-/* $Id$ */
-
-/* Copyright (c) 2007 Arnaud Legrand, Bruno Donassolo.
- All rights reserved. */
+/* 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"
-/** \defgroup m_host_management Management functions of Hosts
- * \brief This section describes the host structure of MSG
- *
- * \htmlonly <!-- DOXYGEN_NAVBAR_LABEL="Hosts" --> \endhtmlonly
- * (#m_host_t) and the functions for managing it.
- *
- * A <em>location</em> (or <em>host</em>) is any possible place where
- * a process may run. Thus it may be represented as a
- * <em>physical resource with computing capabilities</em>, some
- * <em>mailboxes</em> to enable running process to communicate with
- * remote ones, and some <em>private data</em> that can be only
- * accessed by local process.
- * \see m_host_t
- */
+XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_host, simix,
+ "Logging specific to SIMIX (hosts)");
-/********************************* Host **************************************/
-smx_host_t __SIMIX_host_create(const char *name,
- void *workstation,
- void *data)
+static void SIMIX_execution_finish(smx_action_t action);
+
+/**
+ * \brief Internal function to create a SIMIX host.
+ * \param name name of the host to create
+ * \param workstation the SURF workstation to encapsulate
+ * \param data some user data (may be NULL)
+ */
+smx_host_t SIMIX_host_create(const char *name,
+ void *workstation, void *data)
{
- simdata_host_t simdata = xbt_new0(s_simdata_host_t,1);
- smx_host_t host = xbt_new0(s_smx_host_t,1);
+ smx_host_t smx_host = xbt_new0(s_smx_host_t, 1);
s_smx_process_t proc;
/* Host structure */
- host->name = xbt_strdup(name);
- host->simdata = simdata;
- host->data = data;
-
- simdata->host = workstation;
+ smx_host->name = xbt_strdup(name);
+ smx_host->data = data;
+ smx_host->host = workstation;
+ smx_host->process_list =
+ xbt_swag_new(xbt_swag_offset(proc, host_proc_hookup));
- simdata->process_list = xbt_swag_new(xbt_swag_offset(proc, host_proc_hookup));
/* Update global variables */
+ xbt_lib_set(host_lib,smx_host->name,SIMIX_HOST_LEVEL,smx_host);
- xbt_fifo_unshift(simix_global->host, host);
-
- return host;
+ return smx_host;
}
-/** \ingroup m_host_management
+/**
+ * \brief Internal function to destroy a SIMIX host.
*
- * \brief Set the user data of a #m_host_t.
- *
- * This functions checks whether some data has already been associated to \a host
- or not and attach \a data to \a host if it is possible.
+ * \param h the host to destroy (a smx_host_t)
*/
-void SIMIX_host_set_data(smx_host_t host, void *data)
+void SIMIX_host_destroy(void *h)
{
- xbt_assert0((host!=NULL), "Invalid parameters");
- xbt_assert0((host->data == NULL), "Data already set");
+ smx_host_t host = (smx_host_t) h;
- /* Assign data */
- host->data = data;
+ xbt_assert((host != NULL), "Invalid parameters");
+
+ /* Clean Simulator data */
+ if (xbt_swag_size(host->process_list) != 0) {
+ char *msg =
+ bprintf("Shutting down host %s, but it's not empty:", host->name);
+ char *tmp;
+ smx_process_t process = NULL;
+
+ xbt_swag_foreach(process, host->process_list) {
+ tmp = bprintf("%s\n\t%s", msg, process->name);
+ free(msg);
+ msg = tmp;
+ }
+ 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 */
+ free(host->name);
+ free(host);
- return ;
+ return;
}
-/** \ingroup m_host_management
+/**
+ * \brief Returns a dict of all hosts.
*
- * \brief Return the user data of a #m_host_t.
- *
- * This functions checks whether \a host is a valid pointer or not and return
- the user data associated to \a host if it is possible.
+ * \return List of all hosts (as a #xbt_dict_t)
*/
-void *SIMIX_host_get_data(smx_host_t host)
+xbt_dict_t SIMIX_host_get_dict(void)
{
+ 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;
+}
- xbt_assert0((host != NULL), "Invalid parameters");
+smx_host_t SIMIX_host_get_by_name(const char *name)
+{
+ xbt_assert(((simix_global != NULL)
+ && (host_lib != NULL)),
+ "Environment not set yet");
- /* Return data */
- return (host->data);
+ return xbt_lib_get_or_null(host_lib, name, SIMIX_HOST_LEVEL);
}
-/** \ingroup m_host_management
- *
- * \brief Return the name of the #m_host_t.
- *
- * This functions checks whether \a host is a valid pointer or not and return
- its name.
