static void proc_errors(int fd, short args, void *cbdata)
{
orte_state_caddy_t *caddy = (orte_state_caddy_t*)cbdata;
orte_job_t *jdata;
orte_proc_t *pptr;
orte_process_name_t *proc = &caddy->name;
orte_proc_state_t state = caddy->proc_state;
orte_proc_t *child, *ptr;
opal_buffer_t *alert;
orte_plm_cmd_flag_t cmd;
int rc=ORTE_SUCCESS;
orte_vpid_t null=ORTE_VPID_INVALID;
orte_ns_cmp_bitmask_t mask=ORTE_NS_CMP_ALL;
int i;
/*
* if orte is trying to shutdown, just let it
*/
if (orte_finalizing) {
goto cleanup;
}
OPAL_OUTPUT_VERBOSE((2, orte_errmgr_base.output,
"%s errmgr:default_orted:proc_errors process %s error state %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(proc),
orte_proc_state_to_str(state)));
/* if this is a heartbeat failure, let the HNP handle it */
if (ORTE_PROC_STATE_HEARTBEAT_FAILED == state) {
goto cleanup;
}
/* if this was a failed comm, then see if it was to our
* lifeline
*/
if (ORTE_PROC_STATE_COMM_FAILED == state) {
/* if it is our own connection, ignore it */
if (OPAL_EQUAL == orte_util_compare_name_fields(ORTE_NS_CMP_ALL, ORTE_PROC_MY_NAME, proc)) {
goto cleanup;
}
/* was it a daemon? */
if (proc->jobid != ORTE_PROC_MY_NAME->jobid) {
/* nope - ignore */
goto cleanup;
}
OPAL_OUTPUT_VERBOSE((2, orte_errmgr_base.output,
"%s errmgr:default:orted daemon %s exited",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(proc)));
/* see if this was a lifeline */
if (ORTE_SUCCESS != orte_routed.route_lost(proc)) {
OPAL_OUTPUT_VERBOSE((2, orte_errmgr_base.output,
"%s errmgr:orted daemon %s was a lifeline - exiting",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(proc)));
/* kill our children */
killprocs(ORTE_JOBID_WILDCARD, ORTE_VPID_WILDCARD);
/* terminate - our routed children will see
* us leave and automatically die
*/
ORTE_TERMINATE(ORTE_ERROR_DEFAULT_EXIT_CODE);
goto cleanup;
}
/* are any of my children still alive */
for (i=0; i < orte_local_children->size; i++) {
if (NULL != (child = (orte_proc_t*)opal_pointer_array_get_item(orte_local_children, i))) {
if (child->alive && child->state < ORTE_PROC_STATE_UNTERMINATED) {
goto cleanup;
}
}
}
/* if all my routes and children are gone, then terminate
ourselves nicely (i.e., this is a normal termination) */
if (0 == orte_routed.num_routes()) {
OPAL_OUTPUT_VERBOSE((2, orte_errmgr_base.output,
"%s errmgr:default:orted all routes gone - exiting",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
ORTE_TERMINATE(0);
} else {
OPAL_OUTPUT_VERBOSE((2, orte_errmgr_base.output,
"%s errmgr:default:orted not exiting, num_routes() == %d",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
(int)orte_routed.num_routes()));
}
/* if not, then we can continue */
goto cleanup;
}
/* get the job object */
if (NULL == (jdata = orte_get_job_data_object(proc->jobid))) {
/* must already be complete */
goto cleanup;
}
pptr = (orte_proc_t*)opal_pointer_array_get_item(jdata->procs, proc->vpid);
/* if there are no local procs for this job, we can
* ignore this call
*/
if (0 == jdata->num_local_procs) {
goto cleanup;
}
/* find this proc in the local children */
child = NULL;
for (i=0; i < orte_local_children->size; i++) {
if (NULL == (ptr = (orte_proc_t*)opal_pointer_array_get_item(orte_local_children, i))) {
continue;
}
if (OPAL_EQUAL == orte_util_compare_name_fields(mask, &ptr->name, proc)) {
child = ptr;
break;
}
}
if (NULL == child) {
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
goto cleanup;
}
OPAL_OUTPUT_VERBOSE((2, orte_errmgr_base.output,
"%s errmgr:default_orted got state %s for proc %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
orte_proc_state_to_str(state),
ORTE_NAME_PRINT(proc)));
if (ORTE_PROC_STATE_SENSOR_BOUND_EXCEEDED == state) {
child->state = state;
/* Decrement the number of local procs */
jdata->num_local_procs--;
/* kill this proc */
killprocs(proc->jobid, proc->vpid);
goto cleanup;
}
if (ORTE_PROC_STATE_TERM_NON_ZERO == state) {
if (!orte_abort_non_zero_exit) {
/* leave the child in orte_local_children so we can
* later send the state info after full job termination
*/
child->state = state;
child->waitpid_recvd = true;
if (child->iof_complete) {
/* the proc has terminated */
child->alive = false;
/* Clean up the session directory as if we were the process
* itself. This covers the case where the process died abnormally
* and didn't cleanup its own session directory.
