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;
}
int mca_bml_r2_ft_event(int state)
{
static bool first_continue_pass = false;
ompi_proc_t** procs = NULL;
size_t num_procs;
size_t btl_idx;
int ret, p;
int loc_state;
int param_type = -1;
char *param_list = NULL;
if(OPAL_CRS_CHECKPOINT == state) {
/* Do nothing for now */
}
else if(OPAL_CRS_CONTINUE == state) {
first_continue_pass = !first_continue_pass;
/* Since nothing in Checkpoint, we are fine here (unless required by BTL) */
if( ompi_cr_continue_like_restart && !first_continue_pass) {
procs = ompi_proc_all(&num_procs);
if(NULL == procs) {
return OMPI_ERR_OUT_OF_RESOURCE;
}
}
}
else if(OPAL_CRS_RESTART_PRE == state ) {
/* Nothing here */
}
else if(OPAL_CRS_RESTART == state ) {
procs = ompi_proc_all(&num_procs);
if(NULL == procs) {
return OMPI_ERR_OUT_OF_RESOURCE;
}
}
else if(OPAL_CRS_TERM == state ) {
;
}
else {
;
}
/* Never call the ft_event functions attached to the BTLs on the second
* pass of RESTART since on the first pass they were unloaded and therefore
* no longer exist.
*/
if( OPAL_CRS_RESTART != state ) {
if( OPAL_CRS_CONTINUE == state && !first_continue_pass ) {
;
} else {
/* Since we only ever call into the BTLs once during the first restart
* pass, just lie to them on this pass for a bit of local clarity.
*/
if( OPAL_CRS_RESTART_PRE == state ) {
loc_state = OPAL_CRS_RESTART;
} else {
loc_state = state;
}
/*
* Call ft_event in:
* - BTL modules
* - MPool modules
*
* These should be cleaning out stale state, and memory references in
* preparation for being shut down.
*/
for(btl_idx = 0; btl_idx < mca_bml_r2.num_btl_modules; btl_idx++) {
/*
* Notify Mpool
*/
if( NULL != (mca_bml_r2.btl_modules[btl_idx])->btl_mpool &&
NULL != (mca_bml_r2.btl_modules[btl_idx])->btl_mpool->mpool_ft_event ) {
opal_output_verbose(10, ompi_cr_output,
"bml:r2: ft_event: Notify the %s MPool.\n",
(mca_bml_r2.btl_modules[btl_idx])->btl_mpool->mpool_component->mpool_version.mca_component_name);
if(OMPI_SUCCESS != (ret = (mca_bml_r2.btl_modules[btl_idx])->btl_mpool->mpool_ft_event(loc_state) ) ) {
continue;
}
}
/*
* Notify BTL
*/
if( NULL != (mca_bml_r2.btl_modules[btl_idx])->btl_ft_event) {
opal_output_verbose(10, ompi_cr_output,
"bml:r2: ft_event: Notify the %s BTL.\n",
(mca_bml_r2.btl_modules[btl_idx])->btl_component->btl_version.mca_component_name);
if(OMPI_SUCCESS != (ret = (mca_bml_r2.btl_modules[btl_idx])->btl_ft_event(loc_state) ) ) {
continue;
}
}
}
} /* OPAL_CRS_CONTINUE == state && !first_continue_pass */
}
if(OPAL_CRS_CHECKPOINT == state) {
;
}
else if(OPAL_CRS_CONTINUE == state) {
/* Matches OPAL_CRS_RESTART_PRE */
if( ompi_cr_continue_like_restart && first_continue_pass) {
if( OMPI_SUCCESS != (ret = mca_bml_r2_finalize()) ) {
opal_output(0, "bml:r2: ft_event(Restart): Failed to finalize BML framework\n");
return ret;
}
}
/* Matches OPAL_CRS_RESTART */
else if( ompi_cr_continue_like_restart && !first_continue_pass ) {
/*
* Barrier to make all processes have been successfully restarted before
* we try to remove some restart only files.
*/
if (OMPI_SUCCESS != (ret = orte_grpcomm.barrier())) {
opal_output(0, "bml:r2: ft_event(Restart): Failed in orte_grpcomm.barrier (%d)", ret);
return ret;
}
opal_output_verbose(10, ompi_cr_output,
"bml:r2: ft_event(Restart): Cleanup restart files\n");
opal_crs_base_cleanup_flush();
/*
* Re-open the BTL framework to get the full list of components.
