Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
Compilation with warning flags
[simgrid.git] / src / simix / smx_process.c
index 86422fd..a25b215 100644 (file)
@@ -1,10 +1,10 @@
-/* Copyright (c) 2007, 2008, 2009, 2010. The SimGrid Team.
+/* Copyright (c) 2007-2012. The SimGrid Team.
  * All rights reserved.                                                     */
 
 /* This program is free software; you can redistribute it and/or modify it
  * under the terms of the license (GNU LGPL) which comes with this package. */
 
-#include "private.h"
+#include "smx_private.h"
 #include "xbt/sysdep.h"
 #include "xbt/log.h"
 #include "xbt/dict.h"
@@ -56,7 +56,7 @@ void SIMIX_process_cleanup(smx_process_t process)
 
     if (action->comm.src_proc == process) {
       XBT_DEBUG("Found an unfinished send comm %p (detached = %d), state %d, src = %p, dst = %p",
-          action, action->comm.detached, action->state, action->comm.src_proc, action->comm.dst_proc);
+          action, action->comm.detached, (int)action->state, action->comm.src_proc, action->comm.dst_proc);
       action->comm.src_proc = NULL;
 
       if (action->comm.detached) {
@@ -76,7 +76,7 @@ void SIMIX_process_cleanup(smx_process_t process)
     }
     else if (action->comm.dst_proc == process){
       XBT_DEBUG("Found an unfinished recv comm %p, state %d, src = %p, dst = %p",
-          action, action->state, action->comm.src_proc, action->comm.dst_proc);
+          action, (int)action->state, action->comm.src_proc, action->comm.dst_proc);
       action->comm.dst_proc = NULL;
 
       if (action->comm.detached && action->comm.refcount == 1
@@ -96,6 +96,7 @@ void SIMIX_process_cleanup(smx_process_t process)
   xbt_swag_remove(process, simix_global->process_list);
   xbt_swag_remove(process, process->smx_host->process_list);
   xbt_swag_insert(process, simix_global->process_to_destroy);
+  process->context->iwannadie = 0;
 }
 
 /** 
@@ -117,6 +118,8 @@ void SIMIX_process_empty_trash(void)
 
     xbt_fifo_free(process->comms);
 
+    xbt_dynar_free(&process->on_exit);
+
     free(process->name);
     free(process);
   }
@@ -136,14 +139,39 @@ void SIMIX_create_maestro_process()
   maestro->running_ctx = xbt_new(xbt_running_ctx_t, 1);
   XBT_RUNNING_CTX_INITIALIZE(maestro->running_ctx);
   maestro->context = SIMIX_context_new(NULL, 0, NULL, NULL, maestro);
-  maestro->request.issuer = maestro;
+  maestro->simcall.issuer = maestro;
 
   simix_global->maestro_process = maestro;
   return;
 }
+/**
+ * \brief Stops a process.
+ *
+ * Stops the process, execute all the registered on_exit functions,
+ * register it to the list of the process to restart if needed
+ * and stops its context.
+ */
+void SIMIX_process_stop(smx_process_t arg) {
+  /* execute the on_exit functions */
+  SIMIX_process_on_exit_runall(arg);
+  /* Add the process to the list of process to restart, only if
+   * the host is down
+   */
+  if (arg->auto_restart && !SIMIX_host_get_state(arg->smx_host)) {
+    SIMIX_host_add_auto_restart_process(arg->smx_host,arg->name,arg->code, arg->data,
+                                        arg->smx_host->name,
+                                        arg->kill_time,
+                                        arg->argc,arg->argv,arg->properties,
+                                        arg->auto_restart);
+  }
+  XBT_DEBUG("Process %s (%s) is dead",arg->name,arg->smx_host->name);
+  /* stop the context */
+  SIMIX_context_stop(arg->context);
+}
 
 /**
  * \brief Same as SIMIX_process_create() but with only one argument (used by timers).
+ * This function frees the argument.
  * \return the process created
  */
 smx_process_t SIMIX_process_create_from_wrapper(smx_process_arg_t args) {
@@ -155,10 +183,12 @@ smx_process_t SIMIX_process_create_from_wrapper(smx_process_arg_t args) {
       args->code,
       args->data,
       args->hostname,
+      args->kill_time,
       args->argc,
       args->argv,
-      args->properties);
-
+      args->properties,
+      args->auto_restart);
+  xbt_free(args);
   return process;
 }
 
