228 int SIMIX_host_get_nb_pstates(smx_host_t host) {
229           xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
230
231           return surf_workstation_get_nb_pstates(host);
232 }
233
234
235 void SIMIX_host_set_pstate(smx_host_t host, int pstate_index) {
236           xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
237
238           surf_workstation_set_pstate(host, pstate_index);
239 }
240
241 double SIMIX_host_get_consumed_energy(smx_host_t host) {
242           xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
243           return surf_workstation_get_consumed_energy(host);
244 }
245
246 int SIMIX_host_get_state(smx_host_t host){
247   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
248
249   return surf_resource_get_state(surf_workstation_resource_priv(host));
250 }
251
252 void _SIMIX_host_free_process_arg(void *data)
253 {
254   smx_process_arg_t arg = *(void**)data;
255   int i;
256   for (i = 0; i < arg->argc; i++)
257     xbt_free(arg->argv[i]);
258   xbt_free(arg->argv);
259   xbt_free(arg->name);
260   xbt_free(arg);
261 }
262 /**
263  * \brief Add a process to the list of the processes that the host will restart when it comes back
264  * This function add a process to the list of the processes that will be restarted when the host comes
265  * back. It is expected that this function is called when the host is down.
266  * The processes will only be restarted once, meaning that you will have to register the process
267  * again to restart the process again.
268  */
269 void SIMIX_host_add_auto_restart_process(smx_host_t host,
270                                          const char *name,
271                                          xbt_main_func_t code,
272                                          void *data,
273                                          const char *hostname,
274                                          double kill_time,
275                                          int argc, char **argv,
276                                          xbt_dict_t properties,
277                                          int auto_restart)
278 {
279   if (!SIMIX_host_priv(host)->auto_restart_processes) {
280     SIMIX_host_priv(host)->auto_restart_processes = xbt_dynar_new(sizeof(smx_process_arg_t),_SIMIX_host_free_process_arg);
281   }
282   smx_process_arg_t arg = xbt_new(s_smx_process_arg_t,1);
283   arg->name = xbt_strdup(name);
284   arg->code = code;
285   arg->data = data;
286   arg->hostname = hostname;
287   arg->kill_time = kill_time;
288   arg->argc = argc;
289
290   arg->argv = xbt_new(char*,argc + 1);
291
292   int i;
293   for (i = 0; i < argc; i++) {
294     arg->argv[i] = xbt_strdup(argv[i]);
295   }
296   arg->argv[argc] = NULL;
297
298   arg->properties = properties;
299   arg->auto_restart = auto_restart;
300
301   if( SIMIX_host_get_state(host) == SURF_RESOURCE_OFF
302       && !xbt_dict_get_or_null(watched_hosts_lib,sg_host_name(host))){
303     xbt_dict_set(watched_hosts_lib,sg_host_name(host),host,NULL);
304     XBT_DEBUG("Have pushed host %s to watched_hosts_lib because state == SURF_RESOURCE_OFF",sg_host_name(host));
305   }
306   xbt_dynar_push_as(SIMIX_host_priv(host)->auto_restart_processes,smx_process_arg_t,arg);
307 }
308 /**
309  * \brief Restart the list of processes that have been registered to the host
310  */
311 void SIMIX_host_restart_processes(smx_host_t host)
312 {
313   unsigned int cpt;
314   smx_process_arg_t arg;
315   xbt_dynar_t process_list = SIMIX_host_priv(host)->auto_restart_processes;
316   if (!process_list)
317     return;
318
319   xbt_dynar_foreach (process_list, cpt, arg) {
320
321     smx_process_t process;
322
323     XBT_DEBUG("Restarting Process %s(%s) right now", arg->argv[0], arg->hostname);
324     if (simix_global->create_process_function) {
325       simix_global->create_process_function(&process,
326                                             arg->argv[0],
327                                             arg->code,