*/
if( OMPI_SUCCESS != (ret = mca_btl_base_open()) ) {
opal_output(0, "bml:r2: ft_event(Restart): Failed to open BTL framework\n");
return ret;
}
/*
* Re-select the BTL components/modules
* This will cause the BTL components to discover the available
* network options on this machine, and post proper modex informaiton.
*/
if( OMPI_SUCCESS != (ret = mca_btl_base_select(OMPI_ENABLE_PROGRESS_THREADS,
OMPI_ENABLE_MPI_THREADS) ) ) {
opal_output(0, "bml:r2: ft_event(Restart): Failed to select in BTL framework\n");
return ret;
}
/*
* Clear some structures so we can properly repopulate them
*/
mca_bml_r2.btls_added = false;
for(p = 0; p < (int)num_procs; ++p) {
if( NULL != procs[p]->proc_bml) {
OBJ_RELEASE(procs[p]->proc_bml);
procs[p]->proc_bml = NULL;
}
OBJ_RELEASE(procs[p]);
}
if( NULL != procs ) {
free(procs);
procs = NULL;
}
}
}
else if(OPAL_CRS_RESTART_PRE == state ) {
opal_output_verbose(10, ompi_cr_output,
"bml:r2: ft_event(Restart): Finalize BML\n");
/*
* Finalize the BML
* - Flush progress functions
* - Flush module references
* - mca_btl_base_close()
* Need to do this because we may have BTL components that were
* unloaded in the first selection that may be available now.
* Conversely we may have BTL components loaded now that
* are not available now.
*/
if( OMPI_SUCCESS != (ret = mca_bml_r2_finalize()) ) {
opal_output(0, "bml:r2: ft_event(Restart): Failed to finalize BML framework\n");
return ret;
}
}
else if(OPAL_CRS_RESTART == state ) {
/*
* Barrier to make all processes have been successfully restarted before
* we try to remove some restart only files.
*/
if (OMPI_SUCCESS != (ret = orte_grpcomm.barrier())) {
opal_output(0, "bml:r2: ft_event(Restart): Failed in orte_grpcomm.barrier (%d)", ret);
return ret;
}
opal_output_verbose(10, ompi_cr_output,
"bml:r2: ft_event(Restart): Cleanup restart files\n");
opal_crs_base_cleanup_flush();
/*
* Re-open the BTL framework to get the full list of components.
* - but first clear the MCA value that was there
*/
param_type = mca_base_param_find("btl", NULL, NULL);
mca_base_param_lookup_string(param_type, ¶m_list);
opal_output_verbose(11, ompi_cr_output,
"Restart (Previous BTL MCA): <%s>\n", param_list);
if( NULL != param_list ) {
free(param_list);
param_list = NULL;
}
/* Deregister the old value, and refresh the file cache to grab any updates */
mca_base_param_deregister(param_type);
mca_base_param_recache_files(false);
if( OMPI_SUCCESS != (ret = mca_btl_base_open()) ) {
opal_output(0, "bml:r2: ft_event(Restart): Failed to open BTL framework\n");
return ret;
}
param_type = mca_base_param_find("btl", NULL, NULL);
mca_base_param_lookup_string(param_type, ¶m_list);
opal_output_verbose(11, ompi_cr_output,
"Restart (New BTL MCA): <%s>\n", param_list);
if( NULL != param_list ) {
free(param_list);
param_list = NULL;
}
/*
* Re-select the BTL components/modules
* This will cause the BTL components to discover the available
* network options on this machine, and post proper modex informaiton.
*/
if( OMPI_SUCCESS != (ret = mca_btl_base_select(OMPI_ENABLE_PROGRESS_THREADS,
OMPI_ENABLE_MPI_THREADS) ) ) {
opal_output(0, "bml:r2: ft_event(Restart): Failed to select in BTL framework\n");
return ret;
}
/*
* Clear some structures so we can properly repopulate them
*/
mca_bml_r2.btls_added = false;
for(p = 0; p < (int)num_procs; ++p) {
if( NULL != procs[p]->proc_bml) {
OBJ_RELEASE(procs[p]->proc_bml);
procs[p]->proc_bml = NULL;
}
OBJ_RELEASE(procs[p]);
}
if( NULL != procs ) {
free(procs);
procs = NULL;
}
}
else if(OPAL_CRS_TERM == state ) {
;
}
else {
;
}
return OMPI_SUCCESS;
}
