X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/e276e9953a9ab5894bbb623fc17ec79512efab5d..23f5bf52cc58cf837840f231cf638e9fcd45097b:/src/simix/smx_host.c diff --git a/src/simix/smx_host.c b/src/simix/smx_host.c index fec6e6990b..2308f27d4b 100644 --- a/src/simix/smx_host.c +++ b/src/simix/smx_host.c @@ -1,18 +1,18 @@ -/* 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" +#include "mc/mc.h" XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_host, simix, "Logging specific to SIMIX (hosts)"); - static void SIMIX_execution_finish(smx_action_t action); /** @@ -35,8 +35,7 @@ smx_host_t SIMIX_host_create(const char *name, xbt_swag_new(xbt_swag_offset(proc, host_proc_hookup)); /* Update global variables */ - xbt_dict_set(simix_global->host, smx_host->name, smx_host, - &SIMIX_host_destroy); + xbt_lib_set(host_lib,smx_host->name,SIMIX_HOST_LEVEL,smx_host); return smx_host; } @@ -50,7 +49,7 @@ void SIMIX_host_destroy(void *h) { smx_host_t host = (smx_host_t) h; - xbt_assert0((host != NULL), "Invalid parameters"); + xbt_assert((host != NULL), "Invalid parameters"); /* Clean Simulator data */ if (xbt_swag_size(host->process_list) != 0) { @@ -65,9 +64,9 @@ void SIMIX_host_destroy(void *h) msg = tmp; } SIMIX_display_process_status(); - THROW1(arg_error, 0, "%s", msg); + THROWF(arg_error, 0, "%s", msg); } - + xbt_dynar_free(&host->auto_restart_processes); xbt_swag_free(host->process_list); /* Clean host structure */ @@ -84,16 +83,25 @@ void SIMIX_host_destroy(void *h) */ xbt_dict_t SIMIX_host_get_dict(void) { - return simix_global->host; + xbt_dict_t host_dict = xbt_dict_new_homogeneous(NULL); + xbt_lib_cursor_t cursor = NULL; + char *name = NULL; + void **host = NULL; + + xbt_lib_foreach(host_lib, cursor, name, host){ + if(host[SIMIX_HOST_LEVEL]) + xbt_dict_set(host_dict,name,host[SIMIX_HOST_LEVEL], NULL); + } + return host_dict; } smx_host_t SIMIX_host_get_by_name(const char *name) { - xbt_assert0(((simix_global != NULL) - && (simix_global->host != NULL)), + xbt_assert(((simix_global != NULL) + && (host_lib != NULL)), "Environment not set yet"); - return xbt_dict_get_or_null(simix_global->host, name); + return xbt_lib_get_or_null(host_lib, name, SIMIX_HOST_LEVEL); } smx_host_t SIMIX_host_self(void) @@ -102,7 +110,7 @@ smx_host_t SIMIX_host_self(void) return (process == NULL) ? NULL : SIMIX_process_get_host(process); } -/* 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_host_self_get_name(void) { @@ -115,21 +123,21 @@ const char* SIMIX_host_self_get_name(void) const char* SIMIX_host_get_name(smx_host_t host) { - xbt_assert0((host != NULL), "Invalid parameters"); + xbt_assert((host != NULL), "Invalid parameters"); return host->name; } xbt_dict_t SIMIX_host_get_properties(smx_host_t host) { - xbt_assert0((host != NULL), "Invalid parameters (simix host is NULL)"); + xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)"); return surf_workstation_model->extension.workstation.get_properties(host->host); } double SIMIX_host_get_speed(smx_host_t host) { - xbt_assert0((host != NULL), "Invalid parameters (simix host is NULL)"); + xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)"); return surf_workstation_model->extension.workstation. get_speed(host->host, 1.0); @@ -137,7 +145,7 @@ double SIMIX_host_get_speed(smx_host_t host) double SIMIX_host_get_available_speed(smx_host_t host) { - xbt_assert0((host != NULL), "Invalid parameters (simix host is NULL)"); + xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)"); return surf_workstation_model->extension.workstation. get_available_speed(host->host); @@ -145,7 +153,7 @@ double SIMIX_host_get_available_speed(smx_host_t host) int SIMIX_host_get_state(smx_host_t host) { - xbt_assert0((host != NULL), "Invalid parameters (simix