int test_error(char *my_nspace, int my_rank, test_params params)
{
size_t errhandler_refs[MAX_ERR_HANDLERS];
struct timespec ts;
pmix_status_t status;
pmix_proc_t source;
TEST_VERBOSE(("test-error: running error handling test cases"));
/* register specific client error handlers and test their invocation
* by trigerring events from server side*/
status = PMIX_ERR_TIMEOUT;
PMIx_Register_event_handler(&status, 1, NULL, 0,
timeout_errhandler, errhandler_reg_callbk1, &errhandler_refs[0]);
/* reg a handler for comm errors */
status = PMIX_ERR_LOST_PEER_CONNECTION;
PMIx_Register_event_handler(&status, 1, NULL, 0,
comfail_errhandler, errhandler_reg_callbk1, &errhandler_refs[1]);
/* inject error from client */
done = false;
(void)strncpy(source.nspace, my_nspace, PMIX_MAX_NSLEN);
source.rank = my_rank;
/* change error value to test other error notifications */
PMIx_Notify_event(TEST_NOTIFY,
&source, PMIX_RANGE_NAMESPACE,
NULL, 0,
op1_callbk, NULL);
while(!done) {
ts.tv_sec = 0;
ts.tv_nsec = 100000;
nanosleep(&ts, NULL);
}
done = false;
/* dereg all handlers*/
PMIx_Deregister_event_handler( errhandler_refs[0], op1_callbk, NULL);
/* loop until we get callback */
while(!done) {
ts.tv_sec = 0;
ts.tv_nsec = 100000;
nanosleep(&ts, NULL);
}
done = false;
PMIx_Deregister_event_handler( errhandler_refs[1], op1_callbk, NULL);
/* loop until we get callback */
while(!done) {
ts.tv_sec = 0;
ts.tv_nsec = 100000;
nanosleep(&ts, NULL);
}
return PMIX_SUCCESS;
}
int pmixp_lib_init(void)
{
pmix_info_t *kvp = NULL;
pmix_status_t rc;
PMIXP_INFO_ADD(kvp, PMIX_USERID, uint32_t, pmixp_info_jobuid());
#ifdef PMIX_SERVER_TMPDIR
PMIXP_INFO_ADD(kvp, PMIX_SERVER_TMPDIR, string,
pmixp_info_tmpdir_lib());
#endif
/* setup the server library */
if (PMIX_SUCCESS != (rc = PMIx_server_init(&slurm_pmix_cb, kvp,
PMIXP_INFO_SIZE(kvp)))) {
PMIXP_ERROR_STD("PMIx_server_init failed with error %d\n", rc);
return SLURM_ERROR;
}
PMIXP_FREE_KEY(kvp);
/* register the errhandler */
PMIx_Register_event_handler(NULL, 0, NULL, 0, _errhandler,
_errhandler_reg_callbk, NULL);
return SLURM_SUCCESS;
}
static pmix_status_t spawn_debugger(char *appspace, myrel_t *myrel)
{
pmix_status_t rc;
pmix_info_t *dinfo;
pmix_app_t *debugger;
size_t dninfo;
char cwd[1024];
char dspace[PMIX_MAX_NSLEN+1];
mylock_t mylock;
pmix_status_t code = PMIX_ERR_JOB_TERMINATED;
/* setup the debugger */
PMIX_APP_CREATE(debugger, 1);
debugger[0].cmd = strdup("./debuggerd");
PMIX_ARGV_APPEND(rc, debugger[0].argv, "./debuggerd");
getcwd(cwd, 1024); // point us to our current directory
debugger[0].cwd = strdup(cwd);
/* provide directives so the daemons go where we want, and
* let the RM know these are debugger daemons */
dninfo = 6;
PMIX_INFO_CREATE(dinfo, dninfo);
PMIX_INFO_LOAD(&dinfo[0], PMIX_MAPBY, "ppr:1:node", PMIX_STRING); // instruct the RM to launch one copy of the executable on each node
PMIX_INFO_LOAD(&dinfo[1], PMIX_DEBUGGER_DAEMONS, NULL, PMIX_BOOL); // these are debugger daemons
PMIX_INFO_LOAD(&dinfo[1], PMIX_DEBUG_JOB, appspace, PMIX_STRING); // the nspace being debugged
PMIX_INFO_LOAD(&dinfo[2], PMIX_NOTIFY_COMPLETION, NULL, PMIX_BOOL); // notify us when the debugger job completes
PMIX_INFO_LOAD(&dinfo[3], PMIX_DEBUG_WAITING_FOR_NOTIFY, NULL, PMIX_BOOL); // tell the daemon that the proc is waiting to be released
PMIX_INFO_LOAD(&dinfo[4], PMIX_FWD_STDOUT, NULL, PMIX_BOOL); // forward stdout to me
PMIX_INFO_LOAD(&dinfo[5], PMIX_FWD_STDERR, NULL, PMIX_BOOL); // forward stderr to me
/* spawn the daemons */
fprintf(stderr, "Debugger: spawning %s\n", debugger[0].cmd);
if (PMIX_SUCCESS != (rc = PMIx_Spawn(dinfo, dninfo, debugger, 1, dspace))) {
fprintf(stderr, "Debugger daemons failed to launch with error: %s\n", PMIx_Error_string(rc));
PMIX_INFO_FREE(dinfo, dninfo);
PMIX_APP_FREE(debugger, 1);
return rc;
}
/* cleanup */
PMIX_INFO_FREE(dinfo, dninfo);
PMIX_APP_FREE(debugger, 1);
/* register callback for when this job terminates */
myrel->nspace = strdup(dspace);
PMIX_INFO_CREATE(dinfo, 2);
PMIX_INFO_LOAD(&dinfo[0], PMIX_EVENT_RETURN_OBJECT, myrel, PMIX_POINTER);
/* only call me back when this specific job terminates */
PMIX_INFO_LOAD(&dinfo[1], PMIX_NSPACE, dspace, PMIX_STRING);
DEBUG_CONSTRUCT_LOCK(&mylock);
PMIx_Register_event_handler(&code, 1, dinfo, 2,
release_fn, evhandler_reg_callbk, (void*)&mylock);
DEBUG_WAIT_THREAD(&mylock);
rc = mylock.status;
DEBUG_DESTRUCT_LOCK(&mylock);
PMIX_INFO_FREE(dinfo, 2);
return rc;
}
int main(int argc, char **argv)
{
pmix_status_t rc;
pmix_value_t *val;
pmix_proc_t proc;
pmix_info_t *info;
size_t ninfo;
volatile int active;
pmix_query_t *query;
size_t nq, n;
myquery_data_t myquery_data;
fprintf(stderr, "I AM HERE\n");
fflush(stderr);
sleep(10);
exit(0);
/* init us - since we were launched by the RM, our connection info
* will have been provided at startup. */
if (PMIX_SUCCESS != (rc = PMIx_tool_init(&myproc, NULL, 0))) {
fprintf(stderr, "Debugger daemon ns %s rank %d: PMIx_tool_init failed: %d\n", myproc.nspace, myproc.rank, rc);
exit(0);
}
fprintf(stderr, "Debugger daemon ns %s rank %d: Running\n", myproc.nspace, myproc.rank);
/* register our default event handler */
active = -1;
PMIx_Register_event_handler(NULL, 0, NULL, 0,
notification_fn, evhandler_reg_callbk, (void*)&active);
while (-1 == active) {
usleep(10);
}
if (0 != active) {
exit(active);
}
/* get the nspace of the job we are to debug */
(void)strncpy(proc.nspace, myproc.nspace, PMIX_MAX_NSLEN);
proc.rank = PMIX_RANK_WILDCARD;
if (PMIX_SUCCESS != (rc = PMIx_Get(&proc, PMIX_DEBUG_JOB, NULL, 0, &val))) {
fprintf(stderr, "[%s:%d] Failed to get job being debugged - error %d\n", myproc.nspace, myproc.rank, rc);
goto done;
}
if (NULL == val) {
fprintf(stderr, "Got NULL return\n");
goto done;
}
fprintf(stderr, "[%s:%d] Debugging %s\n", myproc.nspace, myproc.rank, val->data.string);
/* get our local proctable - for scalability reasons, we don't want to
* have our "root" debugger process get the proctable for everybody and
* send it out to us. So ask the local PMIx server for the pid's of
* our local target processes */
nq = 1;
PMIX_QUERY_CREATE(query, nq);
PMIX_ARGV_APPEND(rc, query[0].keys, PMIX_QUERY_LOCAL_PROC_TABLE);
query[0].nqual = 1;
PMIX_INFO_CREATE(query[0].qualifiers, 1);
PMIX_INFO_LOAD(&query[0].qualifiers[0], PMIX_NSPACE, val->data.string, PMIX_STRING); // the nspace we are enquiring about
/* setup the caddy to retrieve the data */
myquery_data.info = NULL;
myquery_data.ninfo = 0;
myquery_data.active = true;
/* execute the query */
if (PMIX_SUCCESS != (rc = PMIx_Query_info_nb(query, nq, cbfunc, (void*)&myquery_data))) {
fprintf(stderr, "PMIx_Query_info failed: %d\n", rc);
goto done;
}
while (myquery_data.active) {
usleep(10);
}
fprintf(stderr, "[%s:%d] Local proctable received\n", myproc.nspace, myproc.rank);
/* now that we have the proctable for our local processes, we can do our