328                                             NULL,
329                                             arg->hostname,
330                                             arg->kill_time,
331                                             arg->argc,
332                                             arg->argv,
333                                             arg->properties,
334                                             arg->auto_restart,
335                                             NULL);
336     } else {
337       simcall_process_create(&process,
338                              arg->argv[0],
339                              arg->code,
340                              NULL,
341                              arg->hostname,
342                              arg->kill_time,
343                              arg->argc,
344                              arg->argv,
345                              arg->properties,
346                              arg->auto_restart);
347
348     }
349     /* arg->argv is used by the process created above.  Hide it to
350      * _SIMIX_host_free_process_arg() which is called by xbt_dynar_reset()
351      * below. */
352     arg->argc = 0;
353     arg->argv = NULL;
354   }
355   xbt_dynar_reset(process_list);
356 }
357
358 void SIMIX_host_autorestart(smx_host_t host)
359 {
360   if(simix_global->autorestart)
361     simix_global->autorestart(host);
362   else
363     xbt_die("No function for simix_global->autorestart");
364 }
365
366 smx_synchro_t SIMIX_host_execute(const char *name,
367     smx_host_t host, double flops_amount, double priority, double bound, unsigned long affinity_mask){
368
369   /* alloc structures and initialize */
370   smx_synchro_t synchro = xbt_mallocator_get(simix_global->synchro_mallocator);
371   synchro->type = SIMIX_SYNC_EXECUTE;
372   synchro->name = xbt_strdup(name);
373   synchro->state = SIMIX_RUNNING;
374   synchro->execution.host = host;
375   synchro->category = NULL;
376
377   /* set surf's action */
378   if (!MC_is_active() && !MC_record_replay_is_active()) {
379
380     synchro->execution.surf_exec = surf_workstation_execute(host, flops_amount);
381     surf_action_set_data(synchro->execution.surf_exec, synchro);
382     surf_action_set_priority(synchro->execution.surf_exec, priority);
383
384     /* Note (hypervisor): for multicore, the bound value being passed to the
385      * surf layer should not be zero (i.e., unlimited). It should be the
386      * capacity of a CPU core. */
387     if (bound == 0)
388       surf_cpu_action_set_bound(synchro->execution.surf_exec, SIMIX_host_get_speed(host));
389     else
390       surf_cpu_action_set_bound(synchro->execution.surf_exec, bound);
391
392     if (affinity_mask != 0) {
393       /* just a double check to confirm that this host is the host where this task is running. */
394       xbt_assert(synchro->execution.host == host);
395       surf_cpu_action_set_affinity(synchro->execution.surf_exec, host, affinity_mask);
396     }
397   }
398
399   XBT_DEBUG("Create execute synchro %p: %s", synchro, synchro->name);
400
401   return synchro;
402 }
403
404 smx_synchro_t SIMIX_host_parallel_execute(const char *name,
405     int host_nb, smx_host_t *host_list,
406     double *flops_amount, double *bytes_amount,
407     double amount, double rate){
408
409   void **workstation_list = NULL;
410   int i;
411
412   /* alloc structures and initialize */
413   smx_synchro_t synchro = xbt_mallocator_get(simix_global->synchro_mallocator);
414   synchro->type = SIMIX_SYNC_PARALLEL_EXECUTE;
415   synchro->name = xbt_strdup(name);
416   synchro->state = SIMIX_RUNNING;
417   synchro->execution.host = NULL; /* FIXME: do we need the list of hosts? */
418   synchro->category = NULL;
419
420   /* set surf's synchro */
421   workstation_list = xbt_new0(void *, host_nb);
422   for (i = 0; i < host_nb; i++)
423     workstation_list[i] = surf_workstation_resource_priv(host_list[i]);
424
425
426   /* FIXME: what happens if host_list contains VMs and PMs. If