*/
orte_session_dir_finalize(&child->name);
/* track job status */
jdata->num_terminated++;
}
/* treat this as normal termination */
goto REPORT_STATE;
}
/* report this as abnormal termination to the HNP */
alert = OBJ_NEW(opal_buffer_t);
/* pack update state command */
cmd = ORTE_PLM_UPDATE_PROC_STATE;
if (ORTE_SUCCESS != (rc = opal_dss.pack(alert, &cmd, 1, ORTE_PLM_CMD))) {
ORTE_ERROR_LOG(rc);
return;
}
/* pack only the data for this proc - have to start with the jobid
* so the receiver can unpack it correctly
*/
if (ORTE_SUCCESS != (rc = opal_dss.pack(alert, &proc->jobid, 1, ORTE_JOBID))) {
ORTE_ERROR_LOG(rc);
return;
}
child->state = state;
/* now pack the child's info */
if (ORTE_SUCCESS != (rc = pack_state_for_proc(alert, child))) {
ORTE_ERROR_LOG(rc);
return;
}
/* remove the child from our local array as it is no longer alive */
opal_pointer_array_set_item(orte_local_children, i, NULL);
/* Decrement the number of local procs */
jdata->num_local_procs--;
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base.output,
"%s errmgr:default_orted reporting proc %s abnormally terminated with non-zero status (local procs = %d)",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&child->name),
jdata->num_local_procs));
/* release the child object */
OBJ_RELEASE(child);
/* send it */
if (0 > (rc = orte_rml.send_buffer_nb(ORTE_PROC_MY_HNP, alert,
ORTE_RML_TAG_PLM, 0,
orte_rml_send_callback, NULL))) {
ORTE_ERROR_LOG(rc);
OBJ_RELEASE(alert);
}
return;
}
if (ORTE_PROC_STATE_FAILED_TO_START == state ||
ORTE_PROC_STATE_FAILED_TO_LAUNCH == state) {
/* update the proc state */
child->state = state;
/* count the proc as having "terminated" */
jdata->num_terminated++;
/* leave the error report in this case to the
* state machine, which will receive notice
* when all local procs have attempted to start
* so that we send a consolidated error report
* back to the HNP
*/
goto cleanup;
}
if (ORTE_PROC_STATE_TERMINATED < state) {
/* if the job hasn't completed and the state is abnormally
* terminated, then we need to alert the HNP right away
*/
alert = OBJ_NEW(opal_buffer_t);
/* pack update state command */
cmd = ORTE_PLM_UPDATE_PROC_STATE;
if (ORTE_SUCCESS != (rc = opal_dss.pack(alert, &cmd, 1, ORTE_PLM_CMD))) {
ORTE_ERROR_LOG(rc);
return;
}
/* pack only the data for this proc - have to start with the jobid
* so the receiver can unpack it correctly
*/
if (ORTE_SUCCESS != (rc = opal_dss.pack(alert, &proc->jobid, 1, ORTE_JOBID))) {
ORTE_ERROR_LOG(rc);
return;
}
child->state = state;
/* now pack the child's info */
if (ORTE_SUCCESS != (rc = pack_state_for_proc(alert, child))) {
ORTE_ERROR_LOG(rc);
return;
}
/* remove the child from our local array as it is no longer alive */
opal_pointer_array_set_item(orte_local_children, i, NULL);
/* Decrement the number of local procs */
jdata->num_local_procs--;
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base.output,
"%s errmgr:default_orted reporting proc %s aborted to HNP (local procs = %d)",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&child->name),