@@ -166,8 +196,8 @@ smx_process_t SIMIX_process_create_from_wrapper(smx_process_arg_t args) {
  * \brief Internal function to create a process.
  *
  * This function actually creates the process.
- * It may be called when a REQ_PROCESS_CREATE request occurs,
- * or directly for SIMIX internal purposes.
+ * It may be called when a SIMCALL_PROCESS_CREATE simcall occurs,
+ * or directly for SIMIX internal purposes. The sure thing is that it's called from maestro context.
  *
  * \return the process created
  */
@@ -176,8 +206,10 @@ void SIMIX_process_create(smx_process_t *process,
                           xbt_main_func_t code,
                           void *data,
                           const char *hostname,
+                          double kill_time,
                           int argc, char **argv,
-                          xbt_dict_t properties) {
+                          xbt_dict_t properties,
+                          int auto_restart) {
 
   *process = NULL;
   smx_host_t host = SIMIX_host_get_by_name(hostname);
@@ -192,18 +224,24 @@ void SIMIX_process_create(smx_process_t *process,
     *process = xbt_new0(s_smx_process_t, 1);
 
     xbt_assert(((code != NULL) && (host != NULL)), "Invalid parameters");
-
     /* Process data */
     (*process)->pid = simix_process_maxpid++;
     (*process)->name = xbt_strdup(name);
     (*process)->smx_host = host;
     (*process)->data = data;
     (*process)->comms = xbt_fifo_new();
-    (*process)->request.issuer = *process;
+    (*process)->simcall.issuer = *process;
+    /* Process data for auto-restart */
+    (*process)->auto_restart = auto_restart;
+    (*process)->code = code;
+    (*process)->argc = argc;
+    (*process)->argv = argv;
+    (*process)->kill_time = kill_time;
+
 
     XBT_VERB("Create context %s", (*process)->name);
     (*process)->context = SIMIX_context_new(code, argc, argv,
-       simix_global->cleanup_process_function, *process);
+      simix_global->cleanup_process_function, *process);
 
     (*process)->running_ctx = xbt_new(xbt_running_ctx_t, 1);
     XBT_RUNNING_CTX_INITIALIZE((*process)->running_ctx);
@@ -221,6 +259,14 @@ void SIMIX_process_create(smx_process_t *process,
     XBT_DEBUG("Inserting %s(%s) in the to_run list", (*process)->name, host->name);
     xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, *process);
   }
+
+  if (kill_time > SIMIX_get_clock()) {
+    if (simix_global->kill_process_function) {
+      XBT_DEBUG("Process %s(%s) will be kill at time %f", (*process)->name,
+          (*process)->smx_host->name, kill_time);
+      SIMIX_timer_set(kill_time, simix_global->kill_process_function, *process);
+    }
+  }
 }
 
 /**
@@ -245,12 +291,13 @@ void SIMIX_process_runall(void)
 /**
  * \brief Internal function to kill a SIMIX process.
  *
- * This function may be called when a REQ_PROCESS_KILL request occurs,
+ * This function may be called when a SIMCALL_PROCESS_KILL simcall occurs,
  * or directly for SIMIX internal purposes.
  *
  * \param process poor victim
+ * \param issuer the process which has sent the PROCESS_KILL. Important to not schedule twice the same process.
  */
-void SIMIX_process_kill(smx_process_t process) {
+void SIMIX_process_kill(smx_process_t process, smx_process_t issuer) {
 
   XBT_DEBUG("Killing process %s on %s", process->name, process->smx_host->name);
 
@@ -274,22 +321,32 @@ void SIMIX_process_kill(smx_process_t process) {
         SIMIX_comm_destroy(process->waiting_action);
         break;
 
-      case SIMIX_ACTION_SLEEP:
-       SIMIX_process_sleep_destroy(process->waiting_action);
-       break;
+        case SIMIX_ACTION_SLEEP:
+          SIMIX_process_sleep_destroy(process->waiting_action);
+          break;
 
-      case SIMIX_ACTION_SYNCHRO:
-       SIMIX_synchro_stop_waiting(process, &process->request);
-       SIMIX_synchro_destroy(process->waiting_action);
-       break;
+        case SIMIX_ACTION_SYNCHRO:
+          SIMIX_synchro_stop_waiting(process, &process->simcall);
+          SIMIX_synchro_destroy(process->waiting_action);
+          break;
+
+        case SIMIX_ACTION_IO:
+          SIMIX_io_destroy(process->waiting_action);
+          break;
+
+        /* **************************************/
+        /* TUTORIAL: New API                    */
+        case SIMIX_ACTION_NEW_API:
+          SIMIX_new_api_destroy(process->waiting_action);
+          break;
+        /* **************************************/
 