host is NULL)"); + xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)"); return surf_workstation_model->extension.workstation. get_state(host->host); @@ -163,27 +171,128 @@ void SIMIX_host_self_set_data(void *data) void* SIMIX_host_get_data(smx_host_t host) { - xbt_assert0((host != NULL), "Invalid parameters (simix host is NULL)"); + xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)"); 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; + xbt_free(arg->name); + xbt_free(arg); +} +/** + * \brief Add a process to the list of the processes that the host will restart when it comes back + * This function add a process to the list of the processes that will be restarted when the host comes + * back. It is expected that this function is called when the host is down. + * The processes will only be restarted once, meaning that you will have to register the process + * again to restart the process again. + */ +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->argv[argc] = NULL; + + arg->properties = properties; + arg->auto_restart = auto_restart; + + if( SIMIX_host_get_state(host) == SURF_RESOURCE_OFF + && !xbt_dict_get_or_null(watched_hosts_lib,host->name)){ + xbt_dict_set(watched_hosts_lib,host->name,host,NULL); + XBT_DEBUG("Have push host %s to watched_hosts_lib because state == SURF_RESOURCE_OFF",host->name); + } + xbt_dynar_push_as(host->auto_restart_processes,smx_process_arg_t,arg); +} +/** + * \brief Restart the list of processes that have been registered to the host + */ +void SIMIX_host_restart_processes(smx_host_t host) +{ + unsigned int cpt; + smx_process_arg_t arg; + xbt_dynar_foreach(host->auto_restart_processes,cpt,arg) { + + smx_process_t process; + + XBT_DEBUG("Restarting Process %s(%s) right now", arg->argv[0], arg->hostname); + if (simix_global->create_process_function) { + simix_global->create_process_function(&process, + arg->argv[0], + arg->code, + NULL, + arg->hostname, + arg->kill_time, + arg->argc, + arg->argv, + arg->properties, + arg->auto_restart); + } + else { + simcall_process_create(&process, + arg->argv[0], + arg->code, + NULL, + arg->hostname, + arg->kill_time, + arg->argc, + arg->argv, + arg->properties, + arg->auto_restart); + + } + } + xbt_dynar_reset(host->auto_restart_processes); +} + +void SIMIX_host_autorestart(smx_host_t host) +{ + if(simix_global->autorestart) + simix_global->autorestart(host); + else + xbt_die("No function for simix_global->autorestart"); +} void SIMIX_host_set_data(smx_host_t host, void *data) { - xbt_assert0((host != NULL), "Invalid parameters"); - xbt_assert0((host->data == NULL), "Data already set"); + xbt_assert((host != NULL), "Invalid parameters"); + xbt_assert((host->data == NULL), "Data already set"); host->data = data; } smx_action_t SIMIX_host_execute(const char *name, smx_host_t host, - double computation_amount) + double computation_amount, + double priority) { /* alloc structures and initialize */ - smx_action_t action = xbt_new0(s_smx_action_t, 1); + smx_action_t action = xbt_mallocator_get(simix_global->action_mallocator); action->type = SIMIX_ACTION_EXECUTE; action->name = xbt_strdup(name); - action->request_list = xbt_fifo_new(); action->state = SIMIX_RUNNING; action->execution.host = host; @@ -191,23 +300,16 @@ smx_action_t SIMIX_host_execute(const char *name, smx_host_t host, action->category = NULL; #endif -#ifdef HAVE_MC /* set surf's action */ - if (!