static int rte_finalize(void)
{
int ret = ORTE_SUCCESS;
if (app_init_complete) {
/* if I am a daemon, finalize using the default procedure */
if (ORTE_PROC_IS_DAEMON) {
if (ORTE_SUCCESS != (ret = orte_ess_base_orted_finalize())) {
ORTE_ERROR_LOG(ret);
}
} else {
/* use the default app procedure to finish */
if (ORTE_SUCCESS != (ret = orte_ess_base_app_finalize())) {
ORTE_ERROR_LOG(ret);
}
/* remove the envars that we pushed into environ
* so we leave that structure intact
*/
unsetenv("OMPI_MCA_grpcomm");
unsetenv("OMPI_MCA_routed");
unsetenv("OMPI_MCA_orte_precondition_transports");
}
}
/* deconstruct my nidmap and jobmap arrays - this
* function protects itself from being called
* before things were initialized
*/
orte_util_nidmap_finalize();
#if OPAL_HAVE_HWLOC
if (NULL != opal_hwloc_topology) {
opal_hwloc_base_free_topology(opal_hwloc_topology);
opal_hwloc_topology = NULL;
}
#endif
return ret;
}
static int rte_finalize(void)
{
int ret;
/* use the default procedure to finish */
if (ORTE_SUCCESS != (ret = orte_ess_base_app_finalize())) {
ORTE_ERROR_LOG(ret);
}
/* remove the envars that we pushed into environ
* so we leave that structure intact
*/
unsetenv("OMPI_MCA_grpcomm");
unsetenv("OMPI_MCA_routed");
/* deconstruct my nidmap and jobmap arrays - this
* function protects itself from being called
* before things were initialized
*/
orte_util_nidmap_finalize();
return ret;
}
static int rte_ft_event(int state)
{
int ret, exit_status = ORTE_SUCCESS;
orte_proc_type_t svtype;
orte_grpcomm_collective_t coll;
OBJ_CONSTRUCT(&coll, orte_grpcomm_collective_t);
coll.id = orte_process_info.peer_init_barrier;
/******** Checkpoint Prep ********/
if(OPAL_CRS_CHECKPOINT == state) {
/*
* Notify SnapC
*/
if( ORTE_SUCCESS != (ret = orte_snapc.ft_event(OPAL_CRS_CHECKPOINT))) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
/*
* Notify Routed
*/
if( ORTE_SUCCESS != (ret = orte_routed.ft_event(OPAL_CRS_CHECKPOINT))) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
/*
* Notify RML -> OOB
*/
if( ORTE_SUCCESS != (ret = orte_rml.ft_event(OPAL_CRS_CHECKPOINT))) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
}
/******** Continue Recovery ********/
else if (OPAL_CRS_CONTINUE == state ) {
OPAL_OUTPUT_VERBOSE((1, orte_ess_base_output,
"ess:env ft_event(%2d) - %s is Continuing",
state, ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
/*
* Notify RML -> OOB
*/
if( ORTE_SUCCESS != (ret = orte_rml.ft_event(OPAL_CRS_CONTINUE))) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
/*
* Notify Routed
*/
if( ORTE_SUCCESS != (ret = orte_routed.ft_event(OPAL_CRS_CONTINUE))) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
/*
* Notify SnapC
*/
if( ORTE_SUCCESS != (ret = orte_snapc.ft_event(OPAL_CRS_CONTINUE))) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
if( orte_cr_continue_like_restart ) {
/*
* Barrier to make all processes have been successfully restarted before
* we try to remove some restart only files.
*/
if (ORTE_SUCCESS != (ret = orte_grpcomm.barrier(&coll))) {
opal_output(0, "ess:env: ft_event(%2d): Failed in orte_grpcomm.barrier (%d)",
state, ret);
exit_status = ret;
goto cleanup;
}
ORTE_WAIT_FOR_COMPLETION(coll.active);
if( orte_cr_flush_restart_files ) {
OPAL_OUTPUT_VERBOSE((1, orte_ess_base_output,
"ess:env ft_event(%2d): %s "
"Cleanup restart files...",
state, ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
opal_crs_base_cleanup_flush();
}
}
}
/******** Restart Recovery ********/
else if (OPAL_CRS_RESTART == state ) {
OPAL_OUTPUT_VERBOSE((1, orte_ess_base_output,
"ess:env ft_event(%2d) - %s is Restarting",
state, ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
/*
* This should follow the ess init() function
*/
/*
* Clear nidmap and jmap
*/
orte_util_nidmap_finalize();
/*
* - Reset Contact information
*/
if( ORTE_SUCCESS != (ret = env_set_name() ) ) {
exit_status = ret;
}
/*
* Notify RML -> OOB
*/
if( ORTE_SUCCESS != (ret = orte_rml.ft_event(OPAL_CRS_RESTART))) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
/*
* Restart the routed framework
* JJH: Lie to the finalize function so it does not try to contact the daemon.