jdata->num_local_procs));
/* release the child object */
OBJ_RELEASE(child);
/* send it */
if (0 > (rc = orte_rml.send_buffer_nb(ORTE_PROC_MY_HNP, alert,
ORTE_RML_TAG_PLM, 0,
orte_rml_send_callback, NULL))) {
ORTE_ERROR_LOG(rc);
}
return;
}
REPORT_STATE:
if (ORTE_PROC_STATE_REGISTERED == state) {
/* see if everyone in this job has registered */
if (all_children_registered(proc->jobid)) {
/* once everyone registers, send their contact info to
* the HNP so it is available to debuggers and anyone
* else that needs it
*/
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base.output,
"%s errmgr:default_orted: sending contact info to HNP",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
alert = OBJ_NEW(opal_buffer_t);
/* pack init routes command */
cmd = ORTE_PLM_INIT_ROUTES_CMD;
if (ORTE_SUCCESS != (rc = opal_dss.pack(alert, &cmd, 1, ORTE_PLM_CMD))) {
ORTE_ERROR_LOG(rc);
return;
}
/* pack the jobid */
if (ORTE_SUCCESS != (rc = opal_dss.pack(alert, &proc->jobid, 1, ORTE_JOBID))) {
ORTE_ERROR_LOG(rc);
return;
}
/* pack all the local child vpids */
for (i=0; i < orte_local_children->size; i++) {
if (NULL == (ptr = (orte_proc_t*)opal_pointer_array_get_item(orte_local_children, i))) {
continue;
}
if (ptr->name.jobid == proc->jobid) {
if (ORTE_SUCCESS != (rc = opal_dss.pack(alert, &ptr->name.vpid, 1, ORTE_VPID))) {
ORTE_ERROR_LOG(rc);
return;
}
}
}
/* pack an invalid marker */
if (ORTE_SUCCESS != (rc = opal_dss.pack(alert, &null, 1, ORTE_VPID))) {
ORTE_ERROR_LOG(rc);
return;
}
/* add in contact info for all procs in the job */
if (ORTE_SUCCESS != (rc = pack_child_contact_info(proc->jobid, alert))) {
ORTE_ERROR_LOG(rc);
OBJ_DESTRUCT(&alert);
return;
}
/* send it */
if (0 > (rc = orte_rml.send_buffer_nb(ORTE_PROC_MY_HNP, alert,
ORTE_RML_TAG_PLM, 0,
orte_rml_send_callback, NULL))) {
ORTE_ERROR_LOG(rc);
}
}
return;
}
/* only other state is terminated - see if anyone is left alive */
if (!any_live_children(proc->jobid)) {
alert = OBJ_NEW(opal_buffer_t);
/* pack update state command */
cmd = ORTE_PLM_UPDATE_PROC_STATE;
if (ORTE_SUCCESS != (rc = opal_dss.pack(alert, &cmd, 1, ORTE_PLM_CMD))) {
ORTE_ERROR_LOG(rc);
return;
}
/* pack the data for the job */
if (ORTE_SUCCESS != (rc = pack_state_update(alert, jdata))) {
ORTE_ERROR_LOG(rc);
return;
}
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base.output,
"%s errmgr:default_orted reporting all procs in %s terminated",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_JOBID_PRINT(jdata->jobid)));
/* remove all of this job's children from the global list - do not lock
* the thread as we are already locked
*/
for (i=0; i < orte_local_children->size; i++) {
if (NULL == (ptr = (orte_proc_t*)opal_pointer_array_get_item(orte_local_children, i))) {
continue;
}
if (jdata->jobid == ptr->name.jobid) {
opal_pointer_array_set_item(orte_local_children, i, NULL);
OBJ_RELEASE(ptr);
}
}
/* ensure the job's local session directory tree is removed */
orte_session_dir_cleanup(jdata->jobid);
/* remove this job from our local job data since it is complete */