-      case SIMIX_ACTION_IO:
-        SIMIX_io_destroy(process->waiting_action);
-        break;
     }
   }
+  if(!xbt_dynar_member(simix_global->process_to_run, &(process)) && process != issuer) {
+    xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
+  }
 
-  xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 }
 
 /**
@@ -302,17 +359,17 @@ void SIMIX_process_killall(smx_process_t issuer)
 
   while ((p = xbt_swag_extract(simix_global->process_list))) {
     if (p != issuer) {
-      SIMIX_process_kill(p);
+      SIMIX_process_kill(p,issuer);
     }
   }
 
-  SIMIX_context_runall(simix_global->process_to_run);
+  SIMIX_context_runall();
 
   SIMIX_process_empty_trash();
 }
 
 void SIMIX_process_change_host(smx_process_t process,
-                              smx_host_t dest)
+             smx_host_t dest)
 {
   xbt_assert((process != NULL), "Invalid parameters");
   xbt_swag_remove(process, process->smx_host->process_list);
@@ -325,24 +382,29 @@ void SIMIX_pre_process_change_host(smx_process_t process, smx_host_t dest)
   process->new_host = dest;
 }
 
-void SIMIX_pre_process_suspend(smx_req_t req)
+void SIMIX_pre_process_suspend(smx_simcall_t simcall)
 {
-  smx_process_t process = req->process_suspend.process;
-  SIMIX_process_suspend(process, req->issuer);
+  smx_process_t process = simcall->process_suspend.process;
+  smx_action_t action_suspend =
+      SIMIX_process_suspend(process, simcall->issuer);
 
-  if (process != req->issuer) {
-    SIMIX_request_answer(req);
+  if (process != simcall->issuer) {
+    SIMIX_simcall_answer(simcall);
+  } else {
+    xbt_fifo_push(action_suspend->simcalls, simcall);
+    process->waiting_action = action_suspend;
+    SIMIX_host_execution_suspend(process->waiting_action);
   }
-  /* If we are suspending ourselves, then just do not replay the request. */
+  /* If we are suspending ourselves, then just do not finish the simcall now */
 }
 
-void SIMIX_process_suspend(smx_process_t process, smx_process_t issuer)
+smx_action_t SIMIX_process_suspend(smx_process_t process, smx_process_t issuer)
 {
   xbt_assert((process != NULL), "Invalid parameters");
 
   if (process->suspended) {
     XBT_DEBUG("Process '%s' is already suspended", process->name);
-    return;
+    return NULL;
   }
 
   process->suspended = 1;
@@ -368,11 +430,21 @@ void SIMIX_process_suspend(smx_process_t process, smx_process_t issuer)
           SIMIX_process_sleep_suspend(process->waiting_action);
           break;
 
+        case SIMIX_ACTION_SYNCHRO:
+          /* Suspension is delayed to when the process is rescheduled. */
+          break;
+
         default:
           xbt_die("Internal error in SIMIX_process_suspend: unexpected action type %d",
-              process->waiting_action->type);
+              (int)process->waiting_action->type);
       }
+      return NULL;
+    } else {
+      /* Suspension is delayed to when the process is rescheduled. */
+      return NULL;
     }
+  } else {
+    return SIMIX_host_execute("suspend", process->smx_host, 0.0, 1.0);
   }
 }
 
@@ -380,11 +452,14 @@ void SIMIX_process_resume(smx_process_t process, smx_process_t issuer)
 {
   xbt_assert((process != NULL), "Invalid parameters");
 
-  if (!process->suspended) {
-    XBT_DEBUG("Process '%s' is not suspended", process->name);
+  XBT_IN("process = %p, issuer = %p", process, issuer);
+
+  if(process->context->iwannadie) {
+    XBT_VERB("Ignoring request to suspend a process that is currently dying.");
     return;
   }
 
+  if(!process->suspended) return;
   process->suspended = 0;
 
   /* If we are resuming another process, resume the action it was waiting for
@@ -408,15 +483,19 @@ void SIMIX_process_resume(smx_process_t process, smx_process_t issuer)
           SIMIX_process_sleep_resume(process->waiting_action);
           break;
 
