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, *proct;
orte_process_name_t *proc = &caddy->name;
orte_proc_state_t state = caddy->proc_state;
int i;
int32_t i32, *i32ptr;
OPAL_OUTPUT_VERBOSE((1, orte_errmgr_base_framework.framework_output,
"%s errmgr:default_hnp: for proc %s state %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(proc),
orte_proc_state_to_str(state)));
/*
* if orte is trying to shutdown, just let it
*/
if (orte_finalizing) {
goto cleanup;
}
/* get the job object */
if (NULL == (jdata = orte_get_job_data_object(proc->jobid))) {
/* could be a race condition */
goto cleanup;
}
pptr = (orte_proc_t*)opal_pointer_array_get_item(jdata->procs, proc->vpid);
/* we MUST handle a communication failure before doing anything else
* as it requires some special care to avoid normal termination issues
* for local application procs
*/
if (ORTE_PROC_STATE_COMM_FAILED == state) {
/* is this to a daemon? */
if (ORTE_PROC_MY_NAME->jobid != proc->jobid) {
/* nope - ignore it */
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base_framework.framework_output,
"%s Comm failure to non-daemon proc - ignoring it",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
goto cleanup;
}
/* if this is my own connection, ignore it */
if (ORTE_PROC_MY_NAME->vpid == proc->vpid) {
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base_framework.framework_output,
"%s Comm failure on my own connection - ignoring it",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
goto cleanup;
}
/* mark the daemon as gone */
ORTE_FLAG_UNSET(pptr, ORTE_PROC_FLAG_ALIVE);
/* if we have ordered orteds to terminate or abort
* is in progress, record it */
if (orte_orteds_term_ordered || orte_abnormal_term_ordered) {
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base_framework.framework_output,
"%s Comm failure: daemons terminating - recording daemon %s as gone",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), ORTE_NAME_PRINT(proc)));
/* remove from dependent routes, if it is one */
orte_routed.route_lost(proc);
/* if all my routes and local children are gone, then terminate ourselves */
if (0 == orte_routed.num_routes()) {
for (i=0; i < orte_local_children->size; i++) {
if (NULL != (proct = (orte_proc_t*)opal_pointer_array_get_item(orte_local_children, i)) &&
ORTE_FLAG_TEST(pptr, ORTE_PROC_FLAG_ALIVE) && proct->state < ORTE_PROC_STATE_UNTERMINATED) {
/* at least one is still alive */
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base_framework.framework_output,
"%s Comm failure: at least one proc (%s) still alive",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&proct->name)));
goto cleanup;
}
}
/* call our appropriate exit procedure */
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base_framework.framework_output,
"%s errmgr_hnp: all routes and children gone - ordering exit",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
ORTE_ACTIVATE_JOB_STATE(NULL, ORTE_JOB_STATE_DAEMONS_TERMINATED);
} else {
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base_framework.framework_output,
"%s Comm failure: %d routes remain alive",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
(int)orte_routed.num_routes()));
}
goto cleanup;
}
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base_framework.framework_output,
"%s Comm failure: daemon %s - aborting",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), ORTE_NAME_PRINT(proc)));
/* record the first one to fail */
if (!ORTE_FLAG_TEST(jdata, ORTE_JOB_FLAG_ABORTED)) {
/* output an error message so the user knows what happened */
orte_show_help("help-errmgr-base.txt", "node-died", true, pptr->node->name);
/* mark the daemon job as failed */
jdata->state = ORTE_JOB_STATE_COMM_FAILED;
/* point to the lowest rank to cause the problem */
orte_set_attribute(&jdata->attributes, ORTE_JOB_ABORTED_PROC, ORTE_ATTR_LOCAL, pptr, OPAL_PTR);
/* retain the object so it doesn't get free'd */
OBJ_RETAIN(pptr);
ORTE_FLAG_SET(jdata, ORTE_JOB_FLAG_ABORTED);
/* update our exit code */
ORTE_UPDATE_EXIT_STATUS(pptr->exit_code);