opal_pointer_array_set_item(orte_job_data, ORTE_LOCAL_JOBID(jdata->jobid), NULL);
OBJ_RELEASE(jdata);
/* send it */
if (0 > (rc = orte_rml.send_buffer_nb(ORTE_PROC_MY_HNP, alert,
ORTE_RML_TAG_PLM, 0,
orte_rml_send_callback, NULL))) {
ORTE_ERROR_LOG(rc);
}
return;
}
cleanup:
OBJ_RELEASE(caddy);
}
static int update_state(orte_jobid_t job,
orte_job_state_t jobstate,
orte_process_name_t *proc,
orte_proc_state_t state,
pid_t pid,
orte_exit_code_t exit_code)
{
opal_list_item_t *item, *next;
orte_odls_job_t *jobdat = NULL;
orte_odls_child_t *child;
opal_buffer_t *alert;
orte_plm_cmd_flag_t cmd;
int rc=ORTE_SUCCESS;
orte_vpid_t null=ORTE_VPID_INVALID;
orte_ns_cmp_bitmask_t mask;
/*
* if orte is trying to shutdown, just let it
*/
if (orte_finalizing) {
return ORTE_SUCCESS;
}
OPAL_OUTPUT_VERBOSE((10, orte_errmgr_base.output,
"errmgr:default_orted:update_state() %s) "
"------- %s state updated for process %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
((NULL == proc) ? "App. Process" :
(proc->jobid == ORTE_PROC_MY_HNP->jobid ? "Daemon" : "App. Process")),
(NULL == proc) ? "NULL" : ORTE_NAME_PRINT(proc)));
/* if this is a heartbeat failure, let the HNP handle it */
if (ORTE_JOB_STATE_HEARTBEAT_FAILED == jobstate ||
ORTE_PROC_STATE_HEARTBEAT_FAILED == state) {
return ORTE_SUCCESS;
}
/*** UPDATE COMMAND FOR A JOB ***/
if (NULL == proc) {
/* this is an update for an entire job */
if (ORTE_JOBID_INVALID == job) {
/* whatever happened, we don't know what job
* it happened to
*/
orte_show_help("help-orte-errmgr.txt", "errmgr:unknown-job-error",
true, orte_job_state_to_str(jobstate));
alert = OBJ_NEW(opal_buffer_t);
/* pack update state command */
cmd = ORTE_PLM_UPDATE_PROC_STATE;
if (ORTE_SUCCESS != (rc = opal_dss.pack(alert, &cmd, 1, ORTE_PLM_CMD))) {
ORTE_ERROR_LOG(rc);
return rc;
}
/* pack the "invalid" jobid */
if (ORTE_SUCCESS != (rc = opal_dss.pack(alert, &job, 1, ORTE_JOBID))) {
ORTE_ERROR_LOG(rc);
return rc;
}
if (0 > (rc = orte_rml.send_buffer_nb(ORTE_PROC_MY_HNP, alert, ORTE_RML_TAG_PLM, 0, cbfunc, NULL))) {
ORTE_ERROR_LOG(rc);
} else {
rc = ORTE_SUCCESS;
}
return rc;
}
/* lookup the local jobdat for this job */
jobdat = NULL;
for (item = opal_list_get_first(&orte_local_jobdata);
item != opal_list_get_end(&orte_local_jobdata);
item = opal_list_get_next(item)) {
jobdat = (orte_odls_job_t*)item;
/* is this the specified job? */
if (jobdat->jobid == job) {
break;
}
}
if (NULL == jobdat) {
return ORTE_ERR_NOT_FOUND;
}
switch (jobstate) {
case ORTE_JOB_STATE_FAILED_TO_START:
failed_start(jobdat, exit_code);
break;
case ORTE_JOB_STATE_RUNNING:
/* update all local child states */
update_local_children(jobdat, jobstate, ORTE_PROC_STATE_RUNNING);
break;
case ORTE_JOB_STATE_SENSOR_BOUND_EXCEEDED:
/* update all procs in job */
update_local_children(jobdat, jobstate, ORTE_PROC_STATE_SENSOR_BOUND_EXCEEDED);
/* order all local procs for this job to be killed */
killprocs(jobdat->jobid, ORTE_VPID_WILDCARD);
case ORTE_JOB_STATE_COMM_FAILED:
/* kill all local procs */