* magic debugger stuff and attach to them. We then send a "release" event
* to them - i.e., it's the equivalent to setting the MPIR breakpoint. We
* do this with the event notification system */
(void)strncpy(proc.nspace, val->data.string, PMIX_MAX_NSLEN);
proc.rank = PMIX_RANK_WILDCARD;
/* we send the notification to just the local procs of the job being debugged */
ninfo = 1;
PMIX_INFO_CREATE(info, ninfo);
PMIX_INFO_LOAD(&info[0], PMIX_EVENT_CUSTOM_RANGE, &proc, PMIX_PROC); // deliver to the target nspace
fprintf(stderr, "[%s:%u] Sending release\n", myproc.nspace, myproc.rank);
PMIx_Notify_event(PMIX_ERR_DEBUGGER_RELEASE,
NULL, PMIX_RANGE_LOCAL,
info, ninfo, NULL, NULL);
/* do some debugger magic */
n = 0;
fprintf(stderr, "[%s:%u] Hanging around awhile, doing debugger magic\n", myproc.nspace, myproc.rank);
while (n < 5) {
usleep(1000);
++n;
}
done:
/* finalize us */
fprintf(stderr, "Debugger daemon ns %s rank %d: Finalizing\n", myproc.nspace, myproc.rank);
if (PMIX_SUCCESS != (rc = PMIx_Finalize(NULL, 0))) {
fprintf(stderr, "Debugger daemon ns %s rank %d:PMIx_Finalize failed: %d\n", myproc.nspace, myproc.rank, rc);
} else {
fprintf(stderr, "Debugger daemon ns %s rank %d:PMIx_Finalize successfully completed\n", myproc.nspace, myproc.rank);
}
fflush(stderr);
return(0);
}
int main(int argc, char **argv)
{
int rc;
pmix_value_t value;
pmix_value_t *val = &value;
char *tmp;
pmix_proc_t proc;
uint32_t nprocs, n;
pmix_info_t *info;
bool flag;
volatile int active;
pmix_status_t dbg = PMIX_ERR_DEBUGGER_RELEASE;
/* init us - note that the call to "init" includes the return of
* any job-related info provided by the RM. This includes any
* debugger flag instructing us to stop-in-init. If such a directive
* is included, then the process will be stopped in this call until
* the "debugger release" notification arrives */
if (PMIX_SUCCESS != (rc = PMIx_Init(&myproc, NULL, 0))) {
fprintf(stderr, "Client ns %s rank %d: PMIx_Init failed: %d\n", myproc.nspace, myproc.rank, rc);
exit(0);
}
fprintf(stderr, "Client ns %s rank %d: Running\n", myproc.nspace, myproc.rank);
/* register our default event handler - again, this isn't strictly
* required, but is generally good practice */
active = -1;
PMIx_Register_event_handler(NULL, 0, NULL, 0,
notification_fn, evhandler_reg_callbk, (void*)&active);
while (-1 == active) {
sleep(1);
}
if (0 != active) {
fprintf(stderr, "[%s:%d] Default handler registration failed\n", myproc.nspace, myproc.rank);
exit(active);
}
/* job-related info is found in our nspace, assigned to the
* wildcard rank as it doesn't relate to a specific rank. Setup
* a name to retrieve such values */
PMIX_PROC_CONSTRUCT(&proc);
(void)strncpy(proc.nspace, myproc.nspace, PMIX_MAX_NSLEN);
proc.rank = PMIX_RANK_WILDCARD;
/* check to see if we have been instructed to wait for a debugger
* to attach to us. We won't get both a stop-in-init AND a
* wait-for-notify directive, so we should never stop twice. This
* directive is provided so that something like an MPI implementation
* can do some initial setup in MPI_Init prior to pausing for the
* debugger */
if (PMIX_SUCCESS == (rc = PMIx_Get(&proc, PMIX_DEBUG_WAIT_FOR_NOTIFY, NULL, 0, &val))) {
/* register for debugger release */
active = -1;
PMIx_Register_event_handler(&dbg, 1, NULL, 0,
release_fn, evhandler_reg_callbk, (void*)&active);
/* wait for registration to complete */
while (-1 == active) {
sleep(1);
}
if (0 != active) {
fprintf(stderr, "[%s:%d] Debug handler registration failed\n", myproc.nspace, myproc.rank);
exit(active);
}
/* wait for debugger release */
while (waiting_for_debugger) {
sleep(1);
}
}
/* get our universe size */
if (PMIX_SUCCESS != (rc = PMIx_Get(&proc, PMIX_UNIV_SIZE, NULL, 0, &val))) {
fprintf(stderr, "Client ns %s rank %d: PMIx_Get universe size failed: %d\n", myproc.nspace, myproc.rank, rc);
goto done;
}
fprintf(stderr, "Client %s:%d universe size %d\n", myproc.nspace, myproc.rank, val->data.uint32);
/* get the number of procs in our job - univ size is the total number of allocated
* slots, not the number of procs in the job */
if (PMIX_SUCCESS != (rc = PMIx_Get(&proc, PMIX_JOB_SIZE, NULL, 0, &val))) {
fprintf(stderr, "Client ns %s rank %d: PMIx_Get job size failed: %d\n", myproc.nspace, myproc.rank, rc);
goto done;
}
nprocs = val->data.uint32;
PMIX_VALUE_RELEASE(val);
fprintf(stderr, "Client %s:%d num procs %d\n", myproc.nspace, myproc.rank, nprocs);
/* put a few values */
if (0 > asprintf(&tmp, "%s-%d-internal", myproc.nspace, myproc.rank)) {
exit(1);
}
value.type = PMIX_UINT32;
value.data.uint32 = 1234;
if (PMIX_SUCCESS != (rc = PMIx_Store_internal(&myproc, tmp, &value))) {
fprintf(stderr, "Client ns %s rank %d: PMIx_Store_internal failed: %d\n", myproc.nspace, myproc.rank, rc);
goto done;
}
free(tmp);
if (0 > asprintf(&tmp, "%s-%d-local", myproc.nspace, myproc.rank)) {
exit(1);
}
value.type = PMIX_UINT64;
value.data.uint64 = 1234;
if (PMIX_SUCCESS != (rc = PMIx_Put(PMIX_LOCAL, tmp, &value))) {
fprintf(stderr, "Client ns %s rank %d: PMIx_Put internal failed: %d\n", myproc.nspace, myproc.rank, rc);
goto done;
}
free(tmp);
if (0 > asprintf(&tmp, "%s-%d-remote", myproc.nspace, myproc.rank)) {
exit(1);
}
value.type = PMIX_STRING;
value.data.string = "1234";
if (PMIX_SUCCESS != (rc = PMIx_Put(PMIX_REMOTE, tmp, &value))) {
fprintf(stderr, "Client ns %s rank %d: PMIx_Put internal failed: %d\n", myproc.nspace, myproc.rank, rc);
goto done;
}
free(tmp);
/* push the data to our PMIx server */
if (PMIX_SUCCESS != (rc = PMIx_Commit())) {
fprintf(stderr, "Client ns %s rank %d: PMIx_Commit failed: %d\n", myproc.nspace, myproc.rank, rc);
goto done;
}
/* call fence to synchronize with our peers - instruct
* the fence operation to collect and return all "put"
* data from our peers */
PMIX_INFO_CREATE(info, 1);
flag = true;
PMIX_INFO_LOAD(info, PMIX_COLLECT_DATA, &flag, PMIX_BOOL);
if (PMIX_SUCCESS != (rc = PMIx_Fence(&proc, 1, info, 1))) {
fprintf(stderr, "Client ns %s rank %d: PMIx_Fence failed: %d\n", myproc.nspace, myproc.rank, rc);
goto done;
}
PMIX_INFO_FREE(info, 1);
/* check the returned data */
for (n=0; n < nprocs; n++) {
if (0 > asprintf(&tmp, "%s-%d-local", myproc.nspace, myproc.rank)) {
exit(1);
}
if (PMIX_SUCCESS != (rc = PMIx_Get(&myproc, tmp, NULL, 0, &val))) {
fprintf(stderr, "Client ns %s rank %d: PMIx_Get %s failed: %d\n", myproc.nspace, myproc.rank, tmp, rc);
goto done;
}
if (PMIX_UINT64 != val->type) {
fprintf(stderr, "Client ns %s rank %d: PMIx_Get %s returned wrong type: %d\n", myproc.nspace, myproc.rank, tmp, val->type);
PMIX_VALUE_RELEASE(val);
free(tmp);
goto done;
}
if (1234 != val->data.uint64) {
fprintf(stderr, "Client ns %s rank %d: PMIx_Get %s returned wrong value: %d\n", myproc.nspace, myproc.rank, tmp, (int)val->data.uint64);
PMIX_VALUE_RELEASE(val);
free(tmp);
goto done;
