ompi_mpi_abort: fix corner cases, simplify logic
I recently found a case where ompi_mpi_abort() segv's:
{{{
$ mpirun --mca btl non_existent_btl_name ...
}}}
In this case, the BML init fails because we have no paths to any
peers. It calls ompi_mpi_abort(), but this is before ompi_comm_self
has been setup. ompi_mpi_abort() assumes that if the comm parameter
is != NULL, it can be used. But since we aborted so early in
MPI_INIT, that's a false assumption.
(note that this isn't happening on v1.8 because the check for
INIT/FINALIZE in ompi_mpi_abort() is a little different. Hence: this
is a trunk issue -- at least for now)
When fixing this problem, I noticed a few other problems in ompi_mpi_abort():
* the group access was incorrect (it didn't use accessor functions)
* it wasn't clear that ORTE's ompi_rte_abort_peers() returns
NOT_IMPLEMENTED and falls through down to ompi_rte_abort()
* the check for my proc in the communicator was a little more
complicated than necessary
* the logic for checking for aborts early in MPI_INIT wasn't right
* some comments were stale
* the hostname output in error messages would be NULL if MPI_FINALIZE
had been invoked
* it was possible to abort, but still exit with a 0 status
This commit fixes all of the above problems, and makes the logic a
little more straightforward. Thanks to Ralph Castain and George
Bosilca for the assists with this patch.
This commit was SVN r32125.
Этот коммит содержится в:
Jeff Squyres
родитель
843ef1fcb0
Коммит
852af8b834
@ -2,14 +2,14 @@
|
||||
* Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
|
||||
* University Research and Technology
|
||||
* Corporation. All rights reserved.
|
||||
* Copyright (c) 2004-2011 The University of Tennessee and The University
|
||||
* Copyright (c) 2004-2014 The University of Tennessee and The University
|
||||
* of Tennessee Research Foundation. All rights
|
||||
* reserved.
|
||||
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
|
||||
* University of Stuttgart. All rights reserved.
|
||||
* Copyright (c) 2004-2005 The Regents of the University of California.
|
||||
* All rights reserved.
|
||||
* Copyright (c) 2006-2011 Cisco Systems, Inc. All rights reserved.
|
||||
* Copyright (c) 2006-2014 Cisco Systems, Inc. All rights reserved.
|
||||
* Copyright (c) 2010-2011 Oak Ridge National Labs. All rights reserved.
|
||||
* $COPYRIGHT$
|
||||
*
|
||||
@ -32,6 +32,7 @@
|
||||
#ifdef HAVE_NETDB_H
|
||||
#include <netdb.h>
|
||||
#endif
|
||||
#include <errno.h>
|
||||
|
||||
#include "opal/mca/backtrace/backtrace.h"
|
||||
|
||||
@ -43,15 +44,75 @@
|
||||
|
||||
static bool have_been_invoked = false;
|
||||
|
||||
|
||||
/*
|
||||
* Local helper function to build an array of all the procs in a
|
||||
* communicator, excluding this process.
|
||||
*
|
||||
* Killing a just the indicated peers must be implemented for
|
||||
* MPI_Abort() to work according to the standard language for
|
||||
* a 'high-quality' implementation.
