/*
* get_step_image_dir - get the dir to store step task images
* IN cr: checkpoint/restart
* RET image dir on success, NULL on error
*
* NOTE: only can be called in callbak
*/
static char *
get_step_image_dir(int cr)
{
const struct cr_checkpoint_info *ckpt_info;
const struct cr_restart_info *rstrt_info;
const char *dest;
char *rchar, *dir;
if (cr) { /* checkpoint */
ckpt_info = cr_get_checkpoint_info();
if (!ckpt_info) {
error("failed to get checkpoint info: %s",
cr_strerror(errno));
return NULL;
}
dest = ckpt_info->dest;
} else { /* retart */
rstrt_info = cr_get_restart_info();
if (!rstrt_info) {
error("failed to get restart info: %s",
cr_strerror(errno));
return NULL;
}
dest = rstrt_info->src;
}
rchar = strrchr(dest, '/');
if (rchar) {
dir = xstrndup(dest, rchar - dest + 1);
}
xstrfmtcat(dir, "%u.%u", jobid, stepid);
return dir;
}
static int opal_crs_blcr_thread_callback(void *arg) {
const struct cr_checkpoint_info *ckpt_info = cr_get_checkpoint_info();
int ret;
opal_output_verbose(10, mca_crs_blcr_component.super.output_handle,
"crs:blcr: thread_callback()");
OPAL_THREAD_LOCK(&blcr_lock);
blcr_current_state = OPAL_CRS_CHECKPOINT;
/*
* Allow the checkpoint to be taken, if we requested it
*/
#if CRS_BLCR_HAVE_INFO_REQUESTER == 1
if( ckpt_info->requester != my_pid ) {
ret = cr_checkpoint(CR_CHECKPOINT_OMIT);
blcr_current_state = OPAL_CRS_RUNNING;
opal_output_verbose(10, mca_crs_blcr_component.super.output_handle,
"crs:blcr: thread_callback(); WARNING: An external agent attempted to checkpoint this process "
"when it did not expect to be checkpointed. Skipping this checkpoint request."
" [%d != %d].", ckpt_info->requester, my_pid);
return 0;
}
else
#endif
{
ret = cr_checkpoint(0);
}
/*
* Restarting
*/
if ( 0 < ret ) {
opal_output_verbose(10, mca_crs_blcr_component.super.output_handle,
"crs:blcr: thread_callback: Restarting.");
blcr_current_state = OPAL_CRS_RESTART;
}
/*
* Continuing
*/
else {
opal_output_verbose(10, mca_crs_blcr_component.super.output_handle,
"crs:blcr: thread_callback: Continue.");
blcr_current_state = OPAL_CRS_CONTINUE;
}
OPAL_THREAD_UNLOCK(&blcr_lock);
opal_condition_signal(&blcr_cond);
return 0;
}
static int opal_crs_blcr_signal_callback(void *arg) {
const struct cr_checkpoint_info *ckpt_info = cr_get_checkpoint_info();
int ret;
/*
* Allow the checkpoint to be taken, if we requested it
*/
#if CRS_BLCR_HAVE_INFO_REQUESTER == 1
if( ckpt_info->requester != my_pid ) {
ret = cr_checkpoint(CR_CHECKPOINT_OMIT);
return 0;
}
else
#endif
{
ret = cr_checkpoint(0);
}
return 0;
}