+        case SIMIX_ACTION_SYNCHRO:
+          /* I cannot resume it now. This is delayed to when the process is rescheduled at
+           * the end of the synchro. */
+          break;
+
         default:
           xbt_die("Internal error in SIMIX_process_resume: unexpected action type %d",
-              process->waiting_action->type);
+              (int)process->waiting_action->type);
       }
     }
-    else {
-      xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
-    }
-  }
+  } else XBT_WARN("Strange. Process %p is trying to resume himself.", issuer);
+
+  XBT_OUT();
 }
 
 int SIMIX_process_get_maxpid(void) {
@@ -460,7 +539,7 @@ smx_host_t SIMIX_process_get_host(smx_process_t process)
   return process->smx_host;
 }
 
-/* needs to be public and without request because it is called
+/* needs to be public and without simcall because it is called
    by exceptions and logging events */
 const char* SIMIX_process_self_get_name(void) {
 
@@ -498,17 +577,17 @@ xbt_dict_t SIMIX_process_get_properties(smx_process_t process)
   return process->properties;
 }
 
-void SIMIX_pre_process_sleep(smx_req_t req)
+void SIMIX_pre_process_sleep(smx_simcall_t simcall)
 {
   if (MC_IS_ENABLED) {
-    MC_process_clock_add(req->issuer, req->process_sleep.duration);
-    req->process_sleep.result = SIMIX_DONE;
-    SIMIX_request_answer(req);
+    MC_process_clock_add(simcall->issuer, simcall->process_sleep.duration);
+    simcall->process_sleep.result = SIMIX_DONE;
+    SIMIX_simcall_answer(simcall);
     return;
   }
-  smx_action_t action = SIMIX_process_sleep(req->issuer, req->process_sleep.duration);
-  xbt_fifo_push(action->request_list, req);
-  req->issuer->waiting_action = action;
+  smx_action_t action = SIMIX_process_sleep(simcall->issuer, simcall->process_sleep.duration);
+  xbt_fifo_push(action->simcalls, simcall);
+  simcall->issuer->waiting_action = action;
 }
 
 smx_action_t SIMIX_process_sleep(smx_process_t process, double duration)
@@ -542,14 +621,15 @@ smx_action_t SIMIX_process_sleep(smx_process_t process, double duration)
 
 void SIMIX_post_process_sleep(smx_action_t action)
 {
-  smx_req_t req;
+  smx_simcall_t simcall;
   e_smx_state_t state;
 
-  while ((req = xbt_fifo_shift(action->request_list))) {
+  while ((simcall = xbt_fifo_shift(action->simcalls))) {
 
     switch(surf_workstation_model->action_state_get(action->sleep.surf_sleep)){
       case SURF_ACTION_FAILED:
-        state = SIMIX_SRC_HOST_FAILURE;
+        simcall->issuer->context->iwannadie = 1;
+        //SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
         break;
 
       case SURF_ACTION_DONE:
@@ -560,9 +640,14 @@ void SIMIX_post_process_sleep(smx_action_t action)
         THROW_IMPOSSIBLE;
         break;
     }
-    req->process_sleep.result = state;
-    req->issuer->waiting_action = NULL;
-    SIMIX_request_answer(req);
+    if (surf_workstation_model->extension.
+        workstation.get_state(simcall->issuer->smx_host->host) != SURF_RESOURCE_ON) {
+      simcall->issuer->context->iwannadie = 1;
+    }
+    simcall->process_sleep.result = state;
+    simcall->issuer->waiting_action = NULL;
+    SIMIX_simcall_answer(simcall);
+
   }
   SIMIX_process_sleep_destroy(action);
 }
@@ -603,20 +688,26 @@ void SIMIX_process_yield(smx_process_t self)
   /* Ok, maestro returned control to us */
   XBT_DEBUG("Control returned to me: '%s'", self->name);
 
+  if (self->new_host) {
+    SIMIX_process_change_host(self, self->new_host);
+    self->new_host = NULL;
+  }
+
   if (self->context->iwannadie){
     XBT_DEBUG("I wanna die!");
-    SIMIX_context_stop(self->context);
+    SIMIX_process_stop(self);
+  }
+
+  if(self->suspended) {
+    xbt_assert(!self->doexception, "Gloups! This exception may be lost by subsequent calls.");
+    self->suspended = 0;
+    SIMIX_process_suspend(self,self);
   }
 
   if (self->doexception) {
     XBT_DEBUG("Wait, maestro left me an exception");
     self->doexception = 0;
-    RETHROW;
-  }
-  
-  if (self->new_host) {
-    SIMIX_process_change_host(self, self->new_host);
-    self->new_host = NULL;
+    SMX_THROW();
   }
 }
 
