int opal_crs_self_checkpoint(pid_t pid,
opal_crs_base_snapshot_t *base_snapshot,
opal_crs_base_ckpt_options_t *options,
opal_crs_state_type_t *state)
{
opal_crs_self_snapshot_t *snapshot = OBJ_NEW(opal_crs_self_snapshot_t);
int ret, exit_status = OPAL_SUCCESS;
char * restart_cmd = NULL;
/*
* This function should never be called by a tool
*/
if( opal_cr_is_tool ) {
return OPAL_ERR_NOT_SUPPORTED;
}
if( options->stop ) {
opal_output(0,
"crs:self: checkpoint(): Error: SIGSTOP Not currently supported!");
}
/*
* Setup for snapshot directory creation
*/
snapshot->super = *base_snapshot;
#if 0
snapshot->super.snapshot_directory = strdup(base_snapshot->snapshot_directory);
snapshot->super.metadata_filename = strdup(base_snapshot->metadata_filename);
#endif
opal_output_verbose(10, mca_crs_self_component.super.output_handle,
"crs:self: checkpoint(%d, ---)", pid);
if(!mca_crs_self_component.can_checkpoint) {
opal_show_help("help-opal-crs-self.txt", "self:ckpt_disabled", false);
exit_status = OPAL_ERROR;
goto cleanup;
}
/*
* Update the snapshot metadata
*/
snapshot->super.component_name = strdup(mca_crs_self_component.super.base_version.mca_component_name);
if( NULL == snapshot->super.metadata ) {
if (NULL == (snapshot->super.metadata = fopen(snapshot->super.metadata_filename, "a")) ) {
opal_output(mca_crs_self_component.super.output_handle,
"crs:self: checkpoint(): Error: Unable to open the file (%s)",
snapshot->super.metadata_filename);
exit_status = OPAL_ERROR;
goto cleanup;
}
}
fprintf(snapshot->super.metadata, "%s%s\n", CRS_METADATA_COMP, snapshot->super.component_name);
/*
* Call the user callback function
*/
if(NULL != mca_crs_self_component.ucb_checkpoint_fn) {
mca_crs_self_component.ucb_checkpoint_fn(&restart_cmd);
}
/*
* Save the restart command
*/
if( NULL == restart_cmd) {
*state = OPAL_CRS_ERROR;
opal_show_help("help-opal-crs-self.txt", "self:no-restart-cmd",
true);
exit_status = OPAL_ERROR;
goto cleanup;
}
else {
snapshot->cmd_line = strdup(restart_cmd);
opal_output_verbose(10, mca_crs_self_component.super.output_handle,
"crs:self: checkpoint: Restart Command (%s)", snapshot->cmd_line);
}
/*
* The best we can do is update the metadata file with the
* application argv and argc we started with.
*/
if( OPAL_SUCCESS != (ret = self_update_snapshot_metadata(snapshot)) ) {
*state = OPAL_CRS_ERROR;
opal_output(mca_crs_self_component.super.output_handle,
"crs:self: checkpoint(): Error: Unable to update metadata for snapshot (%s).",
snapshot->super.metadata_filename);
exit_status = ret;
goto cleanup;
}
*state = OPAL_CRS_CONTINUE;
/*
* Call their continue routine for completeness
*/
if(NULL != mca_crs_self_component.ucb_continue_fn) {
mca_crs_self_component.ucb_continue_fn();
}
base_snapshot = &(snapshot->super);
cleanup:
if( NULL != restart_cmd) {
free(restart_cmd);
restart_cmd = NULL;
}
return exit_status;
}
int opal_crs_self_checkpoint(pid_t pid, opal_crs_base_snapshot_t *base_snapshot, opal_crs_state_type_t *state)
{
opal_crs_self_snapshot_t *snapshot = OBJ_NEW(opal_crs_self_snapshot_t);
int ret, exit_status = OPAL_SUCCESS;
char * restart_cmd = NULL;
/*
* This function should never be called by a tool
*/
if( opal_cr_is_tool ) {
return OPAL_ERR_NOT_SUPPORTED;
}
/*
* Setup for snapshot directory creation
*/
if(NULL != snapshot->super.reference_name)
free(snapshot->super.reference_name);
snapshot->super.reference_name = strdup(base_snapshot->reference_name);
if(NULL != snapshot->super.local_location)
free(snapshot->super.local_location);
snapshot->super.local_location = strdup(base_snapshot->local_location);
if(NULL != snapshot->super.remote_location)
free(snapshot->super.remote_location);
snapshot->super.remote_location = strdup(base_snapshot->remote_location);
opal_output_verbose(10, mca_crs_self_component.super.output_handle,
"crs:self: checkpoint(%d, ---)", pid);
if(!mca_crs_self_component.can_checkpoint) {
opal_show_help("help-opal-crs-self.txt", "self:ckpt_disabled", false);
exit_status = OPAL_ERROR;
goto cleanup;
}
/*
* Update the snapshot metadata
*/
snapshot->super.component_name = strdup(mca_crs_self_component.super.base_version.mca_component_name);
if( OPAL_SUCCESS != (ret = opal_crs_base_metadata_write_token(NULL, CRS_METADATA_COMP, snapshot->super.component_name) ) ) {
opal_output(mca_crs_self_component.super.output_handle,
"crs:self: checkpoint(): Error: Unable to write component name to the directory for (%s).",
snapshot->super.reference_name);
exit_status = ret;
goto cleanup;
}
/*
* Call the user callback function
*/
if(NULL != mca_crs_self_component.ucb_checkpoint_fn) {
mca_crs_self_component.ucb_checkpoint_fn(&restart_cmd);
}
/*
* Save the restart command
*/
if( NULL == restart_cmd) {
*state = OPAL_CRS_ERROR;
opal_show_help("help-opal-crs-self.txt", "self:no-restart-cmd",
true);
exit_status = OPAL_ERROR;
goto cleanup;
}
else {
snapshot->cmd_line = strdup(restart_cmd);
opal_output_verbose(10, mca_crs_self_component.super.output_handle,
"crs:self: checkpoint: Restart Command (%s)", snapshot->cmd_line);
}
/*
* The best we can do is update the metadata file with the
* application argv and argc we started with.
*/
if( OPAL_SUCCESS != (ret = self_update_snapshot_metadata(snapshot)) ) {
*state = OPAL_CRS_ERROR;
opal_output(mca_crs_self_component.super.output_handle,
"crs:self: checkpoint(): Error: Unable to update metadata for snapshot (%s).",
snapshot->super.reference_name);
exit_status = ret;
goto cleanup;
}
*state = OPAL_CRS_CONTINUE;
/*
* Call their continue routine for completeness
*/
if(NULL != mca_crs_self_component.ucb_continue_fn) {
mca_crs_self_component.ucb_continue_fn();
}
base_snapshot = &(snapshot->super);
cleanup:
if( NULL != restart_cmd) {
free(restart_cmd);
restart_cmd = NULL;
}
return exit_status;
}