*/
svtype = orte_process_info.proc_type;
orte_process_info.proc_type = ORTE_PROC_TOOL;
if (ORTE_SUCCESS != (ret = orte_routed.finalize()) ) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
orte_process_info.proc_type = svtype;
if (ORTE_SUCCESS != (ret = orte_routed.initialize()) ) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
/*
* Group Comm - Clean out stale data
*/
orte_grpcomm.finalize();
if (ORTE_SUCCESS != (ret = orte_grpcomm.init())) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
if (ORTE_SUCCESS != (ret = orte_db.remove(NULL, NULL))) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
/*
* Restart the PLM - Does nothing at the moment, but included for completeness
*/
if (ORTE_SUCCESS != (ret = orte_plm.finalize())) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
if (ORTE_SUCCESS != (ret = orte_plm.init())) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
/*
* RML - Enable communications
*/
if (ORTE_SUCCESS != (ret = orte_rml.enable_comm())) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
/*
* Notify Routed
*/
if( ORTE_SUCCESS != (ret = orte_routed.ft_event(OPAL_CRS_RESTART))) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
/* if one was provided, build my nidmap */
if (ORTE_SUCCESS != (ret = orte_util_nidmap_init(orte_process_info.sync_buf))) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
/*
* Barrier to make all processes have been successfully restarted before
* we try to remove some restart only files.
*/
if (ORTE_SUCCESS != (ret = orte_grpcomm.barrier(&coll))) {
opal_output(0, "ess:env ft_event(%2d): Failed in orte_grpcomm.barrier (%d)",
state, ret);
exit_status = ret;
goto cleanup;
}
ORTE_WAIT_FOR_COMPLETION(coll.active);
if( orte_cr_flush_restart_files ) {
OPAL_OUTPUT_VERBOSE((1, orte_ess_base_output,
"ess:env ft_event(%2d): %s "
"Cleanup restart files...",
state, ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
opal_crs_base_cleanup_flush();
}
/*
* Session directory re-init
*/
if (orte_create_session_dirs) {
if (ORTE_SUCCESS != (ret = orte_session_dir(true,
orte_process_info.tmpdir_base,
orte_process_info.nodename,
NULL, /* Batch ID -- Not used */
ORTE_PROC_MY_NAME))) {
exit_status = ret;
}
opal_output_set_output_file_info(orte_process_info.proc_session_dir,
"output-", NULL, NULL);
}
/*
* Notify SnapC
*/
if( ORTE_SUCCESS != (ret = orte_snapc.ft_event(OPAL_CRS_RESTART))) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
}
else if (OPAL_CRS_TERM == state ) {
/* Nothing */
}
else {
/* Error state = Nothing */
}
cleanup:
OBJ_DESTRUCT(&coll);
return exit_status;
}
static int rte_ft_event(int state)
{
int ret, exit_status = ORTE_SUCCESS;
orte_proc_type_t svtype;
/******** Checkpoint Prep ********/
if(OPAL_CRS_CHECKPOINT == state) {
/*
* Notify SnapC
*/
if( ORTE_SUCCESS != (ret = orte_snapc.ft_event(OPAL_CRS_CHECKPOINT))) {
exit_status = ret;
goto cleanup;
}
/*
* Notify Routed
*/
if( ORTE_SUCCESS != (ret = orte_routed.ft_event(OPAL_CRS_CHECKPOINT))) {
exit_status = ret;
goto cleanup;
}
/*
* Notify RML -> OOB
*/
if( ORTE_SUCCESS != (ret = orte_rml.ft_event(OPAL_CRS_CHECKPOINT))) {
exit_status = ret;
goto cleanup;
}
}
/******** Continue Recovery ********/
else if (OPAL_CRS_CONTINUE == state ) {