|
||||
*
|
||||
* It would be nifty if we could differentiate between the
|
||||
* abort scenarios (but we don't, currently):
|
||||
* - MPI_Abort()
|
||||
* - MPI_ERRORS_ARE_FATAL
|
||||
* - Victim of MPI_Abort()
|
||||
*/
|
||||
static void try_kill_peers(ompi_communicator_t *comm,
|
||||
int errcode)
|
||||
{
|
||||
int nprocs;
|
||||
ompi_process_name_t *procs;
|
||||
|
||||
nprocs = ompi_comm_size(comm);
|
||||
/* ompi_comm_remote_size() returns 0 if not an intercomm, so
|
||||
this is safe */
|
||||
nprocs += ompi_comm_remote_size(comm);
|
||||
|
||||
procs = (ompi_process_name_t*) calloc(nprocs, sizeof(ompi_process_name_t));
|
||||
if (NULL == procs) {
|
||||
/* quick clean orte and get out */
|
||||
ompi_rte_abort(errno, "Abort: unable to alloc memory to kill procs");
|
||||
}
|
||||
|
||||
/* put all the local group procs in the abort list */
|
||||
int rank, i, count;
|
||||
rank = ompi_comm_rank(comm);
|
||||
for (count = i = 0; i < ompi_comm_size(comm); ++i) {
|
||||
if (rank == i) {
|
||||
/* Don't include this process in the array */
|
||||
--nprocs;
|
||||
} else {
|
||||
assert(count <= nprocs);
|
||||
procs[count++] =
|
||||
ompi_group_get_proc_ptr(comm->c_remote_group, i)->proc_name;
|
||||
}
|
||||
}
|
||||
|
||||
/* if requested, kill off remote group procs too */
|
||||
for (i = 0; i < ompi_comm_remote_size(comm); ++i) {
|
||||
assert(count <= nprocs);
|
||||
procs[count++] =
|
||||
ompi_group_get_proc_ptr(comm->c_remote_group, i)->proc_name;
|
||||
}
|
||||
|
||||
if (nprocs > 0) {
|
||||
ompi_rte_abort_peers(procs, nprocs, errcode);
|
||||
}
|
||||
|
||||
/* We could fall through here if ompi_rte_abort_peers() fails, or
|
||||
if (nprocs == 0). Either way, tidy up and let the caller
|
||||
handle it. */
|
||||
free(procs);
|
||||
}
|
||||
|
||||
int
|
||||
ompi_mpi_abort(struct ompi_communicator_t* comm,
|
||||
int errcode)
|
||||
{
|
||||
int count = 0, i, ret;
|
||||
char *msg, *host, hostname[MAXHOSTNAMELEN];
|
||||
pid_t pid = 0;
|
||||
ompi_process_name_t *abort_procs;
|
||||
int32_t nabort_procs;
|
||||
|
||||
/* Protection for recursive invocation */
|
||||
if (have_been_invoked) {
|
||||
@ -59,9 +120,9 @@ ompi_mpi_abort(struct ompi_communicator_t* comm,
|
||||
}
|
||||
have_been_invoked = true;
|
||||
|
||||
/* If MPI is initialized, we know we have a runtime nodename, so use that. Otherwise, call
|
||||
gethostname. */
|
||||
if (ompi_mpi_initialized) {
|
||||
/* If MPI is initialized, we know we have a runtime nodename, so
|
||||
use that. Otherwise, call gethostname. */
|
||||
if (ompi_rte_initialized) {
|
||||
host = ompi_process_info.nodename;
|
||||
} else {
|
||||
gethostname(hostname, sizeof(hostname));
|
||||
@ -123,79 +184,32 @@ ompi_mpi_abort(struct ompi_communicator_t* comm,
|
||||
}
|
||||
}
|
||||
|
||||
/* If OMPI isn't setup yet/any more, then don't even try killing
|
||||
everyone. OMPI's initialized period covers all runtime
|
||||
initialized period of time, so no need to check that here.
|
||||
Sorry, Charlie... */
|
||||
|
||||
if ((!ompi_mpi_initialized || ompi_mpi_finalized) && !ompi_rte_initialized) {
|
||||
fprintf(stderr, "[%s:%d] Local abort %s completed successfully; not able to aggregate error messages, and not able to guarantee that all other processes were killed!\n",
|
||||
/* If the RTE isn't setup yet/any more, then don't even try
|
||||
killing everyone. Sorry, Charlie... */
|
||||
if (!ompi_rte_initialized) {
|
||||
fprintf(stderr, "[%s:%d] Local abort %s completed successfully, but am not able to aggregate error messages, and not able to guarantee that all other processes were killed!\n",
|
||||
host, (int) pid, ompi_mpi_finalized ?
