/**
* Terminate the orteds for a given job
*/
static int plm_alps_terminate_orteds(void)
{
int rc;
/* deregister the waitpid callback to ensure we don't make it look like
* alps failed when it didn't. Since the alps may have already completed,
* do NOT ERROR_LOG any return code to avoid confusing, duplicate error
* messages
*/
orte_wait_cb_cancel(alps_pid);
/* now tell them to die */
if (orte_abnormal_term_ordered) {
/* cannot know if a daemon is able to
* tell us it died, so just ensure they
* all terminate
*/
if (ORTE_SUCCESS != (rc = orte_plm_base_orted_exit(ORTE_DAEMON_HALT_VM_CMD))) {
ORTE_ERROR_LOG(rc);
}
} else {
/* we need them to "phone home", though,
* so we can know that they have exited
*/
if (ORTE_SUCCESS != (rc = orte_plm_base_orted_exit(ORTE_DAEMON_EXIT_CMD))) {
ORTE_ERROR_LOG(rc);
}
}
return rc;
}
/**
* Terminate the orteds for a given job
*/
static int plm_slurm_terminate_orteds(void)
{
int rc=ORTE_SUCCESS;
orte_job_t *jdata;
/* check to see if the primary pid is set. If not, this indicates
* that we never launched any additional daemons, so we cannot
* not wait for a waitpid to fire and tell us it's okay to
* exit. Instead, we simply trigger an exit for ourselves
*/
if (primary_pid_set) {
if (ORTE_SUCCESS != (rc = orte_plm_base_orted_exit(ORTE_DAEMON_EXIT_CMD))) {
ORTE_ERROR_LOG(rc);
}
} else {
OPAL_OUTPUT_VERBOSE((1, orte_plm_base_framework.framework_output,
"%s plm:slurm: primary daemons complete!",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
jdata = orte_get_job_data_object(ORTE_PROC_MY_NAME->jobid);
/* need to set the #terminated value to avoid an incorrect error msg */
jdata->num_terminated = jdata->num_procs;
ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_DAEMONS_TERMINATED);
}
return rc;
}
/**
* Terminate the orteds for a given job
*/
static int plm_slurm_terminate_orteds(void)
{
int rc;
orte_job_t *jdata;
/* tell them to die without sending a reply - we will rely on the
* waitpid to tell us when they have exited!
*/
if (ORTE_SUCCESS != (rc = orte_plm_base_orted_exit(ORTE_DAEMON_EXIT_NO_REPLY_CMD))) {
ORTE_ERROR_LOG(rc);
}
/* check to see if the primary pid is set. If not, this indicates
* that we never launched any additional daemons, so we cannot
* not wait for a waitpid to fire and tell us it's okay to
* exit. Instead, we simply trigger an exit for ourselves
*/
if (!primary_pid_set) {
OPAL_OUTPUT_VERBOSE((1, orte_plm_globals.output,
"%s plm:slurm: primary daemons complete!",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
jdata = orte_get_job_data_object(ORTE_PROC_MY_NAME->jobid);
jdata->state = ORTE_JOB_STATE_TERMINATED;
/* need to set the #terminated value to avoid an incorrect error msg */
jdata->num_terminated = jdata->num_procs;
orte_trigger_event(&orteds_exit);
}
return rc;
}
/**
* Terminate the orteds for a given job
*/
static int plm_alps_terminate_orteds(void)
{
int rc;
orte_job_t *jdata;
OPAL_OUTPUT_VERBOSE((10, orte_plm_base_framework.framework_output,
"%s plm:alps: terminating orteds",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
/* deregister the waitpid callback to ensure we don't make it look like
* alps failed when it didn't. Since the alps may have already completed,
* do NOT ERROR_LOG any return code to avoid confusing, duplicate error
* messages
*/
if (NULL != alpsrun) {
orte_wait_cb_cancel(alpsrun);
}
/* now tell them to die */
if (ORTE_SUCCESS != (rc = orte_plm_base_orted_exit(ORTE_DAEMON_EXIT_CMD))) {
ORTE_ERROR_LOG(rc);
}
jdata = orte_get_job_data_object(ORTE_PROC_MY_NAME->jobid);
ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_DAEMONS_TERMINATED);
OPAL_OUTPUT_VERBOSE((10, orte_plm_base_framework.framework_output,
"%s plm:alps: terminated orteds",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
return rc;
}
/**
* Terminate the orteds for a given job
*/
static int plm_lsf_terminate_orteds(void)
{
int rc;
if (ORTE_SUCCESS != (rc = orte_plm_base_orted_exit(ORTE_DAEMON_EXIT_CMD))) {
ORTE_ERROR_LOG(rc);
}
return rc;
}
static int isolated_terminate_orteds(void)
{
int rc;
/* send ourselves the halt command */
if (ORTE_SUCCESS != (rc = orte_plm_base_orted_exit(ORTE_DAEMON_EXIT_CMD))) {
ORTE_ERROR_LOG(rc);
}
return rc;
}
/**
* Terminate the orteds for a given job
*/
int plm_tm_terminate_orteds(void)
{
int rc;
/* now tell them to die */
if (orte_abnormal_term_ordered) {
/* cannot know if a daemon is able to
* tell us it died, so just ensure they
* all terminate
*/
if (ORTE_SUCCESS != (rc = orte_plm_base_orted_exit(ORTE_DAEMON_HALT_VM_CMD))) {
ORTE_ERROR_LOG(rc);
}
} else {
/* we need them to "phone home", though,
* so we can know that they have exited
*/
if (ORTE_SUCCESS != (rc = orte_plm_base_orted_exit(ORTE_DAEMON_EXIT_CMD))) {
ORTE_ERROR_LOG(rc);
}
}
return rc;
}
/**
* Terminate the orteds for a given job
*/
static int plm_alps_terminate_orteds(void)
{
int rc;
/* deregister the waitpid callback to ensure we don't make it look like
* alps failed when it didn't. Since the alps may have already completed,
* do NOT ERROR_LOG any return code to avoid confusing, duplicate error
* messages
*/
if (NULL != alpsrun) {
orte_wait_cb_cancel(alpsrun);
}
/* now tell them to die */
if (ORTE_SUCCESS != (rc = orte_plm_base_orted_exit(ORTE_DAEMON_EXIT_CMD))) {
ORTE_ERROR_LOG(rc);
}
return rc;
}