}
fprintf(stderr, "Client ns %s rank %d: PMIx_Get %s returned correct\n", myproc.nspace, myproc.rank, tmp);
PMIX_VALUE_RELEASE(val);
free(tmp);
if (0 > asprintf(&tmp, "%s-%d-remote", myproc.nspace, myproc.rank)) {
exit(1);
}
if (PMIX_SUCCESS != (rc = PMIx_Get(&myproc, tmp, NULL, 0, &val))) {
fprintf(stderr, "Client ns %s rank %d: PMIx_Get %s failed: %d\n", myproc.nspace, myproc.rank, tmp, rc);
goto done;
}
if (PMIX_STRING != val->type) {
fprintf(stderr, "Client ns %s rank %d: PMIx_Get %s returned wrong type: %d\n", myproc.nspace, myproc.rank, tmp, val->type);
PMIX_VALUE_RELEASE(val);
free(tmp);
goto done;
}
if (0 != strcmp(val->data.string, "1234")) {
fprintf(stderr, "Client ns %s rank %d: PMIx_Get %s returned wrong value: %s\n", myproc.nspace, myproc.rank, tmp, val->data.string);
PMIX_VALUE_RELEASE(val);
free(tmp);
goto done;
}
fprintf(stderr, "Client ns %s rank %d: PMIx_Get %s returned correct\n", myproc.nspace, myproc.rank, tmp);
PMIX_VALUE_RELEASE(val);
free(tmp);
}
done:
/* finalize us */
fprintf(stderr, "Client ns %s rank %d: Finalizing\n", myproc.nspace, myproc.rank);
if (PMIX_SUCCESS != (rc = PMIx_Finalize(NULL, 0))) {
fprintf(stderr, "Client ns %s rank %d:PMIx_Finalize failed: %d\n", myproc.nspace, myproc.rank, rc);
} else {
fprintf(stderr, "Client ns %s rank %d:PMIx_Finalize successfully completed\n", myproc.nspace, myproc.rank);
}
fflush(stderr);
return(0);
}
int main(int argc, char **argv)
{
int rc;
pmix_value_t value;
pmix_value_t *val = &value;
char *tmp;
pmix_proc_t proc;
uint32_t nprocs, n;
int cnt, j;
bool doabort = false;
volatile bool active;
pmix_info_t info, *iptr;
size_t ninfo;
pmix_status_t code;
if (1 < argc) {
if (0 == strcmp("-abort", argv[1])) {
doabort = true;
}
}
/* init us and declare we are a test programming model */
PMIX_INFO_CREATE(iptr, 2);
PMIX_INFO_LOAD(&iptr[0], PMIX_PROGRAMMING_MODEL, "TEST", PMIX_STRING);
PMIX_INFO_LOAD(&iptr[1], PMIX_MODEL_LIBRARY_NAME, "PMIX", PMIX_STRING);
if (PMIX_SUCCESS != (rc = PMIx_Init(&myproc, iptr, 2))) {
pmix_output(0, "Client ns %s rank %d: PMIx_Init failed: %s",
myproc.nspace, myproc.rank, PMIx_Error_string(rc));
exit(rc);
}
PMIX_INFO_FREE(iptr, 2);
pmix_output(0, "Client ns %s rank %d: Running", myproc.nspace, myproc.rank);
/* test something */
(void)strncpy(proc.nspace, myproc.nspace, PMIX_MAX_NSLEN);
proc.rank = PMIX_RANK_WILDCARD;
if (PMIX_SUCCESS != (rc = PMIx_Get(&proc, PMIX_JOB_SIZE, NULL, 0, &val))) {
pmix_output(0, "Client ns %s rank %d: PMIx_Get failed: %s",
myproc.nspace, myproc.rank, PMIx_Error_string(rc));
exit(rc);
}
PMIX_VALUE_RELEASE(val);
/* test something */
if (PMIX_SUCCESS != (rc = PMIx_Get(&myproc, PMIX_SERVER_URI, NULL, 0, &val))) {
pmix_output(0, "Client ns %s rank %d: PMIx_Get failed: %s",
myproc.nspace, myproc.rank, PMIx_Error_string(rc));
exit(rc);
}
pmix_output(0, "CLIENT SERVER URI: %s", val->data.string);
PMIX_VALUE_RELEASE(val);
/* register a handler specifically for when models declare */
active = true;
ninfo = 1;
PMIX_INFO_CREATE(iptr, ninfo);
PMIX_INFO_LOAD(&iptr[0], PMIX_EVENT_HDLR_NAME, "SIMPCLIENT-MODEL", PMIX_STRING);
code = PMIX_MODEL_DECLARED;
PMIx_Register_event_handler(&code, 1, iptr, ninfo,
model_callback, model_registration_callback, (void*)&active);
while (active) {
usleep(10);
}
PMIX_INFO_FREE(iptr, ninfo);
/* register our errhandler */
active = true;
PMIx_Register_event_handler(NULL, 0, NULL, 0,
notification_fn, errhandler_reg_callbk, (void*)&active);
while (active) {
usleep(10);
}
/* get our universe size */
(void)strncpy(proc.nspace, myproc.nspace, PMIX_MAX_NSLEN);
proc.rank = PMIX_RANK_WILDCARD;
if (PMIX_SUCCESS != (rc = PMIx_Get(&proc, PMIX_UNIV_SIZE, NULL, 0, &val))) {
pmix_output(0, "Client ns %s rank %d: PMIx_Get universe size failed: %s",
myproc.nspace, myproc.rank, PMIx_Error_string(rc));
goto done;
}
nprocs = val->data.uint32;
PMIX_VALUE_RELEASE(val);
pmix_output(0, "Client %s:%d universe size %d", myproc.nspace, myproc.rank, nprocs);
/* put a few values */
(void)asprintf(&tmp, "%s-%d-internal", myproc.nspace, myproc.rank);
value.type = PMIX_UINT32;
value.data.uint32 = 1234;
if (PMIX_SUCCESS != (rc = PMIx_Store_internal(&myproc, tmp, &value))) {
pmix_output(0, "Client ns %s rank %d: PMIx_Store_internal failed: %s",
myproc.nspace, myproc.rank, PMIx_Error_string(rc));
goto done;
}
for (cnt=0; cnt < MAXCNT; cnt++) {
(void)asprintf(&tmp, "%s-%d-local-%d", myproc.nspace, myproc.rank, cnt);
value.type = PMIX_UINT64;
value.data.uint64 = 1234;
if (PMIX_SUCCESS != (rc = PMIx_Put(PMIX_LOCAL, tmp, &value))) {
pmix_output(0, "Client ns %s rank %d: PMIx_Put internal failed: %s",
myproc.nspace, myproc.rank, PMIx_Error_string(rc));
goto done;
}
(void)asprintf(&tmp, "%s-%d-remote-%d", myproc.nspace, myproc.rank, cnt);
value.type = PMIX_STRING;
value.data.string = "1234";
if (PMIX_SUCCESS != (rc = PMIx_Put(PMIX_REMOTE, tmp, &value))) {
pmix_output(0, "Client ns %s rank %d: PMIx_Put internal failed: %s",
myproc.nspace, myproc.rank, PMIx_Error_string(rc));
goto done;
}
if (PMIX_SUCCESS != (rc = PMIx_Commit())) {
pmix_output(0, "Client ns %s rank %d cnt %d: PMIx_Commit failed: %s",
myproc.nspace, myproc.rank, cnt, PMIx_Error_string(rc));
goto done;
}
/* call fence to ensure the data is received */
PMIX_PROC_CONSTRUCT(&proc);
(void)strncpy(proc.nspace, myproc.nspace, PMIX_MAX_NSLEN);
proc.rank = PMIX_RANK_WILDCARD;
if (PMIX_SUCCESS != (rc = PMIx_Fence(&proc, 1, NULL, 0))) {
pmix_output(0, "Client ns %s rank %d cnt %d: PMIx_Fence failed: %s",
myproc.nspace, myproc.rank, cnt, PMIx_Error_string(rc));
goto done;
}
/* check the returned data */
(void)strncpy(proc.nspace, myproc.nspace, PMIX_MAX_NSLEN);
for (j=0; j <= cnt; j++) {
for (n=0; n < nprocs; n++) {
proc.rank = n;
(void)asprintf(&tmp, "%s-%d-local-%d", myproc.nspace, n, j);
if (PMIX_SUCCESS != (rc = PMIx_Get(&proc, tmp, NULL, 0, &val))) {
pmix_output(0, "Client ns %s rank %d cnt %d: PMIx_Get %s failed: %s",
myproc.nspace, myproc.rank, j, tmp, PMIx_Error_string(rc));
continue;
}
if (NULL == val) {
pmix_output(0, "Client ns %s rank %d: NULL value returned",
myproc.nspace, myproc.rank);
break;
}
if (PMIX_UINT64 != val->type) {
pmix_output(0, "Client ns %s rank %d cnt %d: PMIx_Get %s returned wrong type: %d", myproc.nspace, myproc.rank, j, tmp, val->type);
PMIX_VALUE_RELEASE(val);
free(tmp);
continue;
}
if (1234 != val->data.uint64) {
pmix_output(0, "Client ns %s rank %d cnt %d: PMIx_Get %s returned wrong value: %d", myproc.nspace, myproc.rank, j, tmp, (int)val->data.uint64);
PMIX_VALUE_RELEASE(val);
free(tmp);
continue;
}
pmix_output(0, "Client ns %s rank %d cnt %d: PMIx_Get %s returned correct", myproc.nspace, myproc.rank, j, tmp);
PMIX_VALUE_RELEASE(val);
free(tmp);
if (n != myproc.rank) {
(void)asprintf(&tmp, "%s-%d-remote-%d", proc.nspace, n, j);
if (PMIX_SUCCESS != (rc = PMIx_Get(&proc, tmp, NULL, 0, &val))) {
