void opal_stackframe_output(int stream)
{
int traces_size;
char **traces;
/* print out the stack trace */
if (OPAL_SUCCESS == opal_backtrace_buffer(&traces, &traces_size)) {
int i;
/* since we have the opportunity, strip off the bottom two
function calls, which will be this function and
opal_backtrace_buffer(). */
for (i = 2; i < traces_size; ++i) {
opal_output(stream, "%s", traces[i]);
}
} else {
opal_backtrace_print(stderr, NULL, 2);
}
}
char *opal_stackframe_output_string(void)
{
int traces_size, i;
size_t len;
char *output, **traces;
len = 0;
if (OPAL_SUCCESS != opal_backtrace_buffer(&traces, &traces_size)) {
return NULL;
}
/* Calculate the space needed for the string */
for (i = 3; i < traces_size; i++) {
if (NULL == traces[i]) {
break;
}
len += strlen(traces[i]) + 1;
}
output = (char *) malloc(len + 1);
if (NULL == output) {
return NULL;
}
*output = '\0';
for (i = 3; i < traces_size; i++) {
if (NULL == traces[i]) {
break;
}
strcat(output, traces[i]);
strcat(output, "\n");
}
free(traces);
return output;
}
int
ompi_mpi_abort(struct ompi_communicator_t* comm,
int errcode,
bool kill_remote_of_intercomm)
{
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) {
return OMPI_SUCCESS;
}
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) {
host = ompi_process_info.nodename;
} else {
gethostname(hostname, sizeof(hostname));
host = hostname;
}
pid = getpid();
/* Should we print a stack trace? Not aggregated because they
might be different on all processes. */
if (ompi_mpi_abort_print_stack) {
char **messages;
int len, i;
if (OMPI_SUCCESS == opal_backtrace_buffer(&messages, &len)) {
for (i = 0; i < len; ++i) {
fprintf(stderr, "[%s:%d] [%d] func:%s\n", host, (int) pid,
i, messages[i]);
fflush(stderr);
}
free(messages);
} else {
/* This will print an message if it's unable to print the
backtrace, so we don't need an additional "else" clause
if opal_backtrace_print() is not supported. */
opal_backtrace_print(stderr, NULL, 1);
}
}
/* Notify the debugger that we're about to abort */
if (errcode < 0 ||
asprintf(&msg, "[%s:%d] aborting with MPI error %s%s",
host, (int) pid, ompi_mpi_errnum_get_string(errcode),
ompi_mpi_abort_print_stack ?
" (stack trace available on stderr)" : "") < 0) {
msg = NULL;
}
ompi_debugger_notify_abort(msg);
if (NULL != msg) {
free(msg);
}
/* Should we wait for a while before aborting? */
if (0 != ompi_mpi_abort_delay) {
if (ompi_mpi_abort_delay < 0) {
fprintf(stderr ,"[%s:%d] Looping forever (MCA parameter mpi_abort_delay is < 0)\n",
host, (int) pid);
fflush(stderr);
while (1) {
sleep(5);
}
} else {
fprintf(stderr, "[%s:%d] Delaying for %d seconds before aborting\n",
host, (int) pid, ompi_mpi_abort_delay);
do {
sleep(1);
} while (--ompi_mpi_abort_delay > 0);
}
}
/* 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) {
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",
host, (int) pid, ompi_mpi_finalized ?