/* just in case the exit code hadn't been set, do it here - this
* won't override any reported exit code */
ORTE_UPDATE_EXIT_STATUS(ORTE_ERR_COMM_FAILURE);
}
/* abort the system */
default_hnp_abort(jdata);
goto cleanup;
}
/* update the proc state - can get multiple reports on a proc
* depending on circumstances, so ensure we only do this once
*/
if (pptr->state < ORTE_PROC_STATE_TERMINATED) {
pptr->state = state;
}
/* if we were ordered to terminate, mark this proc as dead and see if
* any of our routes or local children remain alive - if not, then
* terminate ourselves. */
if (orte_orteds_term_ordered) {
for (i=0; i < orte_local_children->size; i++) {
if (NULL != (proct = (orte_proc_t*)opal_pointer_array_get_item(orte_local_children, i))) {
if (ORTE_FLAG_TEST(proct, ORTE_PROC_FLAG_ALIVE)) {
goto keep_going;
}
}
}
/* 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_framework.framework_output,
"%s errmgr:default:hnp all routes gone - exiting",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
ORTE_ACTIVATE_JOB_STATE(NULL, ORTE_JOB_STATE_DAEMONS_TERMINATED);
}
}
keep_going:
/* ensure we record the failed proc properly so we can report
* the error once we terminate
*/
switch (state) {
case ORTE_PROC_STATE_KILLED_BY_CMD:
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base_framework.framework_output,
"%s errmgr:hnp: proc %s killed by cmd",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(proc)));
/* we ordered this proc to die, so it isn't an abnormal termination
* and we don't flag it as such
*/
if (jdata->num_terminated >= jdata->num_procs) {
/* this job has terminated */
ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_TERMINATED);
}
/* don't abort the job as this isn't an abnormal termination */
break;
case ORTE_PROC_STATE_ABORTED:
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base_framework.framework_output,
"%s errmgr:hnp: proc %s aborted",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(proc)));
if (!ORTE_FLAG_TEST(jdata, ORTE_JOB_FLAG_ABORTED)) {
jdata->state = ORTE_JOB_STATE_ABORTED;
/* point to the first rank to cause the problem */
orte_set_attribute(&jdata->attributes, ORTE_JOB_ABORTED_PROC, ORTE_ATTR_LOCAL, pptr, OPAL_PTR);
/* retain the object so it doesn't get free'd */
OBJ_RETAIN(pptr);
ORTE_FLAG_SET(jdata, ORTE_JOB_FLAG_ABORTED);
ORTE_UPDATE_EXIT_STATUS(pptr->exit_code);
/* abnormal termination - abort, but only do it once
* to avoid creating a lot of confusion */
default_hnp_abort(jdata);
}
break;
case ORTE_PROC_STATE_ABORTED_BY_SIG:
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base_framework.framework_output,
"%s errmgr:hnp: proc %s aborted by signal",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(proc)));
ORTE_UPDATE_EXIT_STATUS(pptr->exit_code);
/* track the number of non-zero exits */
i32 = 0;
i32ptr = &i32;
orte_get_attribute(&jdata->attributes, ORTE_JOB_NUM_NONZERO_EXIT, (void**)&i32ptr, OPAL_INT32);
++i32;
orte_set_attribute(&jdata->attributes, ORTE_JOB_NUM_NONZERO_EXIT, ORTE_ATTR_LOCAL, i32ptr, OPAL_INT32);
if (orte_abort_non_zero_exit) {
if (!ORTE_FLAG_TEST(jdata, ORTE_JOB_FLAG_ABORTED)) {
jdata->state = ORTE_JOB_STATE_ABORTED_BY_SIG;
/* point to the first rank to cause the problem */
orte_set_attribute(&jdata->attributes, ORTE_JOB_ABORTED_PROC, ORTE_ATTR_LOCAL, pptr, OPAL_PTR);
/* retain the object so it doesn't get free'd */
OBJ_RETAIN(pptr);
ORTE_FLAG_SET(jdata, ORTE_JOB_FLAG_ABORTED);
ORTE_UPDATE_EXIT_STATUS(pptr->exit_code);
/* abnormal termination - abort, but only do it once
* to avoid creating a lot of confusion */
default_hnp_abort(jdata);
}
} else {
/* user requested we consider this normal termination */
if (jdata->num_terminated >= jdata->num_procs) {
/* this job has terminated */
ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_TERMINATED);
}
}
break;
case ORTE_PROC_STATE_TERM_WO_SYNC:
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base_framework.framework_output,
"%s errmgr:hnp: proc %s terminated without sync",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(proc)));