killprocs(ORTE_JOBID_WILDCARD, ORTE_VPID_WILDCARD);
/* tell the caller we can't recover */
return ORTE_ERR_UNRECOVERABLE;
break;
case ORTE_JOB_STATE_HEARTBEAT_FAILED:
/* let the HNP handle this */
return ORTE_SUCCESS;
break;
default:
break;
}
alert = OBJ_NEW(opal_buffer_t);
/* pack update state command */
cmd = ORTE_PLM_UPDATE_PROC_STATE;
if (ORTE_SUCCESS != (rc = opal_dss.pack(alert, &cmd, 1, ORTE_PLM_CMD))) {
ORTE_ERROR_LOG(rc);
goto FINAL_CLEANUP;
}
/* pack the job info */
if (ORTE_SUCCESS != (rc = pack_state_update(alert, jobdat))) {
ORTE_ERROR_LOG(rc);
}
/* send it */
if (0 > (rc = orte_rml.send_buffer_nb(ORTE_PROC_MY_HNP, alert, ORTE_RML_TAG_PLM, 0, cbfunc, NULL))) {
ORTE_ERROR_LOG(rc);
} else {
rc = ORTE_SUCCESS;
}
return rc;
}
/* if this was a failed comm, then see if it was to our
* lifeline
*/
if (ORTE_PROC_STATE_COMM_FAILED == state) {
/* if it is our own connection, ignore it */
if (OPAL_EQUAL == orte_util_compare_name_fields(ORTE_NS_CMP_ALL, ORTE_PROC_MY_NAME, proc)) {
return ORTE_SUCCESS;
}
/* was it a daemon? */
if (proc->jobid != ORTE_PROC_MY_NAME->jobid) {
/* nope - ignore */
return ORTE_SUCCESS;
}
/* see if this was a lifeline */
if (ORTE_SUCCESS != orte_routed.route_lost(proc)) {
/* kill our children */
killprocs(ORTE_JOBID_WILDCARD, ORTE_VPID_WILDCARD);
/* terminate - our routed children will see
* us leave and automatically die
*/
orte_quit();
}
/* if not, then indicate we can continue */
return ORTE_SUCCESS;
}
/* lookup the local jobdat for this job */
jobdat = NULL;
for (item = opal_list_get_first(&orte_local_jobdata);
item != opal_list_get_end(&orte_local_jobdata);
item = opal_list_get_next(item)) {
jobdat = (orte_odls_job_t*)item;
/* is this the specified job? */
if (jobdat->jobid == proc->jobid) {
break;
}
}
if (NULL == jobdat) {
/* must already be complete */
return ORTE_SUCCESS;
}
/* if there are no local procs for this job, we can
* ignore this call
*/
if (0 == jobdat->num_local_procs) {
return ORTE_SUCCESS;
}
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base.output,
"%s errmgr:default_orted got state %s for proc %s pid %d",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
orte_proc_state_to_str(state),
ORTE_NAME_PRINT(proc), pid));
/*** UPDATE COMMAND FOR A SPECIFIC PROCESS ***/
if (ORTE_PROC_STATE_SENSOR_BOUND_EXCEEDED == state) {
/* find this proc in the local children */
for (item = opal_list_get_first(&orte_local_children);
item != opal_list_get_end(&orte_local_children);
item = opal_list_get_next(item)) {
child = (orte_odls_child_t*)item;
mask = ORTE_NS_CMP_ALL;
if (OPAL_EQUAL == orte_util_compare_name_fields(mask, child->name, proc)) {
if (ORTE_PROC_STATE_UNTERMINATED > child->state) {
child->state = state;
child->exit_code = exit_code;
/* Decrement the number of local procs */
jobdat->num_local_procs--;
/* kill this proc */
killprocs(proc->jobid, proc->vpid);
}
return ORTE_SUCCESS;
}
}
}
if (ORTE_PROC_STATE_TERM_NON_ZERO == state) {
if (!orte_abort_non_zero_exit) {
/* treat this as normal termination */
goto REPORT_STATE;
}
}