/* this data should _not_ be found as we are on the same node
* and the data was "put" with a PMIX_REMOTE scope */
pmix_output(0, "Client ns %s rank %d cnt %d: PMIx_Get %s returned correct", myproc.nspace, myproc.rank, j, tmp);
continue;
}
pmix_output(0, "Client ns %s rank %d cnt %d: PMIx_Get %s returned remote data for a local proc",
myproc.nspace, myproc.rank, j, tmp);
PMIX_VALUE_RELEASE(val);
free(tmp);
}
}
}
}
/* now get the data blob for myself */
pmix_output(0, "Client ns %s rank %d testing internal modex blob",
myproc.nspace, myproc.rank);
if (PMIX_SUCCESS == (rc = PMIx_Get(&myproc, NULL, NULL, 0, &val))) {
if (PMIX_DATA_ARRAY != val->type) {
pmix_output(0, "Client ns %s rank %d did not return an array for its internal modex blob",
myproc.nspace, myproc.rank);
PMIX_VALUE_RELEASE(val);
} else if (PMIX_INFO != val->data.darray->type) {
pmix_output(0, "Client ns %s rank %d returned an internal modex array of type %s instead of PMIX_INFO",
myproc.nspace, myproc.rank, PMIx_Data_type_string(val->data.darray->type));
PMIX_VALUE_RELEASE(val);
} else if (0 == val->data.darray->size) {
pmix_output(0, "Client ns %s rank %d returned an internal modex array of zero length",
myproc.nspace, myproc.rank);
PMIX_VALUE_RELEASE(val);
} else {
pmix_info_t *iptr = (pmix_info_t*)val->data.darray->array;
for (n=0; n < val->data.darray->size; n++) {
pmix_output(0, "\tKey: %s", iptr[n].key);
}
PMIX_VALUE_RELEASE(val);
}
} else {
pmix_output(0, "Client ns %s rank %d internal modex blob FAILED with error %s(%d)",
myproc.nspace, myproc.rank, PMIx_Error_string(rc), rc);
}
/* log something */
PMIX_INFO_CONSTRUCT(&info);
PMIX_INFO_LOAD(&info, PMIX_LOG_STDERR, "test log msg", PMIX_STRING);
active = true;
rc = PMIx_Log_nb(&info, 1, NULL, 0, opcbfunc, (void*)&active);
if (PMIX_SUCCESS != rc) {
pmix_output(0, "Client ns %s rank %d - log_nb returned %s",
myproc.nspace, myproc.rank, PMIx_Error_string(rc));
} else {
while (active) {
usleep(10);
}
}
PMIX_INFO_DESTRUCT(&info);
/* if requested and our rank is 0, call abort */
if (doabort) {
if (0 == myproc.rank) {
PMIx_Abort(PMIX_ERR_PROC_REQUESTED_ABORT, "CALLING ABORT", NULL, 0);
} else {
while(!completed) {
usleep(10);
}
}
}
done:
/* finalize us */
pmix_output(0, "Client ns %s rank %d: Finalizing", myproc.nspace, myproc.rank);
if (PMIX_SUCCESS != (rc = PMIx_Finalize(NULL, 0))) {
fprintf(stderr, "Client ns %s rank %d:PMIx_Finalize failed: %s\n",
myproc.nspace, myproc.rank, PMIx_Error_string(rc));
} else {
fprintf(stderr, "Client ns %s rank %d:PMIx_Finalize successfully completed\n", myproc.nspace, myproc.rank);
}
fflush(stderr);
return(rc);
}
int main(int argc, char **argv)
{
char **client_env=NULL;
char **client_argv=NULL;
int rc;
struct stat stat_buf;
struct timeval tv;
double test_start;
cli_state_t order[CLI_TERM+1];
test_params params;
INIT_TEST_PARAMS(params);
int test_fail = 0;
char *tmp;
int ns_nprocs;
gettimeofday(&tv, NULL);
test_start = tv.tv_sec + 1E-6*tv.tv_usec;
/* smoke test */
if (PMIX_SUCCESS != 0) {
TEST_ERROR(("ERROR IN COMPUTING CONSTANTS: PMIX_SUCCESS = %d", PMIX_SUCCESS));
exit(1);
}
TEST_VERBOSE(("Testing version %s", PMIx_Get_version()));
parse_cmd(argc, argv, ¶ms);
TEST_VERBOSE(("Start PMIx_lite smoke test (timeout is %d)", params.timeout));
/* verify executable */
if( 0 > ( rc = stat(params.binary, &stat_buf) ) ){
TEST_ERROR(("Cannot stat() executable \"%s\": %d: %s", params.binary, errno, strerror(errno)));
FREE_TEST_PARAMS(params);
return 0;
} else if( !S_ISREG(stat_buf.st_mode) ){
TEST_ERROR(("Client executable \"%s\": is not a regular file", params.binary));
FREE_TEST_PARAMS(params);
return 0;
}else if( !(stat_buf.st_mode & S_IXUSR) ){
TEST_ERROR(("Client executable \"%s\": has no executable flag", params.binary));
FREE_TEST_PARAMS(params);
return 0;
}
/* setup the server library */
pmix_info_t info[1];
(void)strncpy(info[0].key, PMIX_SOCKET_MODE, PMIX_MAX_KEYLEN);
info[0].value.type = PMIX_UINT32;
info[0].value.data.uint32 = 0666;
if (PMIX_SUCCESS != (rc = PMIx_server_init(&mymodule, info, 1))) {
TEST_ERROR(("Init failed with error %d", rc));
FREE_TEST_PARAMS(params);
return rc;
}
/* register the errhandler */
PMIx_Register_event_handler(NULL, 0, NULL, 0,
errhandler, errhandler_reg_callbk, NULL);
order[CLI_UNINIT] = CLI_FORKED;
order[CLI_FORKED] = CLI_FIN;
order[CLI_CONNECTED] = CLI_UNDEF;
order[CLI_FIN] = CLI_TERM;
order[CLI_DISCONN] = CLI_UNDEF;
order[CLI_TERM] = CLI_UNDEF;
cli_init(params.nprocs, order);
/* set common argv and env */
client_env = pmix_argv_copy(environ);
set_client_argv(¶ms, &client_argv);
tmp = pmix_argv_join(client_argv, ' ');
TEST_VERBOSE(("Executing test: %s", tmp));
free(tmp);
int launched = 0;
/* set namespaces and fork clients */
if (NULL == params.ns_dist) {
/* we have a single namespace for all clients */
ns_nprocs = params.nprocs;
rc = launch_clients(ns_nprocs, params.binary, &client_env, &client_argv);
if (PMIX_SUCCESS != rc) {
FREE_TEST_PARAMS(params);
return rc;
}
launched += ns_nprocs;
} else {
char *pch;
pch = strtok(params.ns_dist, ":");
while (NULL != pch) {
ns_nprocs = (int)strtol(pch, NULL, 10);
if (params.nprocs < (uint32_t)(launched+ns_nprocs)) {
TEST_ERROR(("Total number of processes doesn't correspond number specified by ns_dist parameter."));
FREE_TEST_PARAMS(params);
return PMIX_ERROR;
}
if (0 < ns_nprocs) {
rc = launch_clients(ns_nprocs, params.binary, &client_env, &client_argv);
if (PMIX_SUCCESS != rc) {
FREE_TEST_PARAMS(params);
return rc;
}
}
pch = strtok (NULL, ":");
launched += ns_nprocs;
}
}
if (params.nprocs != (uint32_t)launched) {
TEST_ERROR(("Total number of processes doesn't correspond number specified by ns_dist parameter."));
cli_kill_all();
test_fail = 1;
}
/* hang around until the client(s) finalize */
while (!test_terminated()) {
// To avoid test hang we want to interrupt the loop each 0.1s
double test_current;
// check if we exceed the max time
gettimeofday(&tv, NULL);
test_current = tv.tv_sec + 1E-6*tv.tv_usec;
if( (test_current - test_start) > params.timeout ){
break;
}
cli_wait_all(0);
}
if( !test_terminated() ){
TEST_ERROR(("Test exited by a timeout!"));
cli_kill_all();
test_fail = 1;
}
if( test_abort ){
TEST_ERROR(("Test was aborted!"));
/* do not simply kill the clients as that generates
* event notifications which these tests then print
* out, flooding the log */
// cli_kill_all();
test_fail = 1;
}
if (0 != params.test_spawn) {
PMIX_WAIT_FOR_COMPLETION(spawn_wait);
}
pmix_argv_free(client_argv);
pmix_argv_free(client_env);
/* deregister the errhandler */
PMIx_Deregister_event_handler(0, op_callbk, NULL);
cli_wait_all(1.0);
/* finalize the server library */
if (PMIX_SUCCESS != (rc = PMIx_server_finalize())) {
TEST_ERROR(("Finalize failed with error %d", rc));
}
FREE_TEST_PARAMS(params);
if (0 == test_fail) {
TEST_OUTPUT(("Test finished OK!"));
}
return test_fail;