427    * execute_parallel_task() does not change the state of the model, we can mix
428    * them. */
429   surf_model_t ws_model = surf_resource_model(host_list[0], SURF_WKS_LEVEL);
430   for (i = 1; i < host_nb; i++) {
431     surf_model_t ws_model_tmp = surf_resource_model(host_list[i], SURF_WKS_LEVEL);
432     if (ws_model_tmp != ws_model) {
433       XBT_CRITICAL("mixing VMs and PMs is not supported");
434       DIE_IMPOSSIBLE;
435     }
436   }
437
438   /* set surf's synchro */
439   if (!MC_is_active() && !MC_record_replay_is_active()) {
440     synchro->execution.surf_exec =
441       surf_workstation_model_execute_parallel_task((surf_workstation_model_t)surf_workstation_model,
442                   host_nb, workstation_list, flops_amount, bytes_amount, rate);
443
444     surf_action_set_data(synchro->execution.surf_exec, synchro);
445   }
446   XBT_DEBUG("Create parallel execute synchro %p", synchro);
447
448   return synchro;
449 }
450
451 void SIMIX_host_execution_destroy(smx_synchro_t synchro){
452   XBT_DEBUG("Destroy synchro %p", synchro);
453
454   if (synchro->execution.surf_exec) {
455     surf_action_unref(synchro->execution.surf_exec);
456     synchro->execution.surf_exec = NULL;
457   }
458   xbt_free(synchro->name);
459   xbt_mallocator_release(simix_global->synchro_mallocator, synchro);
460 }
461
462 void SIMIX_host_execution_cancel(smx_synchro_t synchro){
463   XBT_DEBUG("Cancel synchro %p", synchro);
464
465   if (synchro->execution.surf_exec)
466     surf_action_cancel(synchro->execution.surf_exec);
467 }
468
469 double SIMIX_host_execution_get_remains(smx_synchro_t synchro){
470   double result = 0.0;
471
472   if (synchro->state == SIMIX_RUNNING)
473     result = surf_action_get_remains(synchro->execution.surf_exec);
474
475   return result;
476 }
477
478 e_smx_state_t SIMIX_host_execution_get_state(smx_synchro_t synchro){
479   return synchro->state;
480 }
481
482 void SIMIX_host_execution_set_priority(smx_synchro_t synchro, double priority){
483
484   if(synchro->execution.surf_exec)
485         surf_action_set_priority(synchro->execution.surf_exec, priority);
486 }
487
488 void SIMIX_host_execution_set_bound(smx_synchro_t synchro, double bound){
489
490   if(synchro->execution.surf_exec)
491         surf_cpu_action_set_bound(synchro->execution.surf_exec, bound);
492 }
493
494 void SIMIX_host_execution_set_affinity(smx_synchro_t synchro, smx_host_t host, unsigned long mask){
495   xbt_assert(synchro->type == SIMIX_SYNC_EXECUTE);
496
497   if (synchro->execution.surf_exec) {
498     /* just a double check to confirm that this host is the host where this task is running. */
499     xbt_assert(synchro->execution.host == host);
500     surf_cpu_action_set_affinity(synchro->execution.surf_exec, host, mask);
501   }
502 }
503
504 void simcall_HANDLER_host_execution_wait(smx_simcall_t simcall, smx_synchro_t synchro){
505
506   XBT_DEBUG("Wait for execution of synchro %p, state %d", synchro, (int)synchro->state);
507
508   /* Associate this simcall to the synchro */
509   xbt_fifo_push(synchro->simcalls, simcall);
510   simcall->issuer->waiting_synchro = synchro;
511
512   /* set surf's synchro */
513   if (MC_is_active() || MC_record_replay_is_active()) {
514     synchro->state = SIMIX_DONE;
515     SIMIX_execution_finish(synchro);
516     return;
517   }
518
519   /* If the synchro is already finished then perform the error handling */
520   if (synchro->state != SIMIX_RUNNING)
521     SIMIX_execution_finish(synchro);
522 }
523
524 void SIMIX_host_execution_suspend(smx_synchro_t synchro)
525 {
526   if(synchro->execution.surf_exec)
527     surf_action_suspend(synchro->execution.surf_exec);
528 }
529
530 void SIMIX_host_execution_resume(smx_synchro_t synchro)
531 {
532   if(synchro->execution.surf_exec)
533     surf_action_resume(synchro->execution.surf_exec);
534 }
535
536 void SIMIX_execution_finish(smx_synchro_t synchro)
537 {
538   xbt_fifo_item_t item;
539   smx_simcall_t simcall;