if (ORTE_PROC_STATE_TERMINATED < state) {
/* if the job hasn't completed and the state is abnormally
* terminated, then we need to alert the HNP right away
*/
alert = OBJ_NEW(opal_buffer_t);
/* pack update state command */
cmd = ORTE_PLM_UPDATE_PROC_STATE;
if (ORTE_SUCCESS != (rc = opal_dss.pack(alert, &cmd, 1, ORTE_PLM_CMD))) {
ORTE_ERROR_LOG(rc);
goto FINAL_CLEANUP;
}
/* pack only the data for this proc - have to start with the jobid
* so the receiver can unpack it correctly
*/
if (ORTE_SUCCESS != (rc = opal_dss.pack(alert, &proc->jobid, 1, ORTE_JOBID))) {
ORTE_ERROR_LOG(rc);
return rc;
}
/* find this proc in the local children */
for (item = opal_list_get_first(&orte_local_children);
item != opal_list_get_end(&orte_local_children);
item = opal_list_get_next(item)) {
child = (orte_odls_child_t*)item;
mask = ORTE_NS_CMP_ALL;
if (OPAL_EQUAL == orte_util_compare_name_fields(mask, child->name, proc)) {
if (ORTE_PROC_STATE_UNTERMINATED > child->state) {
child->state = state;
child->exit_code = exit_code;
}
/* now pack the child's info */
if (ORTE_SUCCESS != (rc = pack_state_for_proc(alert, child))) {
ORTE_ERROR_LOG(rc);
return rc;
}
/* remove the child from our local list as it is no longer alive */
opal_list_remove_item(&orte_local_children, &child->super);
/* Decrement the number of local procs */
jobdat->num_local_procs--;
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base.output,
"%s errmgr:default_orted reporting proc %s aborted to HNP (local procs = %d)",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(child->name),
jobdat->num_local_procs));
/* release the child object */
OBJ_RELEASE(child);
/* done with loop */
break;
}
}
/* send it */
if (0 > (rc = orte_rml.send_buffer_nb(ORTE_PROC_MY_HNP, alert, ORTE_RML_TAG_PLM, 0, cbfunc, NULL))) {
ORTE_ERROR_LOG(rc);
} else {
rc = ORTE_SUCCESS;
}
return rc;
}
REPORT_STATE:
/* find this proc in the local children so we can update its state */
for (item = opal_list_get_first(&orte_local_children);
item != opal_list_get_end(&orte_local_children);
item = opal_list_get_next(item)) {
child = (orte_odls_child_t*)item;
mask = ORTE_NS_CMP_ALL;
if (OPAL_EQUAL == orte_util_compare_name_fields(mask, child->name, proc)) {
if (ORTE_PROC_STATE_UNTERMINATED > child->state) {
child->state = state;
if (0 < pid) {
child->pid = pid;
}
child->exit_code = exit_code;
}
/* done with loop */
break;
}
}
if (ORTE_PROC_STATE_REGISTERED == state) {
/* see if everyone in this job has registered */
if (all_children_registered(proc->jobid)) {
/* once everyone registers, send their contact info to
* the HNP so it is available to debuggers and anyone
* else that needs it
*/
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base.output,
"%s errmgr:default_orted: sending contact info to HNP",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
alert = OBJ_NEW(opal_buffer_t);
/* pack init routes command */
cmd = ORTE_PLM_INIT_ROUTES_CMD;
if (ORTE_SUCCESS != (rc = opal_dss.pack(alert, &cmd, 1, ORTE_PLM_CMD))) {
ORTE_ERROR_LOG(rc);
goto FINAL_CLEANUP;
}