}
int main(int argc, char **argv)
{
pmix_status_t rc;
pmix_value_t value;
pmix_value_t *val = &value;
pmix_proc_t proc;
uint32_t nprocs, n;
pmix_info_t *info, *iptr;
bool flag;
mylock_t mylock;
pmix_data_array_t *dptr;
/* init us - note that the call to "init" includes the return of
* any job-related info provided by the RM. */
if (PMIX_SUCCESS != (rc = PMIx_Init(&myproc, NULL, 0))) {
fprintf(stderr, "Client ns %s rank %d: PMIx_Init failed: %d\n", myproc.nspace, myproc.rank, rc);
exit(0);
}
fprintf(stderr, "Client ns %s rank %d: Running\n", myproc.nspace, myproc.rank);
/* register our default event handler - again, this isn't strictly
* required, but is generally good practice */
DEBUG_CONSTRUCT_LOCK(&mylock);
PMIx_Register_event_handler(NULL, 0, NULL, 0,
notification_fn, evhandler_reg_callbk, (void*)&mylock);
/* wait for registration to complete */
DEBUG_WAIT_THREAD(&mylock);
rc = mylock.status;
DEBUG_DESTRUCT_LOCK(&mylock);
if (PMIX_SUCCESS != rc) {
fprintf(stderr, "[%s:%d] Default handler registration failed\n", myproc.nspace, myproc.rank);
goto done;
}
/* job-related info is found in our nspace, assigned to the
* wildcard rank as it doesn't relate to a specific rank. Setup
* a name to retrieve such values */
PMIX_PROC_CONSTRUCT(&proc);
(void)strncpy(proc.nspace, myproc.nspace, PMIX_MAX_NSLEN);
proc.rank = PMIX_RANK_WILDCARD;
/* get our universe size */
if (PMIX_SUCCESS != (rc = PMIx_Get(&proc, PMIX_UNIV_SIZE, NULL, 0, &val))) {
fprintf(stderr, "Client ns %s rank %d: PMIx_Get universe size failed: %d\n", myproc.nspace, myproc.rank, rc);
goto done;
}
nprocs = val->data.uint32;
PMIX_VALUE_RELEASE(val);
fprintf(stderr, "Client %s:%d universe size %d\n", myproc.nspace, myproc.rank, nprocs);
/* inform the RM that we are preemptible, and that our checkpoint methods are
* "signal" on SIGUSR2 and event on PMIX_JCTRL_CHECKPOINT */
PMIX_INFO_CREATE(info, 2);
flag = true;
PMIX_INFO_LOAD(&info[0], PMIX_JOB_CTRL_PREEMPTIBLE, (void*)&flag, PMIX_BOOL);
/* can't use "load" to load a pmix_data_array_t */
(void)strncpy(info[1].key, PMIX_JOB_CTRL_CHECKPOINT_METHOD, PMIX_MAX_KEYLEN);
PMIX_DATA_ARRAY_CREATE(info[1].value.data.darray, 2, PMIX_INFO);
dptr = info[1].value.data.darray;
rc = SIGUSR2;
iptr = (pmix_info_t*)dptr->array;
PMIX_INFO_LOAD(&iptr[0], PMIX_JOB_CTRL_CHECKPOINT_SIGNAL, &rc, PMIX_INT);
rc = PMIX_JCTRL_CHECKPOINT;
PMIX_INFO_LOAD(&iptr[1], PMIX_JOB_CTRL_CHECKPOINT_EVENT, &rc, PMIX_STATUS);
/* since this is informational and not a requested operation, the target parameter
* doesn't mean anything and can be ignored */
DEBUG_CONSTRUCT_LOCK(&mylock);
if (PMIX_SUCCESS != (rc = PMIx_Job_control_nb(NULL, 0, info, 2, infocbfunc, (void*)&mylock))) {
fprintf(stderr, "Client ns %s rank %d: PMIx_Job_control_nb failed: %d\n", myproc.nspace, myproc.rank, rc);
DEBUG_DESTRUCT_LOCK(&mylock);
goto done;
}
DEBUG_WAIT_THREAD(&mylock);
PMIX_INFO_FREE(info, 2);
rc = mylock.status;
DEBUG_DESTRUCT_LOCK(&mylock);
if (PMIX_SUCCESS != rc) {
fprintf(stderr, "Client ns %s rank %d: PMIx_Job_control_nb failed: %d\n", myproc.nspace, myproc.rank, rc);
goto done;
}
/* now request that this process be monitored using heartbeats */
PMIX_INFO_CREATE(iptr, 1);
PMIX_INFO_LOAD(&iptr[0], PMIX_MONITOR_HEARTBEAT, NULL, PMIX_POINTER);
PMIX_INFO_CREATE(info, 3);
PMIX_INFO_LOAD(&info[0], PMIX_MONITOR_ID, "MONITOR1", PMIX_STRING);
n = 5; // require a heartbeat every 5 seconds
PMIX_INFO_LOAD(&info[1], PMIX_MONITOR_HEARTBEAT_TIME, &n, PMIX_UINT32);
n = 2; // two heartbeats can be missed before declaring us "stalled"
PMIX_INFO_LOAD(&info[2], PMIX_MONITOR_HEARTBEAT_DROPS, &n, PMIX_UINT32);
/* make the request */
DEBUG_CONSTRUCT_LOCK(&mylock);
if (PMIX_SUCCESS != (rc = PMIx_Process_monitor_nb(iptr, PMIX_MONITOR_HEARTBEAT_ALERT,
info, 3, infocbfunc, (void*)&mylock))) {
fprintf(stderr, "Client ns %s rank %d: PMIx_Process_monitor_nb failed: %d\n", myproc.nspace, myproc.rank, rc);
DEBUG_DESTRUCT_LOCK(&mylock);
goto done;
}
DEBUG_WAIT_THREAD(&mylock);
PMIX_INFO_FREE(iptr, 1);
PMIX_INFO_FREE(info, 3);
rc = mylock.status;
DEBUG_DESTRUCT_LOCK(&mylock);
if (PMIX_SUCCESS != rc) {
fprintf(stderr, "Client ns %s rank %d: PMIx_Process_monitor_nb failed: %d\n", myproc.nspace, myproc.rank, rc);
goto done;
}
/* send a heartbeat */
PMIx_Heartbeat();
/* call fence to synchronize with our peers - no need to
* collect any info as we didn't "put" anything */
PMIX_INFO_CREATE(info, 1);
flag = false;
PMIX_INFO_LOAD(info, PMIX_COLLECT_DATA, &flag, PMIX_BOOL);
if (PMIX_SUCCESS != (rc = PMIx_Fence(&proc, 1, info, 1))) {
fprintf(stderr, "Client ns %s rank %d: PMIx_Fence failed: %d\n", myproc.nspace, myproc.rank, rc);
goto done;
}
PMIX_INFO_FREE(info, 1);
done:
/* finalize us */
fprintf(stderr, "Client ns %s rank %d: Finalizing\n", myproc.nspace, myproc.rank);
if (PMIX_SUCCESS != (rc = PMIx_Finalize(NULL, 0))) {
fprintf(stderr, "Client ns %s rank %d:PMIx_Finalize failed: %d\n", myproc.nspace, myproc.rank, rc);
} else {
fprintf(stderr, "Client ns %s rank %d:PMIx_Finalize successfully completed\n", myproc.nspace, myproc.rank);
}
fflush(stderr);
return(0);
}
int main(int argc, char **argv)
{
pmix_status_t rc;
pmix_info_t *info, *iptr;
pmix_app_t *app;
size_t ninfo, napps;
char *nspace = NULL;
int i;
pmix_query_t *query;
size_t nq, n;
myquery_data_t myquery_data;
bool cospawn = false, stop_on_exec = false, cospawn_reqd = false;
char cwd[1024];
pmix_status_t code = PMIX_ERR_JOB_TERMINATED;
mylock_t mylock;
myrel_t myrel, launcher_ready, dbrel;
pid_t pid;
pmix_envar_t envar;
char *launchers[] = {
"prun",
"mpirun",
"mpiexec",
"orterun",
NULL
};
pmix_proc_t proc;
bool found;
pmix_data_array_t darray;
char *tmp;
char clientspace[PMIX_MAX_NSLEN+1];
pid = getpid();
/* Process any arguments we were given */
for (i=1; i < argc; i++) {
if (0 == strcmp(argv[i], "-h") ||
0 == strcmp(argv[i], "--help")) {
/* print the usage message and exit */
}
if (0 == strcmp(argv[i], "-a") ||
0 == strcmp(argv[i], "--attach")) {
if (NULL != nspace) {
/* can only support one */
fprintf(stderr, "Cannot attach to more than one nspace\n");
exit(1);
}
/* the next argument must be the nspace */
++i;
if (argc == i) {
/* they goofed */
fprintf(stderr, "The %s option requires an <nspace> argument\n", argv[i]);
exit(1);
}
nspace = strdup(argv[i]);
} else if (0 == strcmp(argv[i], "-c") ||
0 == strcmp(argv[i], "--cospawn")){
cospawn_reqd = true;
break;
}
}
info = NULL;
ninfo = 0;
/* use the system connection first, if available */
PMIX_INFO_CREATE(info, 1);
PMIX_INFO_LOAD(&info[0], PMIX_CONNECT_SYSTEM_FIRST, NULL, PMIX_BOOL);
/* init as a tool */
if (PMIX_SUCCESS != (rc = PMIx_tool_init(&myproc, info, ninfo))) {
fprintf(stderr, "PMIx_tool_init failed: %s(%d)\n", PMIx_Error_string(rc), rc);
exit(rc);
}
PMIX_INFO_FREE(info, ninfo);
fprintf(stderr, "Debugger ns %s rank %d pid %lu: Running\n", myproc.nspace, myproc.rank, (unsigned long)pid);
