Always followup with sigkill when killing local procs as procs can trap sigterm and get stuck
This commit was SVN r24719.
Этот коммит содержится в:
родитель
26034f4a96
Коммит
1b5ca323c6
@ -3000,16 +3000,14 @@ int orte_odls_base_default_kill_local_procs(opal_pointer_array_t *procs,
|
||||
"odls-default:could-not-kill",
|
||||
true, orte_process_info.nodename, child->pid);
|
||||
}
|
||||
}
|
||||
#if OPAL_ENABLE_FT_CR
|
||||
} else {
|
||||
/* Force the SIGKILL just to make sure things are dead
|
||||
* This fixes an issue with process migration/autorecovery
|
||||
* if the application is masking SIGTERM then the child_died()
|
||||
* This fixes an issue that, if the application is masking
|
||||
* SIGTERM, then the child_died()
|
||||
* may return 'true' even though waipid returns with 0.
|
||||
* It does this to avoid a race condition, per documentation
|
||||
* in odls_default_module.c.
|
||||
*/
|
||||
else {
|
||||
kill_local(child->pid, SIGKILL);
|
||||
/* Double check that it actually died this time */
|
||||
if (!child_died(child)) {
|
||||
@ -3018,7 +3016,7 @@ int orte_odls_base_default_kill_local_procs(opal_pointer_array_t *procs,
|
||||
true, orte_process_info.nodename, child->pid);
|
||||
}
|
||||
}
|
||||
#endif
|
||||
|
||||
OPAL_OUTPUT_VERBOSE((5, orte_odls_globals.output,
|
||||
"%s odls:kill_local_proc child %s killed",
|
||||
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
|
||||
|
Загрузка…
Ссылка в новой задаче
Block a user