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;
}
int orte_sstore_central_app_set_attr(orte_sstore_base_handle_t handle, orte_sstore_base_key_t key, char *value)
{
int ret, exit_status = ORTE_SUCCESS;
orte_sstore_central_app_snapshot_info_t *handle_info = NULL;
char *key_str = NULL;
OPAL_OUTPUT_VERBOSE((10, mca_sstore_central_component.super.output_handle,
"sstore:central:(app): set_attr(%d = %s)", key, value));
if( NULL == value ) {
ORTE_ERROR_LOG(ORTE_ERROR);
exit_status = ORTE_ERROR;
goto cleanup;
}
if( key >= SSTORE_METADATA_MAX ) {
ORTE_ERROR_LOG(ORTE_ERROR);
exit_status = ORTE_ERROR;
goto cleanup;
}
/*
* Lookup the handle
*/
handle_info = find_handle_info(handle);
/*
* Access metadata
*/
if( SSTORE_METADATA_LOCAL_CRS_COMP == key ) {
if( NULL != handle_info->crs_comp ) {
free(handle_info->crs_comp);
}
handle_info->crs_comp = strdup(value);
}
else if(SSTORE_METADATA_LOCAL_SKIP_CKPT == key ) {
handle_info->ckpt_skipped = true;
}
else if( SSTORE_METADATA_LOCAL_MKDIR == key ||
SSTORE_METADATA_LOCAL_TOUCH == key ) {
orte_sstore_base_convert_key_to_string(key, &key_str);
if( ORTE_SUCCESS != (ret = metadata_write_str(handle_info, key_str, value))) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
}
else {
exit_status = ORTE_ERROR;
goto cleanup;
}
cleanup:
if( NULL != key_str ) {
free(key_str);
key_str = NULL;
}
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_global_register(orte_sstore_base_handle_t handle)
{
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): register(%d) - Global", handle));
/*
* Lookup the handle
*/
handle_info = find_handle_info(handle);
if( SSTORE_GLOBAL_REG != handle_info->state ) {
handle_info->state = SSTORE_GLOBAL_REG;
} else {
return orte_sstore_central_local_register(handle);
}
orte_sstore_handle_current = handle;
/*
* Associate the metadata
*/
if( handle_info->migrating ) {
if( ORTE_SUCCESS != (ret = metadata_write_int(handle_info,
SSTORE_METADATA_INTERNAL_MIG_SEQ_STR,
handle_info->seq_num)) ) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
} else {
if( ORTE_SUCCESS != (ret = metadata_write_int(handle_info,
SSTORE_METADATA_GLOBAL_SNAP_SEQ_STR,
handle_info->seq_num)) ) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
}
if( ORTE_SUCCESS != (ret = metadata_write_str(handle_info,
SSTORE_METADATA_LOCAL_SNAP_REF_FMT_STR,
orte_sstore_base_local_snapshot_fmt)) ) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
if( ORTE_SUCCESS != (ret = metadata_write_timestamp(handle_info)) ) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
cleanup:
return exit_status;
}
int orte_sstore_central_global_pack(orte_process_name_t* peer, opal_buffer_t* buffer, orte_sstore_base_handle_t handle)
{
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): pack()"));
/*
* Lookup the handle
*/
handle_info = find_handle_info(handle);
/*
* Pack the handle ID
*/
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &handle, 1, ORTE_SSTORE_HANDLE))) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
OPAL_OUTPUT_VERBOSE((10, mca_sstore_central_component.super.output_handle,
"sstore:central:(global): pack(%d, %d, %s)",
handle_info->id,
handle_info->seq_num,
handle_info->ref_name));
/*
* Pack any metadata
*/
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &(handle_info->seq_num), 1, OPAL_INT )) ) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &(handle_info->ref_name), 1, OPAL_STRING )) ) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &(handle_info->app_location_fmt), 1, OPAL_STRING )) ) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