/* construct the debugger termination release */
DEBUG_CONSTRUCT_LOCK(&dbrel.lock);
/* register a default event handler */
DEBUG_CONSTRUCT_LOCK(&mylock);
PMIx_Register_event_handler(NULL, 0, NULL, 0,
notification_fn, evhandler_reg_callbk, (void*)&mylock);
DEBUG_WAIT_THREAD(&mylock);
DEBUG_DESTRUCT_LOCK(&mylock);
/* if we are attaching to a running job, then attach to it */
if (NULL != nspace) {
if (PMIX_SUCCESS != (rc = attach_to_running_job(nspace))) {
fprintf(stderr, "Failed to attach to nspace %s: error code %d\n",
nspace, rc);
goto done;
}
}
done:
DEBUG_DESTRUCT_LOCK(&myrel.lock);
DEBUG_DESTRUCT_LOCK(&dbrel.lock);
PMIx_tool_finalize();
return(rc);
}
int main(int argc, char **argv, const char **environ)
{
pmix_status_t rc;
pmix_info_t *info = NULL;
bool flag;
pmix_status_t retval;
pmix_app_t *spawned_app = NULL;
pmix_info_t *job_info = NULL;
pmix_info_t *proc_info = NULL;
int job_info_count = 0;
int job_info_index = 0;
int proc_info_count = 0;
int proc_info_index = 0;
char spawned_nsp[PMIX_MAX_NSLEN+1];
char *path_to_app = NULL;
char *host_to_use = NULL;
int number_of_clients = 0;
int temp_counter = 0;
done_flag = false;
gethostname(hostn, 500);
int spawned_app_argc = 0;
char **scr_environ = NULL;
int proc_count = 1;
int node_count = 0;
bool blocking_mode = true;
char *node_list = NULL;
bool forward_all_scr_envs = false;
bool pmix_mode = false;
const char *optstring = "+n:N:L:x:bB:pPvhe";
int temp_slen=0;
/* todo: add arg parsing with ompi schizo */
verbose_print = false;
int sleep_max = 30;
const int fixed_sleep = 5;
int c;
while((c = getopt(argc, argv, optstring)) != -1){
switch(c){
case 'h':
print_usage(argv[0]);
exit(0);
break;
case 'n':
proc_count = atoi(optarg);
if(proc_count <= 0 || proc_count > 100){
printf("outside the range of allowable instances to spawn [1-100]\n");
exit(1);
}
if(verbose_print) {
printf("proc_count = %d\n", proc_count);
}
break;
case 'N':
/* node_count = atoi(optarg); */
node_count = 1;
if(verbose_print) {
printf("node_count = %d\n", node_count);
}
break;
case 'B':
blocking_mode = true;
sleep_max = atoi(optarg);
if(sleep_max < 0){
printf("can't sleep for less than 0 seconds\n");
exit(1);
}
if(verbose_print){
printf("blocking mode = %x\n", blocking_mode);
}
break;
case 'b':
blocking_mode = false;
if(verbose_print){
printf("blocking mode = %x\n", blocking_mode);
}
break;
case 'L':
node_list = optarg;
host_to_use = node_list;
if(verbose_print){
printf("node_list = '%s'\n", node_list);
}
break;
case 'x':
temp_slen = strlen(optarg);
/* check if the string is the same length as 'SCR', if so compare them */
if(temp_slen == strlen(SCR_STRING)){
if(strncmp(optarg, SCR_STRING, strlen(SCR_STRING)) == 0){
/* if the string is SCR, then forward all SCR related env vars */
if(verbose_print) printf("all scr envs will be forwarded\n");
forward_all_scr_envs = true;
}
else{
/* handled like a normal env var */
handle_standard_env_var(optarg, &scr_environ);
}
}
else{
/*handled like a normal env var */
handle_standard_env_var(optarg, &scr_environ);
}
break;
case 'v':
verbose_print = true;
break;
case 'p':
pmix_mode = true;
if(verbose_print){
printf("pmix_mode = %x\n", pmix_mode);
}
break;
case 'P':
pmix_mode = false;
if(verbose_print){
printf("pmix_mode = %x\n", pmix_mode);
}
break;
case 'e':
experimental = true;
break;
case '?':
printf("missing a required argument or invalid option: %x\n", optopt);
print_usage(argv[0]);
exit(1);
break;
default:
printf("Unrecognized argument: %c\n", c);
print_usage(argv[0]);
exit(1);
break;
}
}
/* number of instances to spawn */
number_of_clients = proc_count;
/* check to make sure an application was specified to launch */
if( optind < argc ){
/* if optind is < argc, it means there is at least one more arg
* beyond the args for this program */
path_to_app = argv[optind];
spawned_app_argc = argc - optind;
if(verbose_print) {
printf("app to launch: %s @ %s:%d\n",
path_to_app, __FILE__, __LINE__);
}
}
else{
printf("program_to_spawn option was not provded\n");
print_usage(argv[0]);
exit(1);
}
if(verbose_print){
printf("master process will spawn %d instances; app to run: %s\n\n",
number_of_clients, path_to_app);
printf("pmix version: %s (host: %s)\n", PMIx_Get_version(), hostn);
}
/* init pmix */
retval = PMIx_Init(&main_proc, NULL, 0);
if(retval != PMIX_SUCCESS){
error_helper(retval, hostn, "error initializing pmix");
exit(0);
}
if(verbose_print){
printf("rank %d, host '%s', nspace: '%s' init'd pmix succesfully\n\n",
main_proc.rank, hostn, main_proc.nspace);
}
/* we need to attach to a "system" PMIx server so we
* can ask it to spawn applications for us. There can
* only be one such connection on a node, so we will
* instruct the tool library to only look for it */
int ninfo = 1;
PMIX_INFO_CREATE(info, ninfo);
flag = true;
PMIX_INFO_LOAD(&info[0], PMIX_CONNECT_TO_SYSTEM, &flag, PMIX_BOOL);
/* initialize the library and make the connection */
if (PMIX_SUCCESS != (rc = PMIx_tool_init(&tool_proc, NULL, 0 ))) {
fprintf(stderr, "PMIx_tool_init failed: %d\n", rc );
exit(rc);
}
if (0 < ninfo) {
PMIX_INFO_FREE(info, ninfo);
}
/* first call fence to sync all processes */
retval = fence_helper();
if(retval != PMIX_SUCCESS)
{
error_helper(retval, hostn, "error fencing");
exit(retval);
}
/* Process SCR env vars if needed */
if(forward_all_scr_envs){
parse_all_scr_envs(&scr_environ, environ);
}
/* finalize the env array so a NULL is in place */
finalize_array(scr_environ);
/* Setup info structs to pass to this: */
/* pmix_info_t *error_info = NULL; */
/* PMIX_INFO_CREATE(error_info, 1); */
/*
strncpy(error_info[0].key, PMIX_ERROR_GROUP_ABORT, PMIX_MAX_KEYLEN);
error_info[0].value.type = PMIX_BOOL;
error_info[0].value.data.flag = true;
*/
/* strncpy(error_info[0].key, PMIX_ERROR_GROUP_SPAWN, PMIX_MAX_KEYLEN);
int t_val = 1;
pmix_value_load(&error_info[1].value, &t_val, PMIX_BOOL);
*/
/*error_info[1].value.type = PMIX_BOOL;
error_info[1].value.data.flag = true; */
/* strncpy(error_info[2].key, PMIX_ERROR_GROUP_GENERAL, PMIX_MAX_KEYLEN);
error_info[2].value.type = PMIX_BOOL;
error_info[2].value.data.flag = true;
*/
/* TODO: setup error handling when implemented in pmix with the
* following error codes: */
/*
pmix_status_t registered_codes[5];
registered_codes[0] = PMIX_ERR_JOB_TERMINATED;
registered_codes[1] = PMIX_ERR_PROC_ABORTED;
registered_codes[2] = PMIX_ERR_PROC_ABORTING;
*/
PMIx_Register_event_handler(NULL, 0,
NULL, 0,
errhandler_cb,
errhandler_reg_callbk,
(void *) NULL);
/* PMIX_INFO_DESTRUCT(error_info); */
/* allocate memory to hold the spawend app struct */
PMIX_APP_CREATE(spawned_app, 1);
/* maxprocs isn't documented very well, but it appears to control
* how many instances of the spanwed app are created */
spawned_app->maxprocs = number_of_clients;
/* set the app to run */