/*
* Notify RML -> OOB
*/
if( ORTE_SUCCESS != (ret = orte_rml.ft_event(OPAL_CRS_CONTINUE))) {
exit_status = ret;
goto cleanup;
}
/*
* Notify Routed
*/
if( ORTE_SUCCESS != (ret = orte_routed.ft_event(OPAL_CRS_CONTINUE))) {
exit_status = ret;
goto cleanup;
}
/*
* Notify SnapC
*/
if( ORTE_SUCCESS != (ret = orte_snapc.ft_event(OPAL_CRS_CONTINUE))) {
exit_status = ret;
goto cleanup;
}
}
/******** Restart Recovery ********/
else if (OPAL_CRS_RESTART == state ) {
/*
* This should follow the ess init() function
*/
/*
* Clear nidmap and jmap
*/
orte_util_nidmap_finalize();
/*
* - Reset Contact information
*/
if( ORTE_SUCCESS != (ret = slave_set_name() ) ) {
exit_status = ret;
}
/*
* Notify RML -> OOB
*/
if( ORTE_SUCCESS != (ret = orte_rml.ft_event(OPAL_CRS_RESTART))) {
exit_status = ret;
goto cleanup;
}
/*
* Restart the routed framework
* JJH: Lie to the finalize function so it does not try to contact the daemon.
*/
svtype = orte_process_info.proc_type;
orte_process_info.proc_type = ORTE_PROC_TOOL;
if (ORTE_SUCCESS != (ret = orte_routed.finalize()) ) {
exit_status = ret;
goto cleanup;
}
orte_process_info.proc_type = svtype;
if (ORTE_SUCCESS != (ret = orte_routed.initialize()) ) {
exit_status = ret;
goto cleanup;
}
/*
* Group Comm - Clean out stale data
*/
orte_grpcomm.finalize();
if (ORTE_SUCCESS != (ret = orte_grpcomm.init())) {
exit_status = ret;
goto cleanup;
}
if (ORTE_SUCCESS != (ret = orte_grpcomm.purge_proc_attrs())) {
exit_status = ret;
goto cleanup;
}
/*
* Restart the PLM - Does nothing at the moment, but included for completeness
*/
if (ORTE_SUCCESS != (ret = orte_plm.finalize())) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
if (ORTE_SUCCESS != (ret = orte_plm.init())) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
/*
* RML - Enable communications
*/
if (ORTE_SUCCESS != (ret = orte_rml.enable_comm())) {
exit_status = ret;
goto cleanup;
}
/*
* Session directory re-init
*/
if (orte_create_session_dirs) {
if (ORTE_SUCCESS != (ret = orte_session_dir(true,
orte_process_info.tmpdir_base,
orte_process_info.nodename,
NULL, /* Batch ID -- Not used */
ORTE_PROC_MY_NAME))) {
exit_status = ret;
}
opal_output_set_output_file_info(orte_process_info.proc_session_dir,
"output-", NULL, NULL);
}
/*
* Notify Routed
*/
if( ORTE_SUCCESS != (ret = orte_routed.ft_event(OPAL_CRS_RESTART))) {
exit_status = ret;
goto cleanup;
}
/*
* Notify SnapC
*/
if( ORTE_SUCCESS != (ret = orte_snapc.ft_event(OPAL_CRS_RESTART))) {
exit_status = ret;
goto cleanup;
}
/*
* Send new PID to HNP/daemon
* The checkpointer could have used a proxy program to boot us
* so the pid that the orted got from fork() may not be the
* PID of this application.
* - Note: BLCR does this because it tries to preseve the PID
* of the program across checkpointes
*/
if( ORTE_SUCCESS != (ret = ess_slave_ft_event_update_process_info(orte_process_info.my_name, getpid())) ) {
exit_status = ret;
goto cleanup;
}
/* if one was provided, build my nidmap */
if (ORTE_SUCCESS != (ret = orte_util_nidmap_init(orte_process_info.sync_buf))) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
}
else if (OPAL_CRS_TERM == state ) {
/* Nothing */
}
else {
/* Error state = Nothing */
}
cleanup:
return exit_status;
}