if (!ORTE_FLAG_TEST(jdata, ORTE_JOB_FLAG_ABORTED)) {
jdata->state = ORTE_JOB_STATE_ABORTED_WO_SYNC;
/* point to the first rank to cause the problem */
orte_set_attribute(&jdata->attributes, ORTE_JOB_ABORTED_PROC, ORTE_ATTR_LOCAL, pptr, OPAL_PTR);
/* retain the object so it doesn't get free'd */
OBJ_RETAIN(pptr);
ORTE_FLAG_SET(jdata, ORTE_JOB_FLAG_ABORTED);
ORTE_UPDATE_EXIT_STATUS(pptr->exit_code);
/* now treat a special case - if the proc exit'd without a required
* sync, it may have done so with a zero exit code. We want to ensure
* that the user realizes there was an error, so in this -one- case,
* we overwrite the process' exit code with the default error code
*/
ORTE_UPDATE_EXIT_STATUS(ORTE_ERROR_DEFAULT_EXIT_CODE);
/* abnormal termination - abort, but only do it once
* to avoid creating a lot of confusion */
default_hnp_abort(jdata);
}
break;
case ORTE_PROC_STATE_FAILED_TO_START:
case ORTE_PROC_STATE_FAILED_TO_LAUNCH:
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base_framework.framework_output,
"%s errmgr:hnp: proc %s %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(proc),
orte_proc_state_to_str(state)));
if (!ORTE_FLAG_TEST(jdata, ORTE_JOB_FLAG_ABORTED)) {
if (ORTE_PROC_STATE_FAILED_TO_START) {
jdata->state = ORTE_JOB_STATE_FAILED_TO_START;
} else {
jdata->state = ORTE_JOB_STATE_FAILED_TO_LAUNCH;
}
/* point to the first rank to cause the problem */
orte_set_attribute(&jdata->attributes, ORTE_JOB_ABORTED_PROC, ORTE_ATTR_LOCAL, pptr, OPAL_PTR);
/* retain the object so it doesn't get free'd */
OBJ_RETAIN(pptr);
ORTE_FLAG_SET(jdata, ORTE_JOB_FLAG_ABORTED);
ORTE_UPDATE_EXIT_STATUS(pptr->exit_code);
/* abnormal termination - abort, but only do it once
* to avoid creating a lot of confusion */
default_hnp_abort(jdata);
}
/* if this was a daemon, report it */
if (jdata->jobid == ORTE_PROC_MY_NAME->jobid) {
/* output a message indicating we failed to launch a daemon */
orte_show_help("help-errmgr-base.txt", "failed-daemon-launch", true);
}
break;
case ORTE_PROC_STATE_CALLED_ABORT:
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base_framework.framework_output,
"%s errmgr:hnp: proc %s called abort with exit code %d",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(proc), pptr->exit_code));
if (!ORTE_FLAG_TEST(jdata, ORTE_JOB_FLAG_ABORTED)) {
jdata->state = ORTE_JOB_STATE_CALLED_ABORT;
/* point to the first proc to cause the problem */
orte_set_attribute(&jdata->attributes, ORTE_JOB_ABORTED_PROC, ORTE_ATTR_LOCAL, pptr, OPAL_PTR);
/* retain the object so it doesn't get free'd */
OBJ_RETAIN(pptr);
ORTE_FLAG_SET(jdata, ORTE_JOB_FLAG_ABORTED);
ORTE_UPDATE_EXIT_STATUS(pptr->exit_code);
/* abnormal termination - abort, but only do it once
* to avoid creating a lot of confusion */
default_hnp_abort(jdata);
}
break;
case ORTE_PROC_STATE_TERM_NON_ZERO:
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base_framework.framework_output,
"%s errmgr:hnp: proc %s exited with non-zero status %d",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(proc),
pptr->exit_code));
ORTE_UPDATE_EXIT_STATUS(pptr->exit_code);
/* track the number of non-zero exits */
i32 = 0;
i32ptr = &i32;
orte_get_attribute(&jdata->attributes, ORTE_JOB_NUM_NONZERO_EXIT, (void**)&i32ptr, OPAL_INT32);
++i32;
orte_set_attribute(&jdata->attributes, ORTE_JOB_NUM_NONZERO_EXIT, ORTE_ATTR_LOCAL, i32ptr, OPAL_INT32);
if (orte_abort_non_zero_exit) {
if (!ORTE_FLAG_TEST(jdata, ORTE_JOB_FLAG_ABORTED)) {
jdata->state = ORTE_JOB_STATE_NON_ZERO_TERM;
/* point to the first rank to cause the problem */
orte_set_attribute(&jdata->attributes, ORTE_JOB_ABORTED_PROC, ORTE_ATTR_LOCAL, pptr, OPAL_PTR);
/* retain the object so it doesn't get free'd */
OBJ_RETAIN(pptr);
ORTE_FLAG_SET(jdata, ORTE_JOB_FLAG_ABORTED);
/* abnormal termination - abort, but only do it once
* to avoid creating a lot of confusion */
default_hnp_abort(jdata);
}
} else {
/* user requested we consider this normal termination */
if (jdata->num_terminated >= jdata->num_procs) {
/* this job has terminated */
ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_TERMINATED);