_surf_do_model_check) -#endif - { - action->execution.surf_exec = - surf_workstation_model->extension.workstation.execute(host->host, - computation_amount); - surf_workstation_model->action_data_set(action->execution.surf_exec, action); + if (!MC_is_active()) { + action->execution.surf_exec = + surf_workstation_model->extension.workstation.execute(host->host, + computation_amount); + surf_workstation_model->action_data_set(action->execution.surf_exec, action); + surf_workstation_model->set_priority(action->execution.surf_exec, priority); } -#ifdef HAVE_TRACING - TRACE_smx_host_execute(action); - TRACE_surf_action(action->execution.surf_exec, action->category); -#endif - - DEBUG1("Create execute action %p", action); + XBT_DEBUG("Create execute action %p", action); return action; } @@ -221,10 +323,9 @@ smx_action_t SIMIX_host_parallel_execute( const char *name, int i; /* alloc structures and initialize */ - smx_action_t action = xbt_new0(s_smx_action_t, 1); + smx_action_t action = xbt_mallocator_get(simix_global->action_mallocator); action->type = SIMIX_ACTION_PARALLEL_EXECUTE; action->name = xbt_strdup(name); - action->request_list = xbt_fifo_new(); action->state = SIMIX_RUNNING; action->execution.host = NULL; /* FIXME: do we need the list of hosts? */ @@ -237,48 +338,37 @@ smx_action_t SIMIX_host_parallel_execute( const char *name, for (i = 0; i < host_nb; i++) workstation_list[i] = host_list[i]->host; -#ifdef HAVE_MC /* set surf's action */ - if (!_surf_do_model_check) -#endif - { + if (!MC_is_active()) { action->execution.surf_exec = surf_workstation_model->extension.workstation. execute_parallel_task(host_nb, workstation_list, computation_amount, - communication_amount, amount, rate); + communication_amount, rate); - surf_workstation_model->action_data_set(action->execution.surf_exec, action); + surf_workstation_model->action_data_set(action->execution.surf_exec, action); } - DEBUG1("Create parallel execute action %p", action); + XBT_DEBUG("Create parallel execute action %p", action); return action; } void SIMIX_host_execution_destroy(smx_action_t action) { - DEBUG1("Destroy action %p", action); - - if (action->name) - xbt_free(action->name); - - xbt_fifo_free(action->request_list); + XBT_DEBUG("Destroy action %p", action); if (action->execution.surf_exec) { surf_workstation_model->action_unref(action->execution.surf_exec); action->execution.surf_exec = NULL; } - -#ifdef HAVE_TRACING - TRACE_smx_action_destroy(action); -#endif - xbt_free(action); + xbt_free(action->name); + xbt_mallocator_release(simix_global->action_mallocator, action); } void SIMIX_host_execution_cancel(smx_action_t action) { - DEBUG1("Cancel action %p", action); + XBT_DEBUG("Cancel action %p", action); - if (action->execution.surf_exec) + if (action->execution.surf_exec) surf_workstation_model->action_cancel(action->execution.surf_exec); } @@ -299,26 +389,26 @@ e_smx_state_t SIMIX_host_execution_get_state(smx_action_t action) void SIMIX_host_execution_set_priority(smx_action_t action, double priority) { - surf_workstation_model->set_priority(action->execution.surf_exec, priority); + if(action->execution.surf_exec) + surf_workstation_model->set_priority(action->execution.surf_exec, priority); } -void SIMIX_pre_host_execution_wait(smx_req_t req) +void SIMIX_pre_host_execution_wait(smx_simcall_t simcall) { - smx_action_t action = req->host_execution_wait.execution; + smx_action_t action = simcall->host_execution_wait.execution; - DEBUG2("Wait for execution of action %p, state %d", action, action->state); + XBT_DEBUG("Wait for execution of action %p, state %d", action, (int)action->state); - /* Associate this request to the action */ - xbt_fifo_push(action->request_list, req); - req->issuer->waiting_action = action; + /* Associate this simcall to the action */ + xbt_fifo_push(action->simcalls, simcall); + simcall->issuer->waiting_action = action; -#ifdef HAVE_MC /* set surf's action */ - if (_surf_do_model_check){ + if (MC_is_active()) { action->state = SIMIX_DONE; SIMIX_execution_finish(action); + return; } -#endif /* If the action is already