|
||||
"after MPI_FINALIZE" : "before MPI_INIT");
|
||||
exit(errcode);
|
||||
"after MPI_FINALIZE started" : "before MPI_INIT completed");
|
||||
exit(errcode == 0 ? 1 : errcode);
|
||||
}
|
||||
|
||||
/* abort local procs in the communicator. If the communicator is
|
||||
an intercommunicator AND the abort has explicitly requested
|
||||
that we abort the remote procs, then do that as well. */
|
||||
nabort_procs = ompi_comm_size(comm);
|
||||
|
||||
/* ompi_comm_remote_size() returns 0 if not an intercomm, so
|
||||
this is cool */
|
||||
nabort_procs += ompi_comm_remote_size(comm);
|
||||
|
||||
abort_procs = (ompi_process_name_t*)malloc(sizeof(ompi_process_name_t) * nabort_procs);
|
||||
if (NULL == abort_procs) {
|
||||
/* quick clean orte and get out */
|
||||
ompi_rte_abort(errcode, "Abort unable to malloc memory to kill procs");
|
||||
/* If OMPI is initialized and we have a non-NULL communicator,
|
||||
then try to kill just that set of processes */
|
||||
if (ompi_mpi_initialized && !ompi_mpi_finalized && NULL != comm) {
|
||||
try_kill_peers(comm, errcode);
|
||||
}
|
||||
|
||||
/* put all the local procs in the abort list */
|
||||
for (i = 0 ; i < ompi_comm_size(comm) ; ++i) {
|
||||
if (OPAL_EQUAL != ompi_rte_compare_name_fields(OMPI_RTE_CMP_ALL,
|
||||
&comm->c_local_group->grp_proc_pointers[i]->proc_name,
|
||||
OMPI_PROC_MY_NAME)) {
|
||||
assert(count <= nabort_procs);
|
||||
abort_procs[count++] = comm->c_local_group->grp_proc_pointers[i]->proc_name;
|
||||
} else {
|
||||
/* don't terminate me just yet */
|
||||
nabort_procs--;
|
||||
}
|
||||
}
|
||||
/* We can fall through to here in a few cases:
|
||||
|
||||
/* if requested, kill off remote procs too */
|
||||
for (i = 0 ; i < ompi_comm_remote_size(comm) ; ++i) {
|
||||
if (OPAL_EQUAL != ompi_rte_compare_name_fields(OMPI_RTE_CMP_ALL,
|
||||
&comm->c_remote_group->grp_proc_pointers[i]->proc_name,
|
||||
OMPI_PROC_MY_NAME)) {
|
||||
assert(count <= nabort_procs);
|
||||
abort_procs[count++] =
|
||||
comm->c_remote_group->grp_proc_pointers[i]->proc_name;
|
||||
} else {
|
||||
/* don't terminate me just yet */
|
||||
nabort_procs--;
|
||||
}
|
||||
}
|
||||
1. The attempt to kill just a subset of peers via
|
||||
try_kill_peers() failed (e.g., as of July 2014, ORTE does
|
||||
returns NOT_IMPLENTED from orte_rte_abort_peers()).
|
||||
2. MPI wasn't initialized, was already finalized, or we got a
|
||||
NULL communicator.
|
||||
|
||||
if (nabort_procs > 0) {
|
||||
/* This must be implemented for MPI_Abort() to work according to the
|
||||
* standard language for a 'high-quality' implementation.
|
||||
* It would be nifty if we could differentiate between the
|
||||
* abort scenarios:
|
||||
* - MPI_Abort()
|
||||
* - MPI_ERRORS_ARE_FATAL
|
||||
* - Victim of MPI_Abort()
|
||||
*/
|
||||
/*
|
||||
* Abort peers in this communicator group. Does not include self.
|
||||
*/
|
||||
if( OMPI_SUCCESS != (ret = ompi_rte_abort_peers(abort_procs, nabort_procs, errcode)) ) {
|
||||
ompi_rte_abort(errcode, "Open MPI failed to abort all of the procs requested (%d).", ret);
|
||||
}
|
||||
}
|
||||
|
||||
/* now that we've aborted everyone else, gracefully die. */
|
||||
In all of these cases, the only sensible thing left to do is to
|
||||
kill the entire job. Wah wah. */
|
||||
ompi_rte_abort(errcode, NULL);
|
||||
|
||||
return OMPI_SUCCESS;
|
||||
/* Does not return */
|
||||
}
|
||||
|
||||
Загрузка…
Ссылка в новой задаче
Block a user