X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/272ccad1b68b6d9c17069f3c934886925bb15b5d..9d7e857328377861f81e95fbc3138032b51b7da9:/src/simix/smx_host.c

diff --git a/src/simix/smx_host.c b/src/simix/smx_host.c
index 299864a5f9..f663ea260a 100644
--- a/src/simix/smx_host.c
+++ b/src/simix/smx_host.c
@@ -67,7 +67,7 @@ void SIMIX_host_destroy(void *h)
     SIMIX_display_process_status();
     THROWF(arg_error, 0, "%s", msg);
   }
-
+  xbt_dynar_free(&host->auto_restart_processes);
   xbt_swag_free(host->process_list);
 
   /* Clean host structure */
@@ -176,6 +176,50 @@ void* SIMIX_host_get_data(smx_host_t host)
 
   return host->data;
 }
+void _SIMIX_host_free_process_arg(void *);
+void _SIMIX_host_free_process_arg(void *data) {
+  smx_process_arg_t arg = *(void**)data;
+  int i;
+  xbt_free(arg->name);
+  for (i = 0; i < arg->argc; i++) {
+    xbt_free(arg->argv[i]);
+  }
+  xbt_free(arg->argv);
+  xbt_free(arg);
+}
+void SIMIX_host_add_auto_restart_process(smx_host_t host,
+                                         const char *name,
+                                         xbt_main_func_t code,
+                                         void *data,
+                                         const char *hostname,
+                                         double kill_time,
+                                         int argc, char **argv,
+                                         xbt_dict_t properties,
+                                         int auto_restart) {
+  if (!host->auto_restart_processes) {
+    host->auto_restart_processes = xbt_dynar_new(sizeof(smx_process_arg_t),_SIMIX_host_free_process_arg);
+  }
+  smx_process_arg_t arg = xbt_new(s_smx_process_arg_t,1);
+
+  arg->name = xbt_strdup(name);
+  arg->code = code;
+  arg->data = data;
+  arg->hostname = hostname;
+  arg->kill_time = kill_time;
+  arg->argc = argc;
+  arg->argv = xbt_new(char*,argc + 1);
+
+  int i;
+  for (i = 0; i < argc; i++) {
+    arg->argv[i] = xbt_strdup(argv[i]);
+  }
+
+  arg->properties = properties;
+  arg->auto_restart = auto_restart;
+
+  xbt_dynar_push_as(host->auto_restart_processes,smx_process_arg_t,arg);
+}
+
 
 void SIMIX_host_set_data(smx_host_t host, void *data)
 {
@@ -348,13 +392,8 @@ void SIMIX_execution_finish(smx_action_t action)
 
       case SIMIX_FAILED:
         XBT_DEBUG("SIMIX_execution_finished: host '%s' failed", simcall->issuer->smx_host->name);
-        if (simcall->issuer->smx_host == action->execution.host) {
-          // add a reference to the action that will be destroyed when the killed process is cleaned up, and by the end of the current function
-          surf_action_ref(action->execution.surf_exec);
-          SIMIX_process_kill(simcall->issuer);
-        } else {
-          SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
-        }
+        simcall->issuer->context->iwannadie = 1;
+        //SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
         break;
 
       case SIMIX_CANCELED:
@@ -366,7 +405,13 @@ void SIMIX_execution_finish(smx_action_t action)
         xbt_die("Internal error in SIMIX_execution_finish: unexpected action state %d",
             (int)action->state);
     }
-    simcall->issuer->waiting_action = NULL;
+    /* check if the host is down */
+    if (surf_workstation_model->extension.
+        workstation.get_state(simcall->issuer->smx_host->host) != SURF_RESOURCE_ON) {
+      simcall->issuer->context->iwannadie = 1;
+    }
+
+    simcall->issuer->waiting_action =    NULL;
     simcall->host_execution_wait.result = action->state;
     SIMIX_simcall_answer(simcall);
   }
@@ -377,14 +422,18 @@ void SIMIX_execution_finish(smx_action_t action)
 
 void SIMIX_post_host_execute(smx_action_t action)
 {
-  if (surf_workstation_model->extension.workstation.get_state(action->execution.host->host)==SURF_RESOURCE_OFF) {
-    /* if the host running the action failed, notice it so that the asking process can be killed if it runs on that host itself */
+  if (action->type == SIMIX_ACTION_EXECUTE && /* FIMXE: handle resource failure
+                                               * for parallel tasks too */
+      surf_workstation_model->extension.workstation.get_state(action->execution.host->host) == SURF_RESOURCE_OFF) {
+    /* If the host running the action failed, notice it so that the asking
+     * process can be killed if it runs on that host itself */
     action->state = SIMIX_FAILED;
   } else if (surf_workstation_model->action_state_get(action->execution.surf_exec) == SURF_ACTION_FAILED) {
-    /* If the host running the action didn't fail, then the action was canceled */
-     action->state = SIMIX_CANCELED;
+    /* If the host running the action didn't fail, then the action was
+     * canceled */
+    action->state = SIMIX_CANCELED;
   } else {
-     action->state = SIMIX_DONE;
+    action->state = SIMIX_DONE;
   }
 
   if (action->execution.surf_exec) {