(void)asprintf(&spawned_app->cmd, "%s", path_to_app);
/* set argv for spawned app starting with remaining argv */
spawned_app->argv = &argv[optind];
/* set the environment pointer */
spawned_app->env = scr_environ;
/*--START: add all proc level infos */
/* add things to the proc level info */
if(!pmix_mode){
job_info_count++;
}
if(host_to_use != NULL){
proc_info_count++;
}
if(verbose_print){
printf("enabling debug feature for forwarding stdout/stderr\n");
proc_info_count+=2;
/* add PMIX_FWD_STDOUT and PMIX_FWD_STDERR later*/
}
if(experimental){
job_info_count++;
}
if(node_count == 1){
job_info_count++;
}
/*--END: add all proc level infos */
/*--START: append actual proc level info */
PMIX_INFO_CREATE(job_info, job_info_count);
PMIX_INFO_CREATE(proc_info, proc_info_count);
/* PMIX_VAL_set_assign(_v, _field, _val ) */
/* PMIX_VAL_set_strdup(_v, _field, _val ) */
if(host_to_use != NULL){
/* add info struct to the spawned app itself for the host */
/* old way */
strncpy(proc_info[proc_info_index].key, PMIX_HOST, PMIX_MAX_KEYLEN);
//proc_info[proc_info_index].value.type = PMIX_STRING;
/* set the data for host list to use */
//proc_info[proc_info_index].value.data.string = host_to_use;
/* end old way */
if(verbose_print) printf("about to set host val\n");
PMIX_VAL_SET(&(proc_info[proc_info_index].value), string,
host_to_use );
proc_info_index++;
}
if(!pmix_mode){
strncpy(job_info[job_info_index].key, PMIX_NON_PMI,
PMIX_MAX_KEYLEN);
if(verbose_print) printf("about to set non pmix flag\n");
PMIX_VAL_SET(&(job_info[job_info_index].value), flag, true);
job_info_index++;
}
if(verbose_print){
strncpy(proc_info[proc_info_index].key, PMIX_FWD_STDOUT,
PMIX_MAX_KEYLEN);
if(verbose_print) printf("about to set stdout flag\n");
PMIX_VAL_SET(&(proc_info[proc_info_index].value), flag, true );
proc_info_index++;
strncpy(proc_info[proc_info_index].key, PMIX_FWD_STDERR,
PMIX_MAX_KEYLEN);
if(verbose_print) printf("about to set stderr flag\n");
PMIX_VAL_SET(&(proc_info[proc_info_index].value), flag, true );
proc_info_index++;
}
if(experimental){
printf("attempting to perform experiment\n");
bool local_flag = true;
PMIX_INFO_LOAD(&job_info[job_info_index], PMIX_NOTIFY_COMPLETION, &local_flag, PMIX_BOOL);
job_info_index++;
}
if(node_count == 1){
strncpy(job_info[job_info_index].key, PMIX_PPR,
PMIX_MAX_KEYLEN);
PMIX_VAL_SET(&(job_info[job_info_index].value), string,
"1:n");
job_info_index++;
}
/*--END: append actual proc level info */
/* sanity check to make sure we covered all the info structs */
if(proc_info_index != proc_info_count ){
printf("bug: mismatch with appending proc info\n");
exit(1);
}
if(job_info_index != job_info_count){
printf("bug: mismatch with appending job info\n");
exit(1);
}
/* TODO: TEST PMIX_NOTIFY_COMPLETION WHEN IT'S IMPLEMENTED IN PMIX */
/* fill in job_info */
/*
strncpy(job_info[0].key, PMIX_TIMEOUT, PMIX_MAX_KEYLEN);
job_info[0].value.type = PMIX_INT;
job_info[0].value.data.integer = 10; */
/* strncpy(job_info[0].key, PMIX_NOTIFY_COMPLETION, PMIX_MAX_KEYLEN);
job_info[0].value.type = PMIX_BOOL;
job_info[0].value.data.flag = true; */
/*strncpy(spawned_app->info[0].key, PMIX_DISPLAY_MAP, PMIX_MAX_KEYLEN);
job_info[0].value.type = PMIX_BOOL;
job_info[0].value.data.flag = true;*/
/* TODO: TEST PMIX_NOTIFY_COMPLETION WHEN IT'S IMPLEMENTED IN PMIX */
spawned_app->info = proc_info;
spawned_app->ninfo = proc_info_count;
if(verbose_print){
printf("proc level info count: %d\n", proc_info_count);
}
/* call spawn */
retval = PMIx_Spawn(job_info, job_info_count,
spawned_app, 1, spawned_nsp);
if(verbose_print) {
printf("rank %d (host %s) just called spawn; spawned nspace: %s, retval:%d\n",
main_proc.rank,
hostn,
spawned_nsp,
retval);
}
if(retval != PMIX_SUCCESS){
error_helper(retval, hostn, "error with spawn");
goto done;
}
/* TODO: TEMPORARY WORKAROUND TO WAIT FOR A SPAWNED PROCESS */
if(blocking_mode){
sleep(fixed_sleep);
/* wait until app completes: */
while(!done_flag){
sleep(fixed_sleep);
temp_counter++;
if(temp_counter*fixed_sleep >= sleep_max) {
if(verbose_print) printf("broke out early\n");
break;
}
}
if(verbose_print){
if(done_flag == true) {
printf("done_flag was set to true!\n");
}
}
}
done:
/* fence first */
retval = fence_helper();
if(retval != PMIX_SUCCESS){
if(verbose_print) printf("error fencing, finalize may fail ! \n");
}
/* finalize */
PMIx_Deregister_event_handler(_g_errhandler_ref, NULL, NULL);
if(verbose_print){
fprintf(stdout,
"spawn master process (rank %d) (host %s) finalizing\n",
main_proc.rank,
hostn);
}
/* clean up pmix */
retval = PMIx_tool_finalize();
if(retval == PMIX_SUCCESS)
{
if(verbose_print){
printf("spawn master process %d finalize success\n\n",
main_proc.rank);
}
}
else
{
printf("spawn master process %d pmix_finalize FAILURE: %d\n\n",
main_proc.rank,
retval);
}
retval = PMIx_Finalize(NULL, 0);
fflush(stdout);
/* cleanup before returning */
PMIX_INFO_FREE(job_info, job_info_count);
spawned_app->argv = NULL;
PMIX_APP_FREE(spawned_app, 1);
if(verbose_print) printf("%s exiting cleanly :)\n", argv[0]);
return 0;
}
int main(int argc, char **argv)
{
int rc;
pmix_value_t value;
pmix_value_t *val = &value;
pmix_proc_t proc;
uint32_t nprocs;
/* init us */
if (PMIX_SUCCESS != (rc = PMIx_Init(&myproc, NULL, 0))) {
fprintf(stderr, "Client ns %s rank %d: PMIx_Init failed: %d\n", myproc.nspace, myproc.rank, rc);
exit(0);
}
fprintf(stderr, "Client ns %s rank %d: Running\n", myproc.nspace, myproc.rank);
/* get our universe size */
if (PMIX_SUCCESS != (rc = PMIx_Get(&myproc, PMIX_UNIV_SIZE, NULL, 0, &val))) {
fprintf(stderr, "Client ns %s rank %d: PMIx_Get universe size failed: %d\n", myproc.nspace, myproc.rank, rc);
goto done;
}
nprocs = val->data.uint32;
PMIX_VALUE_RELEASE(val);
fprintf(stderr, "Client %s:%d universe size %d\n", myproc.nspace, myproc.rank, nprocs);
completed = false;
/* register our errhandler */
PMIx_Register_event_handler(NULL, 0, NULL, 0,
notification_fn, errhandler_reg_callbk, NULL);
/* call fence to sync */
PMIX_PROC_CONSTRUCT(&proc);
(void)strncpy(proc.nspace, myproc.nspace, PMIX_MAX_NSLEN);
proc.rank = PMIX_RANK_WILDCARD;
if (PMIX_SUCCESS != (rc = PMIx_Fence(&proc, 1, NULL, 0))) {
fprintf(stderr, "Client ns %s rank %d: PMIx_Fence failed: %d\n", myproc.nspace, myproc.rank, rc);
goto done;
}
/* rank=0 calls abort */
if (0 == myproc.rank) {
PMIx_Abort(PMIX_ERR_OUT_OF_RESOURCE, "Eat rocks",
&proc, 1);
fprintf(stderr, "Client ns %s rank %d: Abort called\n", myproc.nspace, myproc.rank);
}
/* everyone simply waits */
while (!completed) {
struct timespec ts;
ts.tv_sec = 0;
ts.tv_nsec = 100000;
nanosleep(&ts, NULL);
}
done:
/* finalize us */
fprintf(stderr, "Client ns %s rank %d: Finalizing\n", myproc.nspace, myproc.rank);
PMIx_Deregister_event_handler(1, op_callbk, NULL);
if (PMIX_SUCCESS != (rc = PMIx_Finalize(NULL, 0))) {