cleanup:
return exit_status;
}
int orte_sstore_central_global_get_attr(orte_sstore_base_handle_t handle, orte_sstore_base_key_t key, char **value)
{
int 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): get_attr()"));
/*
* Lookup the handle
*/
handle_info = find_handle_info(handle);
/*
* Access metadata
*/
if( SSTORE_METADATA_GLOBAL_SNAP_REF == key ) {
*value = strdup(handle_info->ref_name);
}
else if( SSTORE_METADATA_GLOBAL_SNAP_SEQ == key ) {
asprintf(value, "%d", handle_info->seq_num);
}
else if( SSTORE_METADATA_LOCAL_SNAP_REF_FMT == key ) {
*value = strdup(orte_sstore_base_local_snapshot_fmt);
}
/* 'central' does not cache, so these are the same */
else if( SSTORE_METADATA_LOCAL_SNAP_LOC == key ) {
asprintf(value, "%s/%s/%d",
handle_info->base_location,
handle_info->ref_name,
handle_info->seq_num);
}
else if( SSTORE_METADATA_LOCAL_SNAP_REF_LOC_FMT == key ) {
asprintf(value, "%s/%s/%d/%s",
handle_info->base_location,
handle_info->ref_name,
handle_info->seq_num,
orte_sstore_base_local_snapshot_fmt);
}
else {
exit_status = ORTE_ERR_NOT_SUPPORTED;
}
return exit_status;
}
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_global_set_attr(orte_sstore_base_handle_t handle, orte_sstore_base_key_t key, char *value)
{
int ret, exit_status = ORTE_SUCCESS;
orte_sstore_central_global_snapshot_info_t *handle_info = NULL;
char *key_str = NULL;
OPAL_OUTPUT_VERBOSE((10, mca_sstore_central_component.super.output_handle,
"sstore:central:(global): set_attr()"));
/*
* Lookup the handle
*/
handle_info = find_handle_info(handle);
/*
* Process key (Access metadata)
*/
if( key == SSTORE_METADATA_GLOBAL_MIGRATING ) {
handle_info->migrating = true;
}
else {
orte_sstore_base_convert_key_to_string(key, &key_str);
if( NULL == key_str ) {
ORTE_ERROR_LOG(ORTE_ERROR);
exit_status = ORTE_ERROR;
goto cleanup;
}
if( ORTE_SUCCESS != (ret = metadata_write_str(handle_info, key_str, value))) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
}
cleanup:
if( NULL != key_str ) {
free(key_str);
key_str = NULL;
}
return exit_status;
}
int orte_sstore_central_app_sync(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): sync()"));
/*
* Lookup the handle
*/
handle_info = find_handle_info(handle);
/*
* Finalize and close the metadata
*/
if( ORTE_SUCCESS != (ret = metadata_write_timestamp(handle_info)) ) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
if( ORTE_SUCCESS != (ret = metadata_close(handle_info)) ) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
/*
* Push information to the Local coordinator
*/
if( ORTE_SUCCESS != (ret = push_handle_info(handle_info)) ) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
cleanup:
orte_sstore_handle_current = ORTE_SSTORE_HANDLE_INVALID;
return exit_status;
}
int orte_sstore_central_app_get_attr(orte_sstore_base_handle_t handle, orte_sstore_base_key_t key, char **value)
{
int 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): get_attr(%d)", key));
/*
* Lookup the handle
*/
handle_info = find_handle_info(handle);
/*
* Access metadata
*/
if( SSTORE_METADATA_GLOBAL_SNAP_SEQ == key ) {
asprintf(value, "%d", handle_info->seq_num);
OPAL_OUTPUT_VERBOSE((10, mca_sstore_central_component.super.output_handle,
"sstore:central:(app): get_attr(%d, %d) Seq = <%s>", key, handle_info->id, *value));
}
else if( SSTORE_METADATA_LOCAL_SNAP_LOC == key) {
*value = strdup(handle_info->local_location);
}
else if( SSTORE_METADATA_LOCAL_SNAP_META == key ) {
*value = strdup(handle_info->metadata_filename);
}