- */
-const char *SIMIX_host_get_name(smx_host_t host)
+smx_host_t SIMIX_host_self(void)
{
+ smx_process_t process = SIMIX_process_self();
+ return (process == NULL) ? NULL : SIMIX_process_get_host(process);
+}
- xbt_assert0((host != NULL) && (host->simdata != NULL), "Invalid parameters");
+/* needs to be public and without simcall because it is called
+ by exceptions and logging events */
+const char* SIMIX_host_self_get_name(void)
+{
+ smx_host_t host = SIMIX_host_self();
+ if (host == NULL || SIMIX_process_self() == simix_global->maestro_process)
+ return "";
- /* Return data */
- return (host->name);
+ return SIMIX_host_get_name(host);
}
-/** \ingroup m_host_management
- *
- * \brief Return the location on which the current process is executed.
- */
-smx_host_t SIMIX_host_self(void)
+const char* SIMIX_host_get_name(smx_host_t host)
{
- return SIMIX_process_get_host(SIMIX_process_self());
+ xbt_assert((host != NULL), "Invalid parameters");
+
+ return host->name;
}
-/*
- * Real function for destroy a host.
- * MSG_host_destroy is just a front_end that also removes it from
- * msg_global->host
- */
-void __SIMIX_host_destroy(smx_host_t host)
+xbt_dict_t SIMIX_host_get_properties(smx_host_t host)
{
- simdata_host_t simdata = NULL;
+ xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
- xbt_assert0((host != NULL), "Invalid parameters");
+ return surf_workstation_model->extension.workstation.get_properties(host->host);
+}
-
- /* Clean Simulator data */
- simdata = host->simdata;
+double SIMIX_host_get_speed(smx_host_t host)
+{
+ xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
- xbt_assert0((xbt_swag_size(simdata->process_list)==0),
- "Some process are still running on this host");
- xbt_swag_free(simdata->process_list);
+ return surf_workstation_model->extension.workstation.
+ get_speed(host->host, 1.0);
+}
- free(simdata);
+double SIMIX_host_get_available_speed(smx_host_t host)
+{
+ xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
- /* Clean host structure */
- free(host->name);
- free(host);
+ return surf_workstation_model->extension.workstation.
+ get_available_speed(host->host);
+}
- return;
+int SIMIX_host_get_state(smx_host_t host)
+{
+ xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
+
+ return surf_workstation_model->extension.workstation.
+ get_state(host->host);
}
-/** \ingroup m_host_management
- * \brief Return the current number of #m_host_t.
- */
-int SIMIX_get_host_number(void)
+void* SIMIX_host_self_get_data(void)
{
- return (xbt_fifo_size(simix_global->host));
+ return SIMIX_host_get_data(SIMIX_host_self());
}
-/** \ingroup m_host_management
- * \brief Return a array of all the #m_host_t.
- */
-smx_host_t *SIMIX_get_host_table(void)
+void SIMIX_host_self_set_data(void *data)
{
- return ((smx_host_t *)xbt_fifo_to_array(simix_global->host));
+ SIMIX_host_set_data(SIMIX_host_self(), data);
}
+void* SIMIX_host_get_data(smx_host_t host)
+{
+ xbt_assert((host != NULL), "Invalid parameters (simix host is NULL)");
-/** \ingroup m_host_management
- * \brief Return the speed of the processor (in Mflop/s), regardless of
- the current load on the machine.
+ 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
*/
-double SIMIX_get_host_speed(smx_host_t h)
+void SIMIX_host_restart_processes(smx_host_t host)
{
- xbt_assert0((h!= NULL), "Invalid parameters");
+ 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);
+}
- return(surf_workstation_resource->
- extension_public->get_speed(h->simdata->host,1.0));
+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");
}
-/** \ingroup msg_gos_functions
- * \brief Determine if a host is available.