}
}
break;
case ORTE_PROC_STATE_HEARTBEAT_FAILED:
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base_framework.framework_output,
"%s errmgr:hnp: proc %s heartbeat failed",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(proc)));
if (!ORTE_FLAG_TEST(jdata, ORTE_JOB_FLAG_ABORTED)) {
jdata->state = ORTE_JOB_STATE_HEARTBEAT_FAILED;
/* point to the first rank to cause the problem */
orte_set_attribute(&jdata->attributes, ORTE_JOB_ABORTED_PROC, ORTE_ATTR_LOCAL, pptr, OPAL_PTR);
/* retain the object so it doesn't get free'd */
OBJ_RETAIN(pptr);
ORTE_FLAG_SET(jdata, ORTE_JOB_FLAG_ABORTED);
ORTE_UPDATE_EXIT_STATUS(pptr->exit_code);
/* abnormal termination - abort, but only do it once
* to avoid creating a lot of confusion */
default_hnp_abort(jdata);
}
/* remove from dependent routes, if it is one */
orte_routed.route_lost(proc);
break;
case ORTE_PROC_STATE_UNABLE_TO_SEND_MSG:
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base_framework.framework_output,
"%s errmgr:hnp: unable to send message to proc %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(proc)));
/* if this proc is one of my daemons, then we are truly
* hosed - so just exit out
*/
if (ORTE_PROC_MY_NAME->jobid == proc->jobid) {
ORTE_ACTIVATE_JOB_STATE(NULL, ORTE_JOB_STATE_DAEMONS_TERMINATED);
break;
}
if (!ORTE_FLAG_TEST(jdata, ORTE_JOB_FLAG_ABORTED)) {
/* abnormal termination - abort, but only do it once
* to avoid creating a lot of confusion */
default_hnp_abort(jdata);
}
break;
default:
/* shouldn't get this, but terminate job if required */
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base_framework.framework_output,
"%s errmgr:hnp: proc %s default error %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(proc),
orte_proc_state_to_str(state)));
if (jdata->num_terminated == jdata->num_procs) {
ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_TERMINATED);
}
break;
}
/* if the waitpid fired, be sure to let the state machine know */
if (ORTE_FLAG_TEST(pptr, ORTE_PROC_FLAG_WAITPID)) {
ORTE_ACTIVATE_PROC_STATE(&pptr->name, ORTE_PROC_STATE_WAITPID_FIRED);
}
cleanup:
OBJ_RELEASE(caddy);
}
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, *proct;
orte_process_name_t *proc = &caddy->name;
orte_proc_state_t state = caddy->proc_state;
int i;
OPAL_OUTPUT_VERBOSE((1, orte_errmgr_base.output,
"%s errmgr:default_hnp: for proc %s state %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(proc),
orte_proc_state_to_str(state)));
/*
* if orte is trying to shutdown, just let it
*/
if (orte_finalizing) {
goto cleanup;
}
/* get the job object */
if (NULL == (jdata = orte_get_job_data_object(proc->jobid))) {
/* if the orteds are terminating, check job complete */
if (orte_orteds_term_ordered) {
opal_output(0, "TERM ORDERED - CHECKING COMPLETE");
goto cleanup;
} else {
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
goto cleanup;
}
}
pptr = (orte_proc_t*)opal_pointer_array_get_item(jdata->procs, proc->vpid);
/* we MUST handle a communication failure before doing anything else
* as it requires some special care to avoid normal termination issues
* for local application procs
*/
if (ORTE_PROC_STATE_COMM_FAILED == state) {
/* is this to a daemon? */
if (ORTE_PROC_MY_NAME->jobid != proc->jobid) {
/* nope - ignore it */
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base.output,
"%s Comm failure to non-daemon proc - ignoring it",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
goto cleanup;
}
/* if this is my own connection, ignore it */
if (ORTE_PROC_MY_NAME->vpid == proc->vpid) {
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base.output,
"%s Comm failure on my own connection - ignoring it",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
goto cleanup;
}
/* if we have ordered orteds to terminate or abort
* is in progress, record it */
if (orte_orteds_term_ordered || orte_abnormal_term_ordered) {
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base.output,
"%s Comm failure: daemons terminating - recording daemon %s as gone",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), ORTE_NAME_PRINT(proc)));