else if( SSTORE_METADATA_GLOBAL_SNAP_REF == key ) {
*value = strdup(handle_info->global_ref_name);
OPAL_OUTPUT_VERBOSE((10, mca_sstore_central_component.super.output_handle,
"sstore:central:(app): get_attr(%d, %d) Ref = <%s>", key, handle_info->id, *value));
}
else {
exit_status = ORTE_ERR_NOT_SUPPORTED;
goto cleanup;
}
OPAL_OUTPUT_VERBOSE((10, mca_sstore_central_component.super.output_handle,
"sstore:central:(app): get_attr(%d, %d) <%s>", key, handle_info->id, *value));
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;
}
int orte_sstore_central_local_sync(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): sync()"));
/*
* Lookup the handle
*/
handle_info = find_handle_info(handle);
/*
* Wait for all of the applications to update their metadata
*/
if( ORTE_SUCCESS != (ret = wait_all_apps_updated(handle_info))) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
/*
* Push information to the Global coordinator
*/
if( ORTE_SUCCESS != (ret = push_handle_info(handle_info)) ) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
handle_info->status = SSTORE_LOCAL_SYNCED;
cleanup:
return exit_status;
}
static void sstore_central_global_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_global_snapshot_info_t *handle_info = NULL;
if( ORTE_RML_TAG_SSTORE_INTERNAL != tag ) {
return;
}
/*
* If this was an application process contacting us, then act like an orted
* instead of an HNP
*/
if(OPAL_EQUAL != orte_util_compare_name_fields(ORTE_NS_CMP_JOBID,
ORTE_PROC_MY_NAME,
sender)) {
orte_sstore_central_local_recv(status, sender, buffer, tag, cbdata);
return;
}
OPAL_OUTPUT_VERBOSE((10, mca_sstore_central_component.super.output_handle,
"sstore:central:(global): 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
*/
if(NULL == (handle_info = find_handle_info(loc_id)) ) {
; /* JJH big problem */
}
/*
* Process the command
*/
if( ORTE_SSTORE_CENTRAL_PULL == command ) {
process_local_pull(sender, buffer, handle_info);
}
else if( ORTE_SSTORE_CENTRAL_PUSH == command ) {
process_local_push(sender, buffer, handle_info);
}
cleanup:
return;
}
int orte_sstore_central_global_sync(orte_sstore_base_handle_t handle)
{
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): sync()"));
/*
* Lookup the handle
*/
handle_info = find_handle_info(handle);
if( SSTORE_GLOBAL_SYNCING != handle_info->state ) {
handle_info->state = SSTORE_GLOBAL_SYNCING;
if( ORTE_SNAPC_LOCAL_COORD_TYPE == (orte_snapc_coord_type & ORTE_SNAPC_LOCAL_COORD_TYPE) ) {
return orte_sstore_central_local_sync(handle);
}
}
/*
* Synchronize all of the files
*/
while(handle_info->num_procs_synced < handle_info->num_procs_total) {
opal_progress();
}
/*
* Finalize and close the metadata
*/
if( ORTE_SUCCESS != (ret = metadata_write_timestamp(handle_info)) ) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
if( handle_info->migrating ) {
if( ORTE_SUCCESS != (ret = metadata_write_int(handle_info,
SSTORE_METADATA_INTERNAL_DONE_MIG_SEQ_STR,
handle_info->seq_num)) ) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
} else {
if( ORTE_SUCCESS != (ret = metadata_write_int(handle_info,
SSTORE_METADATA_INTERNAL_DONE_SEQ_STR,
handle_info->seq_num)) ) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
}
if( ORTE_SUCCESS != (ret = metadata_close(handle_info)) ) {
ORTE_ERROR_LOG(ret);
exit_status = ret;
goto cleanup;
}
/* JJH: We should lock this var! */
if( !handle_info->migrating ) {
orte_sstore_base_is_checkpoint_available = true;
orte_sstore_handle_last_stable = orte_sstore_handle_current;
}
cleanup:
return exit_status;
}