540
541   xbt_fifo_foreach(synchro->simcalls, item, simcall, smx_simcall_t) {
542
543     switch (synchro->state) {
544
545       case SIMIX_DONE:
546         /* do nothing, synchro done */
547         XBT_DEBUG("SIMIX_execution_finished: execution successful");
548         break;
549
550       case SIMIX_FAILED:
551         XBT_DEBUG("SIMIX_execution_finished: host '%s' failed", sg_host_name(simcall->issuer->smx_host));
552         simcall->issuer->context->iwannadie = 1;
553         SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
554         break;
555
556       case SIMIX_CANCELED:
557         XBT_DEBUG("SIMIX_execution_finished: execution canceled");
558         SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Canceled");
559         break;
560
561       default:
562         xbt_die("Internal error in SIMIX_execution_finish: unexpected synchro state %d",
563             (int)synchro->state);
564     }
565     /* check if the host is down */
566     if (surf_resource_get_state(surf_workstation_resource_priv(simcall->issuer->smx_host)) != SURF_RESOURCE_ON) {
567       simcall->issuer->context->iwannadie = 1;
568     }
569
570     simcall->issuer->waiting_synchro =    NULL;
571     simcall_host_execution_wait__set__result(simcall, synchro->state);
572     SIMIX_simcall_answer(simcall);
573   }
574
575   /* We no longer need it */
576   SIMIX_host_execution_destroy(synchro);
577 }
578
579
580 void SIMIX_post_host_execute(smx_synchro_t synchro)
581 {
582   if (synchro->type == SIMIX_SYNC_EXECUTE && /* FIMXE: handle resource failure
583                                                * for parallel tasks too */
584       surf_resource_get_state(surf_workstation_resource_priv(synchro->execution.host)) == SURF_RESOURCE_OFF) {
585     /* If the host running the synchro failed, notice it so that the asking
586      * process can be killed if it runs on that host itself */
587     synchro->state = SIMIX_FAILED;
588   } else if (surf_action_get_state(synchro->execution.surf_exec) == SURF_ACTION_FAILED) {
589     /* If the host running the synchro didn't fail, then the synchro was
590      * canceled */
591     synchro->state = SIMIX_CANCELED;
592   } else {
593     synchro->state = SIMIX_DONE;
594   }
595
596   if (synchro->execution.surf_exec) {
597     surf_action_unref(synchro->execution.surf_exec);
598     synchro->execution.surf_exec = NULL;
599   }
600
601   /* If there are simcalls associated with the synchro, then answer them */
602   if (xbt_fifo_size(synchro->simcalls)) {
603     SIMIX_execution_finish(synchro);
604   }
605 }
606
607
608 void SIMIX_set_category(smx_synchro_t synchro, const char *category)
609 {
610   if (synchro->state != SIMIX_RUNNING) return;
611   if (synchro->type == SIMIX_SYNC_EXECUTE){
612     surf_action_set_category(synchro->execution.surf_exec, category);
613   }else if (synchro->type == SIMIX_SYNC_COMMUNICATE){
614     surf_action_set_category(synchro->comm.surf_comm, category);
615   }
616 }
617
618 /**
619  * \brief Function to get the parameters of the given the SIMIX host.
620  *
621  * \param host the host to get_phys_host (a smx_host_t)
622  * \param param the parameter object space to be overwritten (a ws_params_t)
623  */
624 void SIMIX_host_get_params(smx_host_t ind_vm, ws_params_t params)
625 {
626   /* jump to ws_get_params(). */
627   surf_workstation_get_params(ind_vm, params);
628 }
629
630 void SIMIX_host_set_params(smx_host_t ind_vm, ws_params_t params)
631 {
632   /* jump to ws_set_params(). */
633   surf_workstation_set_params(ind_vm, params);
634 }
635
636 xbt_dict_t SIMIX_host_get_mounted_storage_list(smx_host_t host){
637   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
638
639   return surf_workstation_get_mounted_storage_list(host);
640 }
641
642 xbt_dynar_t SIMIX_host_get_attached_storage_list(smx_host_t host){
643   xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
644
645   return surf_workstation_get_attached_storage_list(host);
646 }