/* remove from dependent routes, if it is one */
orte_routed.route_lost(proc);
/* if all my routes and local children are gone, then terminate ourselves */
if (0 == orte_routed.num_routes()) {
for (i=0; i < orte_local_children->size; i++) {
if (NULL != (proct = (orte_proc_t*)opal_pointer_array_get_item(orte_local_children, i)) &&
proct->alive && proct->state < ORTE_PROC_STATE_UNTERMINATED) {
/* at least one is still alive */
goto cleanup;
}
}
/* call our appropriate exit procedure */
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base.output,
"%s errmgr_hnp: all routes and children gone - ordering exit",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
ORTE_ACTIVATE_JOB_STATE(NULL, ORTE_JOB_STATE_DAEMONS_TERMINATED);
}
goto cleanup;
}
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base.output,
"%s Comm failure: daemon %s - aborting",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), ORTE_NAME_PRINT(proc)));
/* record the first one to fail */
if (!jdata->abort) {
jdata->state = ORTE_JOB_STATE_COMM_FAILED;
/* point to the lowest rank to cause the problem */
jdata->aborted_proc = pptr;
/* retain the object so it doesn't get free'd */
OBJ_RETAIN(pptr);
jdata->abort = true;
ORTE_UPDATE_EXIT_STATUS(pptr->exit_code);
}
/* abort the system */
default_hnp_abort(jdata);
goto cleanup;
}
/* update the proc state - can get multiple reports on a proc
* depending on circumstances, so ensure we only do this once
*/
if (pptr->state < ORTE_PROC_STATE_TERMINATED) {
pptr->state = state;
jdata->num_terminated++;
}
/* since we only come here if the proc terminated,
* cleanup the local proc, if required
*/
cleanup_local_proc(jdata, proc);
/* ensure we record the failed proc properly so we can report
* the error once we terminate
*/
switch (state) {
case ORTE_PROC_STATE_KILLED_BY_CMD:
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base.output,
"%s errmgr:hnp: proc %s killed by cmd",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(proc)));
/* we ordered this proc to die, so it isn't an abnormal termination
* and we don't flag it as such
*/
if (jdata->num_terminated >= jdata->num_procs) {
/* this job has terminated */
ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_TERMINATED);
}
/* don't abort the job as this isn't an abnormal termination */
break;
case ORTE_PROC_STATE_ABORTED:
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base.output,
"%s errmgr:hnp: proc %s aborted",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(proc)));
if (!jdata->abort) {
jdata->state = ORTE_JOB_STATE_ABORTED;
/* point to the first rank to cause the problem */
jdata->aborted_proc = pptr;
/* retain the object so it doesn't get free'd */
OBJ_RETAIN(pptr);
jdata->abort = true;
ORTE_UPDATE_EXIT_STATUS(pptr->exit_code);
}
/* abnormal termination - abort */
default_hnp_abort(jdata);
break;
case ORTE_PROC_STATE_ABORTED_BY_SIG:
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base.output,
"%s errmgr:hnp: proc %s aborted by signal",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(proc)));
if (!jdata->abort) {
jdata->state = ORTE_JOB_STATE_ABORTED_BY_SIG;
/* point to the first rank to cause the problem */
jdata->aborted_proc = pptr;
/* retain the object so it doesn't get free'd */
OBJ_RETAIN(pptr);
jdata->abort = true;
ORTE_UPDATE_EXIT_STATUS(pptr->exit_code);
}
/* abnormal termination - abort */
default_hnp_abort(jdata);
break;
case ORTE_PROC_STATE_TERM_WO_SYNC:
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base.output,
"%s errmgr:hnp: proc %s terminated without sync",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(proc)));
if (!jdata->abort) {
jdata->state = ORTE_JOB_STATE_ABORTED_WO_SYNC;
/* point to the first rank to cause the problem */
jdata->aborted_proc = pptr;
/* retain the object so it doesn't get free'd */
OBJ_RETAIN(pptr);
jdata->abort = true;
ORTE_UPDATE_EXIT_STATUS(pptr->exit_code);
/* now treat a special case - if the proc exit'd without a required
* sync, it may have done so with a zero exit code. We want to ensure