fprintf(stderr, "Client ns %s rank %d:PMIx_Finalize failed: %d\n", myproc.nspace, myproc.rank, rc);
} else {
fprintf(stderr, "Client ns %s rank %d:PMIx_Finalize successfully completed\n", myproc.nspace, myproc.rank);
}
fflush(stderr);
return(0);
}
int main(int argc, char **argv)
{
pmix_status_t rc;
pmix_value_t *val;
pmix_proc_t proc;
pmix_info_t *info;
size_t ninfo;
pmix_query_t *query;
size_t nq, n;
myquery_data_t myquery_data;
pid_t pid;
pmix_status_t code = PMIX_ERR_JOB_TERMINATED;
mylock_t mylock;
myrel_t myrel;
uint16_t localrank;
char *target = NULL;
pid = getpid();
/* init us - since we were launched by the RM, our connection info
* will have been provided at startup. */
if (PMIX_SUCCESS != (rc = PMIx_tool_init(&myproc, NULL, 0))) {
fprintf(stderr, "Debugger daemon: PMIx_tool_init failed: %d\n", rc);
exit(0);
}
fprintf(stderr, "Debugger daemon ns %s rank %d pid %lu: Running\n", myproc.nspace, myproc.rank, (unsigned long)pid);
/* register our default event handler */
DEBUG_CONSTRUCT_LOCK(&mylock);
PMIx_Register_event_handler(NULL, 0, NULL, 0,
notification_fn, evhandler_reg_callbk, (void*)&mylock);
DEBUG_WAIT_THREAD(&mylock);
if (PMIX_SUCCESS != mylock.status) {
rc = mylock.status;
DEBUG_DESTRUCT_LOCK(&mylock);
goto done;
}
DEBUG_DESTRUCT_LOCK(&mylock);
/* get the nspace of the job we are to debug - it will be in our JOB info */
#ifdef PMIX_LOAD_PROCID
PMIX_LOAD_PROCID(&proc, myproc.nspace, PMIX_RANK_WILDCARD);
#else
PMIX_PROC_CONSTRUCT(&proc);
(void)strncpy(proc.nspace, myproc.nspace, PMIX_MAX_KEYLEN);
proc.rank = PMIX_RANK_WILDCARD;
#endif
if (PMIX_SUCCESS != (rc = PMIx_Get(&proc, PMIX_DEBUG_JOB, NULL, 0, &val))) {
fprintf(stderr, "[%s:%d:%lu] Failed to get job being debugged - error %s\n",
myproc.nspace, myproc.rank,
(unsigned long)pid, PMIx_Error_string(rc));
goto done;
}
if (NULL == val || PMIX_STRING != val->type || NULL == val->data.string) {
fprintf(stderr, "[%s:%d:%lu] Failed to get job being debugged - NULL data returned\n",
myproc.nspace, myproc.rank, (unsigned long)pid);
goto done;
}
/* save it for later */
target = strdup(val->data.string);
PMIX_VALUE_RELEASE(val);
fprintf(stderr, "[%s:%d:%lu] Debugging %s\n", myproc.nspace, myproc.rank,
(unsigned long)pid, target);
/* get my local rank so I can determine which local proc is "mine"
* to debug */
val = NULL;
if (PMIX_SUCCESS != (rc = PMIx_Get(&myproc, PMIX_LOCAL_RANK, NULL, 0, &val))) {
fprintf(stderr, "[%s:%d:%lu] Failed to get my local rank - error %s\n",
myproc.nspace, myproc.rank,
(unsigned long)pid, PMIx_Error_string(rc));
goto done;
}
if (NULL == val) {
fprintf(stderr, "[%s:%d:%lu] Failed to get my local rank - NULL data returned\n",
myproc.nspace, myproc.rank, (unsigned long)pid);
goto done;
}
if (PMIX_UINT16 != val->type) {
fprintf(stderr, "[%s:%d:%lu] Failed to get my local rank - returned wrong type %s\n",
myproc.nspace, myproc.rank, (unsigned long)pid, PMIx_Data_type_string(val->type));
goto done;
}
/* save the data */
localrank = val->data.uint16;
PMIX_VALUE_RELEASE(val);
fprintf(stderr, "[%s:%d:%lu] my local rank %d\n", myproc.nspace, myproc.rank,
(unsigned long)pid, (int)localrank);
/* register another handler specifically for when the target
* job completes */
DEBUG_CONSTRUCT_LOCK(&myrel.lock);
myrel.nspace = strdup(proc.nspace);
PMIX_INFO_CREATE(info, 2);
PMIX_INFO_LOAD(&info[0], PMIX_EVENT_RETURN_OBJECT, &myrel, PMIX_POINTER);
/* only call me back when this specific job terminates */
PMIX_LOAD_PROCID(&proc, target, PMIX_RANK_WILDCARD);
PMIX_INFO_LOAD(&info[1], PMIX_EVENT_AFFECTED_PROC, &proc, PMIX_PROC);
fprintf(stderr, "[%s:%d:%lu] registering for termination of %s\n", myproc.nspace, myproc.rank,
(unsigned long)pid, proc.nspace);
DEBUG_CONSTRUCT_LOCK(&mylock);
PMIx_Register_event_handler(&code, 1, info, 2,
release_fn, evhandler_reg_callbk, (void*)&mylock);
DEBUG_WAIT_THREAD(&mylock);
if (PMIX_SUCCESS != mylock.status) {
rc = mylock.status;
DEBUG_DESTRUCT_LOCK(&mylock);
PMIX_INFO_FREE(info, 2);
goto done;
}
DEBUG_DESTRUCT_LOCK(&mylock);
PMIX_INFO_FREE(info, 2);
/* get our local proctable - for scalability reasons, we don't want to
* have our "root" debugger process get the proctable for everybody and
* send it out to us. So ask the local PMIx server for the pid's of
* our local target processes */
nq = 1;
PMIX_QUERY_CREATE(query, nq);
PMIX_ARGV_APPEND(rc, query[0].keys, PMIX_QUERY_LOCAL_PROC_TABLE);
query[0].nqual = 1;
PMIX_INFO_CREATE(query[0].qualifiers, 1);
PMIX_INFO_LOAD(&query[0].qualifiers[0], PMIX_NSPACE, target, PMIX_STRING); // the nspace we are enquiring about
/* setup the caddy to retrieve the data */
DEBUG_CONSTRUCT_LOCK(&myquery_data.lock);
myquery_data.info = NULL;
myquery_data.ninfo = 0;
/* execute the query */
if (PMIX_SUCCESS != (rc = PMIx_Query_info_nb(query, nq, cbfunc, (void*)&myquery_data))) {
fprintf(stderr, "PMIx_Query_info failed: %d\n", rc);
goto done;
}
DEBUG_WAIT_THREAD(&myquery_data.lock);
DEBUG_DESTRUCT_LOCK(&myquery_data.lock);
PMIX_QUERY_FREE(query, nq);
if (PMIX_SUCCESS != myquery_data.status) {
rc = myquery_data.status;
goto done;
}
fprintf(stderr, "[%s:%d:%lu] Local proctable received\n", myproc.nspace, myproc.rank, (unsigned long)pid);
/* now that we have the proctable for our local processes, we can do our
* magic debugger stuff and attach to them. We then send a "release" event
* to them - i.e., it's the equivalent to setting the MPIR breakpoint. We
* do this with the event notification system. For this example, we just
* send it to all local procs of the job being debugged */
(void)strncpy(proc.nspace, target, PMIX_MAX_NSLEN);
proc.rank = PMIX_RANK_WILDCARD;
ninfo = 2;
PMIX_INFO_CREATE(info, ninfo);
PMIX_INFO_LOAD(&info[0], PMIX_EVENT_CUSTOM_RANGE, &proc, PMIX_PROC); // deliver to the target nspace
PMIX_INFO_LOAD(&info[1], PMIX_EVENT_NON_DEFAULT, NULL, PMIX_BOOL); // deliver to the target nspace
fprintf(stderr, "[%s:%u:%lu] Sending release\n", myproc.nspace, myproc.rank, (unsigned long)pid);
rc = PMIx_Notify_event(PMIX_ERR_DEBUGGER_RELEASE,
NULL, PMIX_RANGE_CUSTOM,
info, ninfo, NULL, NULL);
if (PMIX_SUCCESS != rc) {
fprintf(stderr, "%s[%s:%u:%lu] Sending release failed with error %s(%d)\n",
myproc.nspace, myproc.rank, (unsigned long)pid, PMIx_Error_string(rc), rc);
goto done;
}
/* do some debugger magic while waiting for the job to terminate */
DEBUG_WAIT_THREAD(&myrel.lock);
done:
if (NULL != target) {
free(target);
}
/* finalize us */
fprintf(stderr, "Debugger daemon ns %s rank %d pid %lu: Finalizing\n", myproc.nspace, myproc.rank, (unsigned long)pid);
if (PMIX_SUCCESS != (rc = PMIx_Finalize(NULL, 0))) {
fprintf(stderr, "Debugger daemon ns %s rank %d:PMIx_Finalize failed: %d\n", myproc.nspace, myproc.rank, rc);
} else {
fprintf(stderr, "Debugger daemon ns %s rank %d pid %lu:PMIx_Finalize successfully completed\n", myproc.nspace, myproc.rank, (unsigned long)pid);
}
fflush(stderr);
return(0);
}