- *
- * \param h host to test
- */
-int SIMIX_host_is_avail (smx_host_t h)
+void SIMIX_host_set_data(smx_host_t host, void *data)
+{
+ 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 priority)
+{
+ /* alloc structures and initialize */
+ smx_action_t action = xbt_mallocator_get(simix_global->action_mallocator);
+ action->type = SIMIX_ACTION_EXECUTE;
+ action->name = xbt_strdup(name);
+ action->state = SIMIX_RUNNING;
+ action->execution.host = host;
+
+#ifdef HAVE_TRACING
+ action->category = NULL;
+#endif
+
+ /* set surf's 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);
+ }
+
+ XBT_DEBUG("Create execute action %p", action);
+
+ return action;
+}
+
+smx_action_t SIMIX_host_parallel_execute( const char *name,
+ int host_nb, smx_host_t *host_list,
+ double *computation_amount, double *communication_amount,
+ double amount, double rate)
+{
+ void **workstation_list = NULL;
+ int i;
+
+ /* alloc structures and initialize */
+ smx_action_t action = xbt_mallocator_get(simix_global->action_mallocator);
+ action->type = SIMIX_ACTION_PARALLEL_EXECUTE;
+ action->name = xbt_strdup(name);
+ action->state = SIMIX_RUNNING;
+ action->execution.host = NULL; /* FIXME: do we need the list of hosts? */
+
+#ifdef HAVE_TRACING
+ action->category = NULL;
+#endif
+
+ /* set surf's action */
+ workstation_list = xbt_new0(void *, host_nb);
+ for (i = 0; i < host_nb; i++)
+ workstation_list[i] = host_list[i]->host;
+
+ /* set surf's action */
+ if (!MC_is_active()) {
+ action->execution.surf_exec =
+ surf_workstation_model->extension.workstation.
+ execute_parallel_task(host_nb, workstation_list, computation_amount,
+ communication_amount, rate);
+
+ surf_workstation_model->action_data_set(action->execution.surf_exec, action);
+ }
+ XBT_DEBUG("Create parallel execute action %p", action);
+
+ return action;
+}
+
+void SIMIX_host_execution_destroy(smx_action_t action)
+{
+ 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;
+ }
+ xbt_free(action->name);
+ xbt_mallocator_release(simix_global->action_mallocator, action);
+}
+
+void SIMIX_host_execution_cancel(smx_action_t action)
+{
+ XBT_DEBUG("Cancel action %p", action);
+
+ if (action->execution.surf_exec)
+ surf_workstation_model->action_cancel(action->execution.surf_exec);
+}
+
+double SIMIX_host_execution_get_remains(smx_action_t action)
+{
+ double result = 0.0;
+
+ if (action->state == SIMIX_RUNNING)
+ result = surf_workstation_model->get_remains(action->execution.surf_exec);
+
+ return result;
+}
+
+e_smx_state_t SIMIX_host_execution_get_state(smx_action_t action)
+{
+ return action->state;
+}
+
+void SIMIX_host_execution_set_priority(smx_action_t action, double priority)
+{
+ if(action->execution.surf_exec)
+ surf_workstation_model->set_priority(action->execution.surf_exec, priority);
+}
+
+void SIMIX_pre_host_execution_wait(smx_simcall_t simcall)
+{
+ smx_action_t action = simcall->host_execution_wait.execution;
+
+ XBT_DEBUG("Wait for execution of action %p, state %d", action, (int)action->state);
+
+ /* Associate this simcall to the action */
+ xbt_fifo_push(action->simcalls, simcall);
+ simcall->issuer->waiting_action = action;
+
+ /* set surf's action */
+ if (MC_is_active()) {
+ action->state = SIMIX_DONE;
+ SIMIX_execution_finish(action);
+ return;
+ }
+
+ /* If the action is already finished then perform the error handling */
+ if (action->state != SIMIX_RUNNING)
+ SIMIX_execution_finish(action);
+}
+
+void SIMIX_host_execution_suspend(smx_action_t action)
{
- e_surf_cpu_state_t cpustate;
- xbt_assert0((h!= NULL), "Invalid parameters");
+ if(action->execution.surf_exec)
+ surf_workstation_model->suspend(action->execution.surf_exec);
+}
- cpustate =
- surf_workstation_resource->extension_public->get_state(h->simdata->host);
+void SIMIX_host_execution_resume(smx_action_t action)
+{
+ if(action->execution.surf_exec)
+ surf_workstation_model->resume(action->execution.surf_exec);
+}
- xbt_assert0((cpustate == SURF_CPU_ON || cpustate == SURF_CPU_OFF),
- "Invalid cpu state");
+void SIMIX_execution_finish(smx_action_t action)
+{
+ xbt_fifo_item_t item;
+ smx_simcall_t simcall;
+
+ xbt_fifo_foreach(action->simcalls, item, simcall, smx_simcall_t) {
+
+ switch (action->state) {
+
+ case SIMIX_DONE:
+ /* do nothing, action done */
+ XBT_DEBUG("SIMIX_execution_finished: execution successful");
+ break;
+
+ case SIMIX_FAILED:
+ 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:
+ XBT_DEBUG("SIMIX_execution_finished: execution canceled");
+ SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Canceled");
+ break;
+
+ default:
+ 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;
+ }
+
+ 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);
+}
- return (cpustate==SURF_CPU_ON);
+void SIMIX_post_host_execute(smx_action_t action)
+{
+ 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 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
+