/* pack the jobid */
if (ORTE_SUCCESS != (rc = opal_dss.pack(alert, &proc->jobid, 1, ORTE_JOBID))) {
ORTE_ERROR_LOG(rc);
goto FINAL_CLEANUP;
}
/* pack all the local child vpids and epochs */
for (item = opal_list_get_first(&orte_local_children);
item != opal_list_get_end(&orte_local_children);
item = opal_list_get_next(item)) {
child = (orte_odls_child_t*)item;
if (child->name->jobid == proc->jobid) {
if (ORTE_SUCCESS != (rc = opal_dss.pack(alert, &child->name->vpid, 1, ORTE_VPID))) {
ORTE_ERROR_LOG(rc);
goto FINAL_CLEANUP;
}
}
}
/* pack an invalid marker */
if (ORTE_SUCCESS != (rc = opal_dss.pack(alert, &null, 1, ORTE_VPID))) {
ORTE_ERROR_LOG(rc);
goto FINAL_CLEANUP;
}
/* add in contact info for all procs in the job */
if (ORTE_SUCCESS != (rc = pack_child_contact_info(proc->jobid, alert))) {
ORTE_ERROR_LOG(rc);
OBJ_DESTRUCT(&alert);
return rc;
}
/* send it */
if (0 > (rc = orte_rml.send_buffer_nb(ORTE_PROC_MY_HNP, alert, ORTE_RML_TAG_PLM, 0, cbfunc, NULL))) {
ORTE_ERROR_LOG(rc);
} else {
rc = ORTE_SUCCESS;
}
}
return rc;
}
/* only other state is terminated - see if anyone is left alive */
if (!any_live_children(proc->jobid)) {
/* lookup the local jobdat for this job */
jobdat = NULL;
for (item = opal_list_get_first(&orte_local_jobdata);
item != opal_list_get_end(&orte_local_jobdata);
item = opal_list_get_next(item)) {
jobdat = (orte_odls_job_t*)item;
/* is this the specified job? */
if (jobdat->jobid == proc->jobid) {
break;
}
}
if (NULL == jobdat) {
/* race condition - may not have been formed yet */
return ORTE_SUCCESS;
}
alert = OBJ_NEW(opal_buffer_t);
/* pack update state command */
cmd = ORTE_PLM_UPDATE_PROC_STATE;
if (ORTE_SUCCESS != (rc = opal_dss.pack(alert, &cmd, 1, ORTE_PLM_CMD))) {
ORTE_ERROR_LOG(rc);
goto FINAL_CLEANUP;
}
/* pack the data for the job */
if (ORTE_SUCCESS != (rc = pack_state_update(alert, jobdat))) {
ORTE_ERROR_LOG(rc);
}
FINAL_CLEANUP:
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base.output,
"%s errmgr:default_orted reporting all procs in %s terminated",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_JOBID_PRINT(jobdat->jobid)));
/* remove all of this job's children from the global list - do not lock
* the thread as we are already locked
*/
for (item = opal_list_get_first(&orte_local_children);
item != opal_list_get_end(&orte_local_children);
item = next) {
child = (orte_odls_child_t*)item;
next = opal_list_get_next(item);
if (jobdat->jobid == child->name->jobid) {
opal_list_remove_item(&orte_local_children, &child->super);
OBJ_RELEASE(child);
}
}
/* ensure the job's local session directory tree is removed */
orte_session_dir_cleanup(jobdat->jobid);
/* remove this job from our local job data since it is complete */
opal_list_remove_item(&orte_local_jobdata, &jobdat->super);
OBJ_RELEASE(jobdat);
/* send it */
if (0 > (rc = orte_rml.send_buffer_nb(ORTE_PROC_MY_HNP, alert, ORTE_RML_TAG_PLM, 0, cbfunc, NULL))) {
ORTE_ERROR_LOG(rc);
} else {
rc = ORTE_SUCCESS;
}
/* indicate that the job is complete */
return rc;
}
return ORTE_SUCCESS;
}