@@ -630,7 +721,7 @@ xbt_running_ctx_t *SIMIX_process_get_running_context(void)
 void SIMIX_process_exception_terminate(xbt_ex_t * e)
 {
   xbt_ex_display(e);
-  abort();
+  xbt_abort();
 }
 
 smx_context_t SIMIX_process_get_context(smx_process_t p) {
@@ -654,11 +745,101 @@ xbt_dynar_t SIMIX_process_get_runnable(void)
  */
 smx_process_t SIMIX_process_from_PID(int PID)
 {
-       smx_process_t proc;
-       xbt_swag_foreach(proc, simix_global->process_list)
-       {
-        if(proc->pid == PID)
-        return proc;
-       }
-       return NULL;
+  smx_process_t proc;
+  xbt_swag_foreach(proc, simix_global->process_list)
+  {
+   if(proc->pid == PID)
+   return proc;
+  }
+  return NULL;
+}
+
+/** @brief returns a dynar containg all currently existing processes */
+xbt_dynar_t SIMIX_processes_as_dynar(void) {
+  smx_process_t proc;
+  xbt_dynar_t res = xbt_dynar_new(sizeof(smx_process_t),NULL);
+  xbt_swag_foreach(proc, simix_global->process_list) {
+    xbt_dynar_push(res,&proc);
+  }
+  return res;
+}
+void SIMIX_process_on_exit_runall(smx_process_t process) {
+  s_smx_process_exit_fun_t exit_fun;
+
+  while (!xbt_dynar_is_empty(process->on_exit)) {
+    exit_fun = xbt_dynar_pop_as(process->on_exit,s_smx_process_exit_fun_t);
+    (exit_fun.fun)(exit_fun.arg);
+  }
+}
+void SIMIX_process_on_exit(smx_process_t process, int_f_pvoid_t fun, void *data) {
+  xbt_assert(process, "current process not found: are you in maestro context ?");
+
+  if (!process->on_exit) {
+    process->on_exit = xbt_dynar_new(sizeof(s_smx_process_exit_fun_t), NULL);
+  }
+
+  s_smx_process_exit_fun_t exit_fun = {fun, data};
+
+  xbt_dynar_push_as(process->on_exit,s_smx_process_exit_fun_t,exit_fun);
+}
+/**
+ * \brief Sets the auto-restart status of the process.
+ * If set to 1, the process will be automatically restarted when its host
+ * comes back.
+ */
+void SIMIX_process_auto_restart_set(smx_process_t process, int auto_restart) {
+  process->auto_restart = auto_restart;
+}
+/**
+ * \brief Restart a process.
+ * Restart a process, starting it again from the beginning.
+ */
+smx_process_t SIMIX_process_restart(smx_process_t process, smx_process_t issuer) {
+  XBT_DEBUG("Restarting process %s on %s", process->name, process->smx_host->name);
+  //retrieve the arguments of the old process
+  //FIXME: Factorise this with SIMIX_host_add_auto_restart_process ?
+  s_smx_process_arg_t arg;
+  arg.code = process->code;
+  arg.hostname = process->smx_host->name;
+  arg.kill_time = process->kill_time;
+  arg.argc = process->argc;
+  arg.data = process->data;
+  int i;
+  arg.argv = xbt_new(char*,process->argc + 1);
+  for (i = 0; i < arg.argc; i++) {
+    arg.argv[i] = xbt_strdup(process->argv[i]);
+  }
+  arg.argv[process->argc] = NULL;
+  arg.properties = NULL;
+  arg.auto_restart = process->auto_restart;
+  //kill the old process
+  SIMIX_process_kill(process,issuer);
+  //start the new process
+  smx_process_t new_process;
+  if (simix_global->create_process_function) {
+    simix_global->create_process_function(&new_process,
+                                          arg.argv[0],
+                                          arg.code,
+                                          arg.data,
+                                          arg.hostname,
+                                          arg.kill_time,
+                                          arg.argc,
+                                          arg.argv,
+                                          arg.properties,
+                                          arg.auto_restart);
+  }
+  else {
+    simcall_process_create(&new_process,
+                                          arg.argv[0],
+                                          arg.code,
+                                          arg.data,
+                                          arg.hostname,
+                                          arg.kill_time,
+                                          arg.argc,
+                                          arg.argv,
+                                          arg.properties,
+                                          arg.auto_restart);
+
+  }
+  return new_process;
 }