* that the user realizes there was an error, so in this -one- case,
* we overwrite the process' exit code with the default error code
*/
ORTE_UPDATE_EXIT_STATUS(ORTE_ERROR_DEFAULT_EXIT_CODE);
}
/* abnormal termination - abort */
default_hnp_abort(jdata);
break;
case ORTE_PROC_STATE_FAILED_TO_START:
case ORTE_PROC_STATE_FAILED_TO_LAUNCH:
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base.output,
"%s errmgr:hnp: proc %s %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(proc),
orte_proc_state_to_str(state)));
if (!jdata->abort) {
if (ORTE_PROC_STATE_FAILED_TO_START) {
jdata->state = ORTE_JOB_STATE_FAILED_TO_START;
} else {
jdata->state = ORTE_JOB_STATE_FAILED_TO_LAUNCH;
}
/* point to the first rank to cause the problem */
jdata->aborted_proc = pptr;
/* retain the object so it doesn't get free'd */
OBJ_RETAIN(pptr);
jdata->abort = true;
ORTE_UPDATE_EXIT_STATUS(pptr->exit_code);
}
/* abnormal termination - abort */
default_hnp_abort(jdata);
break;
case ORTE_PROC_STATE_CALLED_ABORT:
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base.output,
"%s errmgr:hnp: proc %s called abort",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(proc)));
if (!jdata->abort) {
jdata->state = ORTE_JOB_STATE_CALLED_ABORT;
/* point to the first proc to cause the problem */
jdata->aborted_proc = pptr;
/* retain the object so it doesn't get free'd */
OBJ_RETAIN(pptr);
jdata->abort = true;
ORTE_UPDATE_EXIT_STATUS(pptr->exit_code);
}
/* abnormal termination - abort */
default_hnp_abort(jdata);
break;
case ORTE_PROC_STATE_SENSOR_BOUND_EXCEEDED:
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base.output,
"%s errmgr:hnp: proc %s exceeded sensor boundary",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(proc)));
if (!jdata->abort) {
jdata->state = ORTE_JOB_STATE_SENSOR_BOUND_EXCEEDED;
/* point to the lowest rank to cause the problem */
jdata->aborted_proc = pptr;
/* retain the object so it doesn't get free'd */
OBJ_RETAIN(pptr);
jdata->abort = true;
ORTE_UPDATE_EXIT_STATUS(pptr->exit_code);
}
/* abnormal termination - abort */
default_hnp_abort(jdata);
break;
case ORTE_PROC_STATE_TERM_NON_ZERO:
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base.output,
"%s errmgr:hnp: proc %s exited with non-zero status %d",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(proc),
pptr->exit_code));
ORTE_UPDATE_EXIT_STATUS(pptr->exit_code);
/* track the number of non-zero exits */
jdata->num_non_zero_exit++;
if (orte_abort_non_zero_exit) {
if (!jdata->abort) {
jdata->state = ORTE_JOB_STATE_NON_ZERO_TERM;
/* point to the first rank to cause the problem */
jdata->aborted_proc = pptr;
/* retain the object so it doesn't get free'd */
OBJ_RETAIN(pptr);
jdata->abort = true;
}
/* user requested we abort in this scenario */
default_hnp_abort(jdata);
} else {
/* user requested we consider this normal termination */
if (jdata->num_terminated >= jdata->num_procs) {
/* this job has terminated */
ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_TERMINATED);
}
}
break;
case ORTE_PROC_STATE_HEARTBEAT_FAILED:
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base.output,
"%s errmgr:hnp: proc %s heartbeat failed",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(proc)));
if (!jdata->abort) {
jdata->state = ORTE_JOB_STATE_HEARTBEAT_FAILED;
/* point to the first rank to cause the problem */
jdata->aborted_proc = pptr;
/* retain the object so it doesn't get free'd */
OBJ_RETAIN(pptr);
jdata->abort = true;
ORTE_UPDATE_EXIT_STATUS(pptr->exit_code);
}
/* remove from dependent routes, if it is one */
orte_routed.route_lost(proc);
/* kill all jobs */
default_hnp_abort(jdata);
break;
default:
/* shouldn't get this, but terminate job if required */
OPAL_OUTPUT_VERBOSE((5, orte_errmgr_base.output,
"%s errmgr:hnp: proc %s default error %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(proc),
orte_proc_state_to_str(state)));
if (jdata->num_terminated == jdata->num_procs) {
ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_TERMINATED);
}
break;
}
cleanup:
OBJ_RELEASE(caddy);
}
