int orte_sstore_central_global_request_checkpoint_handle(orte_sstore_base_handle_t *handle, int seq, orte_jobid_t jobid)
{
int ret, exit_status = ORTE_SUCCESS;
orte_sstore_central_global_snapshot_info_t *handle_info = NULL;
OPAL_OUTPUT_VERBOSE((10, mca_sstore_central_component.super.output_handle,
"sstore:central:(global): request_checkpoint_handle()"));
/*
* Construct a handle
* - Associate all of the necessary information
*/
handle_info = create_new_handle_info(seq, SSTORE_HANDLE_TYPE_CKPT, jobid);
/*
* Create the global checkpoint directory
*/
if( ORTE_SUCCESS != (ret = init_global_snapshot_directory(handle_info)) ) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
/*
* Return the handle
*/
*handle = handle_info->id;
cleanup:
return exit_status;
}
int orte_sstore_central_local_register(orte_sstore_base_handle_t handle)
{
int ret, exit_status = ORTE_SUCCESS;
orte_sstore_central_local_snapshot_info_t *handle_info = NULL;
OPAL_OUTPUT_VERBOSE((10, mca_sstore_central_component.super.output_handle,
"sstore:central:(local): register()"));
/*
* Create a handle
*/
if( NULL == (handle_info = find_handle_info(handle)) ) {
handle_info = create_new_handle_info(handle);
}
/*
* Get basic information from Global SStore
*/
if( ORTE_SUCCESS != (ret = pull_handle_info(handle_info)) ) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
/*
* Wait here until the pull request has been satisfied
*/
while(SSTORE_LOCAL_READY != handle_info->status &&
SSTORE_LOCAL_ERROR != handle_info->status ) {
opal_progress();
}
cleanup:
return exit_status;
}
void orte_sstore_central_local_recv(int status,
orte_process_name_t* sender,
opal_buffer_t* buffer,
orte_rml_tag_t tag,
void* cbdata)
{
int ret;
orte_sstore_central_cmd_flag_t command;
orte_std_cntr_t count;
orte_sstore_base_handle_t loc_id;
orte_sstore_central_local_snapshot_info_t *handle_info = NULL;
if( ORTE_RML_TAG_SSTORE_INTERNAL != tag ) {
return;
}
OPAL_OUTPUT_VERBOSE((10, mca_sstore_central_component.super.output_handle,
"sstore:central:(local): process_cmd(%s)",
ORTE_NAME_PRINT(sender)));
count = 1;
if (ORTE_SUCCESS != (ret = opal_dss.unpack(buffer, &command, &count, ORTE_SSTORE_CENTRAL_CMD))) {
ORTE_ERROR_LOG(ret);
goto cleanup;
}
count = 1;
if (ORTE_SUCCESS != (ret = opal_dss.unpack(buffer, &loc_id, &count, ORTE_SSTORE_HANDLE )) ) {
ORTE_ERROR_LOG(ret);
goto cleanup;
}
/*
* Find the referenced handle (Create if it does not exist)
*/
if(NULL == (handle_info = find_handle_info(loc_id)) ) {
handle_info = create_new_handle_info(loc_id);
}
/*
* Process the command
*/
if( ORTE_SSTORE_CENTRAL_PULL == command ) {
if(OPAL_EQUAL == orte_util_compare_name_fields(ORTE_NS_CMP_ALL, ORTE_PROC_MY_HNP, sender)) {
process_global_pull(sender, buffer, handle_info);
} else {
process_app_pull(sender, buffer, handle_info);
}
}
else if( ORTE_SSTORE_CENTRAL_PUSH == command ) {
if(OPAL_EQUAL == orte_util_compare_name_fields(ORTE_NS_CMP_ALL, ORTE_PROC_MY_HNP, sender)) {
process_global_push(sender, buffer, handle_info);
} else {
process_app_push(sender, buffer, handle_info);
}
}
cleanup:
return;
}
int orte_sstore_central_local_unpack(orte_process_name_t* peer, opal_buffer_t* buffer, orte_sstore_base_handle_t *handle)
{
int ret, exit_status = ORTE_SUCCESS;
orte_sstore_central_local_snapshot_info_t *handle_info = NULL;
orte_std_cntr_t count;
OPAL_OUTPUT_VERBOSE((10, mca_sstore_central_component.super.output_handle,
"sstore:central:(local): unpack()"));
/*
* Unpack the handle id
*/
count = 1;
if (ORTE_SUCCESS != (ret = opal_dss.unpack(buffer, handle, &count, ORTE_SSTORE_HANDLE))) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
/*
* Lookup the handle
*/
if( NULL == (handle_info = find_handle_info(*handle)) ) {
handle_info = create_new_handle_info(*handle);
}
/*
* Unpack the metadata piggybacked on this message
*/
if( ORTE_SUCCESS != (ret = process_global_push(peer, buffer, handle_info))) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
OPAL_OUTPUT_VERBOSE((10, mca_sstore_central_component.super.output_handle,
"sstore:central:(local): unpack(%d, %d, %s)",
handle_info->id,
handle_info->seq_num,
handle_info->global_ref_name));
cleanup:
return exit_status;
}
int orte_sstore_central_app_register(orte_sstore_base_handle_t handle)
{
int ret, exit_status = ORTE_SUCCESS;
orte_sstore_central_app_snapshot_info_t *handle_info = NULL;
OPAL_OUTPUT_VERBOSE((10, mca_sstore_central_component.super.output_handle,
"sstore:central:(app): register(%d)", (int)handle));
/*
* Create a handle
*/
orte_sstore_handle_current = handle;
handle_info = find_handle_info(handle);
if( NULL != handle_info ) {
/* Remove the old, stale handle */
opal_list_remove_item(active_handles, &(handle_info->super));
}
handle_info = create_new_handle_info(handle);
/*
* Get basic information from Local SStore
*/
if( ORTE_SUCCESS != (ret = pull_handle_info(handle_info)) ) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
/*
* Setup the storage directory
*/
if( ORTE_SUCCESS != (ret = init_local_snapshot_directory(handle_info)) ) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
cleanup:
return exit_status;
}