"after MPI_FINALIZE" : "before MPI_INIT");
exit(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);
if (kill_remote_of_intercomm) {
/* 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");
}
/* 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--;
}
}
/* if requested, kill off remote procs too */
if (kill_remote_of_intercomm) {
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--;
}
}
}
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. */
ompi_rte_abort(errcode, NULL);
return OMPI_SUCCESS;
}
int
ompi_mpi_abort(struct ompi_communicator_t* comm,
int errcode,
bool kill_remote_of_intercomm)
{
int count = 0, i;
char *msg, *host, hostname[MAXHOSTNAMELEN];
pid_t pid = 0;
orte_process_name_t *abort_procs;
orte_std_cntr_t nabort_procs;
/* Protection for recursive invocation */
if (have_been_invoked) {
return OMPI_SUCCESS;
}
have_been_invoked = true;
/* If ORTE is initialized, use its nodename. Otherwise, call
gethostname. */
if (orte_initialized) {
host = orte_process_info.nodename;
} else {
gethostname(hostname, sizeof(hostname));
host = hostname;
}
pid = getpid();
/* Should we print a stack trace? */
if (ompi_mpi_abort_print_stack) {
char **messages;
int len, i;
if (OMPI_SUCCESS == opal_backtrace_buffer(&messages, &len)) {
for (i = 0; i < len; ++i) {
fprintf(stderr, "[%s:%d] [%d] func:%s\n", host, (int) pid,
i, messages[i]);
fflush(stderr);
}
free(messages);
} else {
/* This will print an message if it's unable to print the
backtrace, so we don't need an additional "else" clause
if opal_backtrace_print() is not supported. */
opal_backtrace_print(stderr);
}
}
/* Notify the debugger that we're about to abort */
if (errcode < 0 ||
asprintf(&msg, "[%s:%d] aborting with MPI error %s%s",
host, (int) pid, ompi_mpi_errnum_get_string(errcode),
ompi_mpi_abort_print_stack ?
" (stack trace available on stderr)" : "") < 0) {
msg = NULL;
}
ompi_debugger_notify_abort(msg);
if (NULL != msg) {
free(msg);
}
/* Should we wait for a while before aborting? */
if (0 != ompi_mpi_abort_delay) {
if (ompi_mpi_abort_delay < 0) {
fprintf(stderr ,"[%s:%d] Looping forever (MCA parameter mpi_abort_delay is < 0)\n",
host, (int) pid);
fflush(stderr);
while (1) {
sleep(5);
}
} else {
fprintf(stderr, "[%s:%d] Delaying for %d seconds before aborting\n",
host, (int) pid, ompi_mpi_abort_delay);
do {
sleep(1);
} while (--ompi_mpi_abort_delay > 0);
}
}
/* If OMPI isn't setup yet/any more, then don't even try killing
everyone. Ditto for ORTE (e.g., ORTE may be initialized before
MPI_INIT is over, but ompi_initialized will be false because
communicators are not setup yet). Sorry, Charlie... */
if (!orte_initialized || !ompi_mpi_initialized || ompi_mpi_finalized) {
fprintf(stderr, "[%s:%d] Abort %s completed successfully; 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);
}
/* 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);
if (kill_remote_of_intercomm) {
/* ompi_comm_remote_size() returns 0 if not an intercomm, so
this is cool */
nabort_procs += ompi_comm_remote_size(comm);
}
abort_procs = (orte_process_name_t*)malloc(sizeof(orte_process_name_t) * nabort_procs);
if (NULL == abort_procs) {
/* quick clean orte and get out */
orte_errmgr.abort(errcode, "Abort unable to malloc memory to kill procs");
}
/* put all the local procs in the abort list */
for (i = 0 ; i < ompi_comm_size(comm) ; ++i) {
if (OPAL_EQUAL != orte_util_compare_name_fields(ORTE_NS_CMP_ALL,
&comm->c_local_group->grp_proc_pointers[i]->proc_name,
ORTE_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--;
}
}
/* if requested, kill off remote procs too */
if (kill_remote_of_intercomm) {
for (i = 0 ; i < ompi_comm_remote_size(comm) ; ++i) {
if (OPAL_EQUAL != orte_util_compare_name_fields(ORTE_NS_CMP_ALL,
&comm->c_remote_group->grp_proc_pointers[i]->proc_name,
ORTE_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--;
}
}
}
if (nabort_procs > 0) {
#if 0
int ret = orte_errmgr.abort_procs_request(abort_procs, nabort_procs);
if (OMPI_SUCCESS != ret) {
orte_errmgr.abort(ret, "Open MPI failed to abort procs as requested (%d). Exiting.", ret);
}
#endif
}
/* now that we've aborted everyone else, gracefully die. */
orte_errmgr.abort(errcode, NULL);
return OMPI_SUCCESS;
}
int
ompi_mpi_abort(struct ompi_communicator_t* comm,
int errcode)
{
char *msg, *host, hostname[MAXHOSTNAMELEN];
pid_t pid = 0;
/* Protection for recursive invocation */
if (have_been_invoked) {
return OMPI_SUCCESS;
}
have_been_invoked = true;
/* 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));
host = hostname;
}
pid = getpid();
/* Should we print a stack trace? Not aggregated because they
might be different on all processes. */
if (ompi_mpi_abort_print_stack) {
char **messages;
int len, i;
if (OMPI_SUCCESS == opal_backtrace_buffer(&messages, &len)) {
for (i = 0; i < len; ++i) {
fprintf(stderr, "[%s:%d] [%d] func:%s\n", host, (int) pid,
i, messages[i]);
fflush(stderr);
}
free(messages);
} else {
/* This will print an message if it's unable to print the
backtrace, so we don't need an additional "else" clause
if opal_backtrace_print() is not supported. */
opal_backtrace_print(stderr, NULL, 1);
}
}
/* Notify the debugger that we're about to abort */
if (errcode < 0 ||
asprintf(&msg, "[%s:%d] aborting with MPI error %s%s",
host, (int) pid, ompi_mpi_errnum_get_string(errcode),
ompi_mpi_abort_print_stack ?