finished then perform the error handling */ if (action->state != SIMIX_RUNNING) @@ -327,74 +417,97 @@ void SIMIX_pre_host_execution_wait(smx_req_t req) void SIMIX_host_execution_suspend(smx_action_t action) { - surf_workstation_model->suspend(action->execution.surf_exec); + if(action->execution.surf_exec) + surf_workstation_model->suspend(action->execution.surf_exec); } void SIMIX_host_execution_resume(smx_action_t action) { - surf_workstation_model->suspend(action->execution.surf_exec); + if(action->execution.surf_exec) + surf_workstation_model->resume(action->execution.surf_exec); } void SIMIX_execution_finish(smx_action_t action) { xbt_fifo_item_t item; - smx_req_t req; + smx_simcall_t simcall; - xbt_fifo_foreach(action->request_list, item, req, smx_req_t) { + xbt_fifo_foreach(action->simcalls, item, simcall, smx_simcall_t) { switch (action->state) { case SIMIX_DONE: - /* do nothing, action done*/ - DEBUG0("SIMIX_execution_finished: execution successful"); + /* do nothing, action done */ + XBT_DEBUG("SIMIX_execution_finished: execution successful"); break; case SIMIX_FAILED: - TRY { - DEBUG1("SIMIX_execution_finished: host '%s' failed", req->issuer->smx_host->name); - THROW0(host_error, 0, "Host failed"); - } - CATCH(req->issuer->running_ctx->exception) { - req->issuer->doexception = 1; - } - break; + XBT_DEBUG("SIMIX_execution_finished: host '%s' failed", simcall->issuer->smx_host->name); + simcall->issuer->context->iwannadie = 1; + //SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed"); + break; case SIMIX_CANCELED: - TRY { - DEBUG0("SIMIX_execution_finished: execution canceled"); - THROW0(cancel_error, 0, "Canceled"); - } - CATCH(req->issuer->running_ctx->exception) { - req->issuer->doexception = 1; - } - break; + XBT_DEBUG("SIMIX_execution_finished: execution canceled"); + SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Canceled"); + break; default: - THROW_IMPOSSIBLE; + xbt_die("Internal error in SIMIX_execution_finish: unexpected action state %d", + (int)action->state); + } + /* 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; } - req->issuer->waiting_action = NULL; - SIMIX_request_answer(req); + + simcall->issuer->waiting_action = NULL; + simcall->host_execution_wait.result = action->state; + SIMIX_simcall_answer(simcall); } + + /* We no longer need it */ + SIMIX_host_execution_destroy(action); } void SIMIX_post_host_execute(smx_action_t action) { - /* FIXME: check if the host running the action failed or not*/ - /*if(surf_workstation_model->extension.workstation.get_state(action->host->host))*/ - - /* If the host running the action didn't fail, then the action was cancelled */ - if (surf_workstation_model->action_state_get(action->execution.surf_exec) == SURF_ACTION_FAILED) - action->state = SIMIX_CANCELED; - else - action->state = SIMIX_DONE; + 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; + } else { + action->state = SIMIX_DONE; + } if (action->execution.surf_exec) { surf_workstation_model->action_unref(action->execution.surf_exec); action->execution.surf_exec = NULL; } - /* If there are requests associated with the action, then answer them */ - if (xbt_fifo_size(action->request_list)) + /* If there are simcalls associated with the action, then answer them */ + if (xbt_fifo_size(action->simcalls)) { SIMIX_execution_finish(action); + } } + +#ifdef HAVE_TRACING +void SIMIX_set_category(smx_action_t action, const char *category) +{ + if (action->state != SIMIX_RUNNING) return; + if (action->type == SIMIX_ACTION_EXECUTE){ + surf_workstation_model->set_category(action->execution.surf_exec, category); + }else if (action->type == SIMIX_ACTION_COMMUNICATE){ + surf_workstation_model->set_category(action->comm.surf_comm, category); + } +} +#endif +