" (stack trace available on stderr)" : "") < 0) {
msg = NULL;
}
ompi_debugger_notify_abort(msg);
if (NULL != msg) {
free(msg);
}
/* Should we wait for a while before aborting? */
if (0 != ompi_mpi_abort_delay) {
if (ompi_mpi_abort_delay < 0) {
fprintf(stderr ,"[%s:%d] Looping forever (MCA parameter mpi_abort_delay is < 0)\n",
host, (int) pid);
fflush(stderr);
while (1) {
sleep(5);
}
} else {
fprintf(stderr, "[%s:%d] Delaying for %d seconds before aborting\n",
host, (int) pid, ompi_mpi_abort_delay);
do {
sleep(1);
} while (--ompi_mpi_abort_delay > 0);
}
}
/* 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 started" : "before MPI_INIT completed");
_exit(errcode == 0 ? 1 : errcode);
}
/* 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);
}
/* We can fall through to here in a few cases:
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.
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);
/* Does not return */
}
int oshmem_shmem_abort(int errcode)
{
char *host, hostname[OPAL_MAXHOSTNAMELEN];
pid_t pid = 0;
/* Protection for recursive invocation */
if (have_been_invoked) {
return OSHMEM_SUCCESS;
}
have_been_invoked = true;
/* If ORTE is initialized, use its nodename. Otherwise, call
gethostname. */
/* 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));
host = hostname;
}
pid = getpid();
opal_show_help("help-shmem-api.txt",
"shmem-abort",
true,
OMPI_PROC_MY_NAME->vpid,
pid,
host,
errcode);
/* Should we print a stack trace? Not aggregated because they
might be different on all processes. */
if (opal_abort_print_stack) {
char **messages;
int len, i;
if (OPAL_SUCCESS == opal_backtrace_buffer(&messages, &len)) {
for (i = 0; i < len; ++i) {
fprintf(stderr,
"[%s:%05d] [%d] func:%s\n",
host,
(int) pid,
i,
messages[i]);
fflush(stderr);
}
free(messages);
} else {
/* This will print an message if it's unable to print the
backtrace, so we don't need an additional "else" clause
if opal_backtrace_print() is not supported. */
opal_backtrace_print(stderr, NULL, 1);
}
}
/* Wait for a while before aborting */
opal_delay_abort();
if (!oshmem_shmem_initialized) {
if (!opal_initialized) {
/* TODO help message from SHMEM not from MPI is needed*/
opal_show_help("help-shmem-runtime.txt",
"oshmem shmem abort:cannot guarantee all killed",
true,
host,
(int) pid);
} else {
fprintf(stderr,
"[%s:%05d] Local abort completed successfully; not able to aggregate error messages, and not able to guarantee that all other processes were killed!\n",
host,
(int) pid);
}
oshmem_shmem_aborted = true;
exit(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. */
oshmem_shmem_aborted = true;
/* now that we've aborted everyone else, gracefully die. */
ompi_rte_abort(errcode, NULL );
return OSHMEM_SUCCESS;
}
